![Community tools [120]](https://www.researchgate.net/profile/Gilson-Yukio-Sato/publication/220660674/figure/fig1/AS:341236520112138@1458368534593/Community-tools-120_Q320.jpg)
Content uploaded by Gilson Yukio Sato
Author content
All content in this area was uploaded by Gilson Yukio Sato
Content may be subject to copyright.
Auton Agent Multi-Agent Syst
DOI 10.1007/s10458-011-9170-9
Agent and multi-agent applications to support
distributed communities of practice: a short review
Gilson Yukio Sato ·Hilton José Silva de Azevedo ·
Jean-Paul A. Barthès
© The Author(s) 2011
Abstract This paper concerns the relationship between agents or multi-agent systems and
distributed communities of practice. It presents a review of a number of agent and multi-
agent applications with features that could contribute to supporting distributed communities
of practice. The association is promising because of features like autonomy, pro-activity,
flexibility or ability to integrate systems that characterize agents and multi-agent systems.
Furthermore, such an association is a step towards building mixed communities of humans
and artificial agents. To understand how agents and multi-agent systems could answer some of
the needs of distributed communities of practice, we organize the analyzed applications into
five different categories defined by considering the main activities of a community, namely:
Individual Participation, Synchronous Interactions, Asynchronous Interactions, Publishing
and Community Cultivation. Such a classification helps us identify the relevant features of
the current technology and determine some that should be further developed, e.g. to sup-
port community coordination or gather information related to virtual communities. For each
application we selected, we present its main approach and point out its potential interest.
Keywords Multi-agent system ·Intelligent agent ·Community of practice ·
Applications for virtual communities
G. Y. Sato (B
)
Post-Graduate Program in Biomedical Engineering, PPGEB and Electronics Department, DAELN,
Federal University of Technology—Paraná, Av. Sete de Setembro, 3165, 80230-901 Curitiba, Paraná,
Brazil
e-mail: sato@utfpr.edu.br
H. J. S. de Azevedo
Post-Graduate Program in Technology, PPGTE and Electronics Department, DAELN, Federal University
of Technology—Paraná, Av. Sete de Setembro, 3165, 80230-901 Curitiba, Paraná, Brazil
e-mail: hilton@utfpr.edu.br
J.-P. A. Barthès
CMR UMR6599 HEUDIASYC, Computer Science Department, Centre de Recherches Royallieu,
Université de Technologie de Compiègne, BP 529, 60205, Compiègne, France
e-mail: barthes@utc.fr
123
Auton Agent Multi-Agent Syst
1 Introduction
The notion of community of practice (CoP) has been used in different domains, like social
sciences, education or knowledge management, to study and understand learning and col-
laborative work [6,13,23,33,37,46,47,54,84,97,119].
The proliferation of CoPs and the widely spread use of Internet tools, ranging from email to
virtual environments, led to the creation of distributed CoPs (DCoPs). DCoP members face
more difficulties to perform their communal activities than members of co-located CoPs.
Such difficulties are due to distance among members (who cannot rely on face-to-face inter-
actions), lack of awareness concerning other members’ activities, high number of members,
or different cultural mindsets [119]. Clearly in this context, systems able to support DCoP
activities are highly desirable.
We argue that, since its beginning, the research in multi-agent systems was associated with
notions of social science. From such a premise, we built our argument to justify the association
between multi-agent and agent technologies with distributed communities of practice. From
the technical point of view, agent and multi-agent technologies possess characteristics such
as autonomy, pro-activity, flexibility and ability to integrate systems, making them eligible
as support systems. Furthermore, in the long term, such systems will lead to communities
that mingle humans and artificial agents.
Given the potential of the association of MAS and DCoP, we decided to look at the
state-of-the-art of agent and multi-agent systems from the perspective of supporting DCoPs.
Therefore, to classify the systems and to structure our study, we used the categorization pro-
posed by Wenger et al. [120] that considers five types of activities performed by members
of DCoPs: Synchronous Interactions, Asynchronous Interactions, Publishing, Community
Cultivation and Individual Participation. Thus, we could identify which DCoP activities are
supported by current agent and multi-agent systems, and envisage some features and char-
acteristics to improve such a support.
In the paper we first define distributed communities of practice, mentioning some of
their particular features. We then present agents and multi-agent systems (MAS) discussing
arguments that justify their interest for supporting a community. We then explain how we
conducted the review and present the analysis and categorization of the selected systems. We
end up with some comments and conclusions on the work.
2 Communities of practice
In this section, we discuss the notion of communities of practice and some related concepts
such as identity, trajectory, and multi-membership. Such concepts help justify the use of
agents and multi-agent systems as tools to support DCoPs and to understand the potential
usefulness of the systems reviewed in the present paper.
2.1 An evolving notion
The notion of community of practice (CoP) was proposed by Lave and Wenger [58]intheir
seminal work “Situated Learning: Legitimate Peripheral Participation”. Since then, it has
evolved. Cox [23] and Kimble [54] agree that its evolution underwent three periods, each
period being associated with a major work. The first period is characterized by the work
of Lave and Wenger. The second period is defined by the book “Communities of Practice:
Learning, Meaning and Identity” by Wenger [118]. The third period can be associated with the
123
Auton Agent Multi-Agent Syst
book “Cultivating Communities of Practice: a Guide to Managing Knowledge” by Wenger
et al. [119].
Cox [23] and Kimble [54] agree that the changes in the notion of communities of practice
were considerable. In the first period, Lave and Wenger [58] focus on the concepts of Situated
Learning and Legitimate Peripheral Participation, and leave the notion of CoP intuitive. They
consider that a CoP is a set of relations between people, activities and the world. For them,
a CoP is an intrinsic condition for the existence of knowledge not only because it provides
a background to interpret information, but also because the social practice defines what is
possible to learn.
In the second period, Wenger [118] puts the notion of CoPs in the center stage, develop-
ing it and its relations with other concepts like identity, meaning and learning. He does not
provide a straightforward definition for CoP, but characterizes it by describing three aspects
of the relation between community and social practice: mutual engagement, common enter-
prise and shared enterprise. The mutual engagement brings a sense of belonging to CoP
members. Belonging to a CoP means engaging in its practice. The mutual engagement is
built and maintained by shared informal activities. A common enterprise gives coherence to
the CoP. Such coherence is created and kept by the continuous negotiation of the common
enterprise. The common enterprise creates mutual accountability between members helping
them define what is relevant and acceptable. In the pursuit of the common enterprise, CoP
members elaborate several resources that form a shared repertoire. Such a repertoire includes
routines, words, tools, procedures, etc.
The third period is more prescriptive. Wenger et al. [119] develop recommendations on
how to apply CoPs to Knowledge Management initiatives. In this period, Wenger et al. define
a CoP as “a group of people who share a concern, a set of problems, or a passion about a topic,
and who deepen their knowledge and expertise in the corresponding area by interacting on an
ongoing basis.” They also present a structural model of CoP that includes three elements: (i)
a domain of knowledge; (ii) a community of people; and (iii) a shared practice. The domain
defines a set of issues and legitimizes the community by asserting its purpose and value to
its members. The CoP’s domain motivates its members to participate and contribute. Such a
domain also helps members define what activities should be performed. The community cre-
ates the social fabric of learning and fosters interactions and relationships based on mutual
respect and trust. This kind of relationship creates an environment encouraging people to
share ideas, to expose their ignorance, to ask questions and to listen carefully. The practice
is a “set of frameworks, ideas, tools, information, styles, languages, stories and documents
that community members share.” It represents the knowledge that the community creates,
shares and maintains [119].
It is possible to observe that in the first period, the notion of CoP is more abstract and open
than in the subsequent periods, making it more difficult to be applied in the analysis of tools
to support distributed communities of practice (DCoPs). Despite the absence of a definition
of CoP, the second period provides a theoretical framework with concepts, such as identity,
that are more appropriate to analyze tools able to support DCoPs. The approach of the third
period is not as deep and complex as the previous ones but contains some concepts, such
as a structural model of a CoP, which can be useful in the analysis of tools able to support
DCoPs.
2.2 Distributed communities of practice
Although suggested in the second period when the notion of locality is discussed, distributed
communities of practice are defined in the third period. A DCoP is a CoP in which members
123
Auton Agent Multi-Agent Syst
are distributed geographically and thus, cannot count on frequent face-to-face (formal or
informal) meetings to interact. Since DCoP members do not work close to one another, they
use technological means like videoconferencing, email or virtual environments to interact
[119]. DCoPs can bring together people spread in different units of an organization or even
in different organizations.
DCoPs must overcome a set of obstacles to succeed. The most obvious is the distance. A
DCoP is less “present” in the members’ routines. A community makes itself “present” when
members run into one another at the cafeteria, or in the hall, or when they hold a face-to-face
meeting. In a local community it is easier to connect informally with other members, and
members are more visible. Even if a member does not expose his opinion in a meeting, other
members can observe his passive participation. Newcomers can speak in private when they
run into another member in the cafeteria to network informally. In a DCoP, it is difficult
to observe who is participating passively by just reading the messages on the discussion
board. It is also difficult to meet other members fortuitously or to connect informally during
teleconferences.
DCoPs have more potential members than local ones, so they tend to be larger, making it
difficult to know everybody personally and to establish relationships based on trust and mutual
respect. The number of members can also impact the community structure. For example, a
large distributed community can be divided into sub-communities with local coordination.
DCoPs can also face cultural issues. As they can have members distributed all over the
world, distinct national cultures can affect the way a community works considerably. For
example, in some cultures, newcomers can feel extremely uncomfortable to criticize or even
ask questions to senior members. Language is also an important factor. For example, members
that are not proficient in the DCoP “official” language can find it difficult to express their
viewpoints clearly. All these peculiarities make the development of DCoPs more difficult
[12,61,100,119]. Considering such characteristics, tools able to support the development of
such communities are desirable.
2.3 Identity, trajectories and multi-membership
The notions of identity and trajectories were introduced in the first period, but they were only
fully developed in the second one.
The notion of CoP belongs to a wider theoretical framework, described in the first period,
as Situated Learning. A CoP is the place where the process of Legitimate Peripheral Partic-
ipation takes place. In this process newcomers build an identity in the process of becoming
old-timers. Newcomers start from a peripheral participation to achieve a full participation
[58]. In this context, the concept of identity plays a major role, because the development of
an identity implies learning the CoP practice.
In the second period, Wenger [118] argues that newcomers feel more acquainted with
some communities than others and when joining some of them, they continue forging their
identity. A newcomer will learn the community practices, will become more proficient in
its domain and will share knowledge as a way to belong to this community. Belonging to a
community helps building an identity because it helps defining what is to be known and what
can be ignored.
Identities are not static, they change in time. They have trajectories inside communities
that represent the past, the present and the future of the community members. The analysis
of emblematic trajectories could help newcomers envision their future and allow old-timers
to revisit their own history.
123
Auton Agent Multi-Agent Syst
Identities also develop through space, they cross boundaries among communities. People
participate in several communities and they can neither use a specific identity for each of
them nor use the same identity in each of them. Each person develops different aspects of
the same identity in order to participate in different communities. Multi-membership is an
inherent aspect of identities being a bridge between different communities and constituting
another way to expand identities [118].
2.4 Life cycle
The notion of CoP life cycle is developed in the third period. Wenger et al. [119] argue that a
CoP has a life cycle. Such a cycle has five stages: potential, coalescing, maturing, steward-
ship and transformation. In the potential stage, possible members are loosely connected. As
such members start to build connections, they coalesce into a CoP. In the maturation stage,
more members join the CoP and the knowledge sharing between members is deeper. In the
stewardship stage, CoP members develop their knowledge and practices consciously. In the
transformation stage, CoPs can: disappear, split into sub-communities, become a different
type of group (e.g. a department) or merge with other communities.
3 Agents and multi-agent systems
In this section, we briefly discuss agents and multi-agent systems. We seek to justify the use
of Agent and MAS technology to develop systems able to support DCoPs.
3.1 Agents
Wooldridge [121] considers that “the notion of an agent as an isolated system is evident in
early AI (artificial intelligence) literature.” However, not until the 1980s, would agents consti-
tute a central concern for the AI research community [50]. Even if agents have been considered
a research subject since then, there is not a consensual definition of agent [7,18,25,50,114].
Most AI researchers accept the two different types of agency defined by Wooldridge and
Jennings [114,122]: weak and strong. A “weak” agent has the following properties: auton-
omy, social ability, reactivity and pro-activity. A “strong” agent has the same properties, but
is conceptualized or implemented using notions that would be usually applied to humans
such as: beliefs, intentions or knowledge [122].
In a later work, Jennings et al. [50] adapt a definition from Wooldridge and Jennings [122]:
“an agent is a computer system, situated in some environment […] capable of flexible auton-
omous action in order to meet its design objectives.” In such a definition, three concepts are
emphasized: situatedness, autonomy and flexibility. Situatedness refers to the agent’s ability
to receive data from the environment and act to change such an environment. Autonomy is
the capacity of acting without the direct intervention of humans and controlling its internal
state and actions. To be flexible, agents should be: responsive, pro-active and social. An
agent is responsive if it can respond to changes in its environment in a timely fashion. To be
pro-active, an agent should present a goal-directed behavior and initiative when appropriate.
An agent is social if it is able to interact with other artificial agents and humans in order to
achieve its goals and help other agents [50]. Capacities such as the ones mentioned above
would be useful in systems able to support DCoPs.
Case et al. [17] do not refer specifically to DCoPs, but argue that intelligent agents can
provide an appropriate platform to provide support for electronic communities. As DCoPs,
123
Auton Agent Multi-Agent Syst
electronic communities connect geographically dispersed people. Case et al. consider that
an intelligent agent should possess some or all of the following capacities: cooperation, pro-
activity and adaptability. For example, an agent with adaptive behavior could retrieve more
personalized information using its user’s previous information searches. Pro-active agents
could execute the search for a piece of information without user intervention. Although dif-
ferent from the one used by Jennings et al. [50], the set of characteristics that define an
intelligent agent for Case et al. looks similar. Notions like pro-activity and sociability are
present in both sets.
We consider that even “weak” agents or stand-alone agents could be useful to support
the activities of DCoP members because they possess capabilities like autonomy and pro-
activity. Such capabilities make them interesting to provide services such as personalized
information retrieval and collaborative filtering. Since the DCoP members’ activities are per-
formed in parallel with other activities like those of a project team, members do not have
much time to devote to community tasks. Agents that are autonomous and pro-active can help
them decrease their workload by performing tasks that can be automated such as information
retrieval.
However, although useful when supporting DCoPs, “weak” agents and stand-alone agents
can only offer a limited support. For example, different agents could require different user
profiles to work efficiently. They could also require different interfaces to interact with their
users. To minimize the effects of such limitations, multi-agent systems could be deployed.
3.2 Multi-agent systems
Despite the gap between researchers in academia and industrial users, Multi-agent systems
(MAS) have been deployed in several domains [79,80] such as concurrent engineering [91],
manufacturing [78–80,91], knowledge management [108], communications [68], air traffic
control/flow [2,80], logistics [80], space exploration [80,92], or e-commerce [86,93].
They can be used to intelligently assist users in specialized or generic tasks. Specialized
tasks include, among others, network management [68] or operation of CAE tools [73].
Generic tasks include handling information (e.g. retrieving, filtering, synthesizing), making
decisions (decision support systems) [55] or capturing lessons learned by a project team
[108].
Based on the work of Durfee and Lesser [27], Jennings et al. [50] and Sycara [105]define
a MAS “as a loosely coupled network of problem solvers that interact to solve problems that
are beyond the individual capabilities or knowledge of each problem solver.” The problem
solver mentioned in the definition is an agent. Sycara [105] also describes the abilities pre-
sented by multi-agent systems that make them an interesting research subject. Such abilities
can also be used to argue that multi-agent technology is a promising technology to implement
systems able to support DCoPs.
The first ability is the capacity of solving problems that are so large that a centralized agent
would not have enough resources to do so. Furthermore, a centralized agent would become
a performance bottleneck or a source of risk because of the possibility of failure at critical
times. In the case of systems supporting DCoPs, it is difficult to estimate whether or not the
problem is large enough to justify the use of a MAS. DCoPs tend to be larger than the local
CoPs, but they can remain small. Wenger et al. [119] mention a large community with about
1.500 geographically distributed members at Shell. Such a community shares resources with
other communities, so there are potentially more people using the same systems to interact.
We consider that, although relevant, the capacity of solving large problems is not a decisive
argument to select MAS as a technology to implement systems to support DCoPs.
123
Auton Agent Multi-Agent Syst
The second ability is making possible the interconnection and interoperation of multiple
legacy systems. The idea is to use an agent wrapper around each system and incorporate them
into a group of agents. It is then possible to make the features of the legacy systems available
to the new systems. An example of the integration of legacy systems is the integration of
agents developed by different groups to simulate the evacuation of civilians from threatened
locations. Such integration was developed in the Teamcore framework, a framework for
building agent organizations [82,112]. Such ability can be very useful for systems able to
support DCoPs. Not just because systems that possess such an ability can incorporate legacy
systems, but because, by using a similar approach, it is possible to incorporate new features
to the current system. As DCoPs have life cycles, it is probable that they would use different
tools to perform their activities as they evolve. For example, a DCoP could start with a small
group of people, with no specific funding, that tends to choose among the available tools (e.g.
freeware). As the DCoP grows, either different types of features might become desirable, or
the DCoP might become “official” in an organization and be obliged to “transfer” its activities
to an “official” system. In either case, the ability to interconnect and to allow interoperation
could be useful. If new features are required, an agent wrapper could be used to incorporate
it to the current system. In the second case, an agent wrapper could be used to incorporate
the old system to the “official” one.
The third ability is to solve problems “that can naturally be regarded as a society of
autonomous interacting components-agents” [105]. We consider that supporting DCoPs is a
problem that fits such a description. Although Hattori et al. [44] do not refer to DCoPs specif-
ically; they argue that a MAS is an attractive technology to support networked communities.
The distributed character of this type of community fits in a distributed architecture like the
multi-agent architecture.
The fourth ability is the capacity of using distributed information sources to solve prob-
lems. As the members of a DCoP are geographically distributed, so are the information
sources. Even if central repositories of information are used, members themselves can also
be considered as sources of information and they are geographically dispersed. Furthermore,
DCoPs develop their activities over the Internet, considered the ultimate distributed infor-
mation source.
The fifth ability is solving problems using distributed expertise. In the case of DCoPs,
the fifth ability is similar to the fourth. Old-time members could also be geographically
distributed.
The sixth ability is the capacity of improving the performance along the dimensions of
computational efficiency, reliability, extensibility, robustness, maintainability, responsive-
ness, flexibility, and reuse. From the above, we consider that the flexibility and the extensi-
bility are likely to be the basis for developing applications that better meet the evolving needs
of a community. CoPs can differ from one another [119]. For example, some communities
can prefer the use of asynchronous tools like email or blogs, while others will prefer using
synchronous tools like chats. Furthermore, DCoPs have life cycles. In this context the tools
used in the community must also change. In this case, the flexibility and extensibility inherent
to a MAS can help provide an always adequate set of tools for the community. Hattori et al.
[44] do not refer to CoPs specifically, but they argue that the support system for a virtual
community should handle the dynamic nature of such a community whose members change
the way they participate in the community.
Thus, analyzing the abilities described by Sycara [105] regarding the characteristics of
DCoPs, it is possible to consider multi-agent technology as a promising technology to develop
systems for supporting DCoPs.
123
Auton Agent Multi-Agent Syst
Furthermore, virtual communities have other issues that can be handled by a MAS. Hattori
et al. [44] argue that the use of personal agents can preserve the individuality of each com-
munity member. Community members share some objectives, but they also have individual
objectives. As community members have different interests, they can participate in different
communities. The inclusion of personal agents in MAS can personalize the support to the
specific needs of the individual and preserve the individuality of members. The personal
agents can also be used to alleviate the workload of community members by unifying the
interface to several systems. Since DCoPs share the same issues, we consider that the authors’
arguments can also be applied to DCoPs.
It is also possible to identify some advantages provided by multi-agent technology when
comparing approaches using a MAS to others using different technologies. Hattori et al.
[44] argue that, as virtual community members are dispersed and the number of potential
members is large (as DCoPs), solid, centralized and monolithic solutions would not be an
adequate solution. Such types of solutions include most groupware and portals.
Barthès and Tacla [5] identify advantages and limitations of using a MAS or a groupware-
based portal to manage knowledge. The groupware is a more mature technology, but has
limited pro-activity and extensibility. In contrast, most limitations of the MAS approach are
related to the fact that it is not as mature as groupware ones. Thus, Barthès and Tacla propose
a combination of both approaches by using an agent to encapsulate some of the groupware
features. By combining agents and groupware, they try to overcome the limited pro-activity
and extensibility of groupware systems. They compare the performance of groupware and
MAS to manage knowledge, leading us to consider that their analysis might also be applied to
systems that support the activities developed by DCoPs. The portal described by Barthès and
Tacla has features similar to the ones found in systems supporting DCoPs such as document
repositories, directories, and calendars.
The research in Computer Supported Cooperative Work (CSCW) and groupware has
several points in common with the research in DCoPs. Both domains have a multidisciplin-
ary character (about such a character in CSCW [28,42] and in CoPs/DCoPs [118]) and are
concerned about facilitating human to human interaction by means of technological tools.
Several systems we surveyed for our review are groupware tools (e.g. [41,126]). The points
in common between the two domains make them overlapping domains.
Despite the existence of common characteristics, it is possible to identify distinctions
between both domains. Ellis et al. [28] mention in their definition of groupware “a group of
people with a common goal or task.” DCoP members could have common goals or tasks,
but usually they join DCoPs to learn by participating in the communal activities [118]. In
DCoPs, goals can be loosely defined or implicit and a specific task (e.g. elaboration of a
manual about the community domain) could be carried out by a small part of the community
members. Boundaries of groups using groupware tend to be clear and DCoPs tend to have
fuzzy boundaries (e.g. lurkers are relevant participants in a DCoP) [119]. Usually groups
using groupware and DCoPs should be managed in different ways. In DCoPs, the attribution
of roles is not so rigid and the management should be “light handed” [119]. Schedules and
deadlines are not central in DCoPs’ activities. In the other hand, groups using groupware
usually have to control schedules, respect deadlines and their members should play their
roles in order to achieve the group’s goals.
The CSCW/groupware domain and the CoPs/DCoPs domain have similarities and dif-
ferences. We think the cross fertilization of both domains can bear valuable fruits. CoPs
and DCoPs can benefit from a well established research tradition and CSCW/groupware can
benefit from a new social perspective.
123
Auton Agent Multi-Agent Syst
The emergence of the semantic web [9] also creates an opportunity to combine tech-
nologies. Pechoucek and Marík [80] consider that the semantic web and service-oriented
architectures affect considerably the way agent technology will be developed and applied
in the future. Chen et al. [21] combine intelligent agents and Web Services to develop an
automated meeting room. The work presented by Chen et al. suggests that it is possible to
integrate, through a MAS, features from legacy systems, intelligent agents and web services
to provide a system to support DCoPs.
We consider that the discussion above strongly suggests that MAS technology is appro-
priate to implement systems able to support the activities developed by DCoPs. We think that
a major argument in favor of the association of multi-agent systems and DCoPs has not yet
been explored in this paper. Such an argument involves a vision of how people and artificial
agents would interact in the future.
3.3 Communities mixing humans and artificial agents
It is possible to argue in favor of the association between multi-agent systems and DCoPs at
a higher level. Research works in both domains are concerned by the social nature of action
and knowledge. The framework in which the notion of communities of practice is developed
considers that learning is an integral part of the social practice. The MAS research can inspire
and be inspired by social studies.
A foundational aspect of the MAS research should be its concern with the “social dimen-
sions of the action and knowledge as a fundamental category of analysis” [32]. Malsch [66]
and Malsch and Schulz-Schaeffer [67] materialized such a concern in a research field called
Socionics. It is possible to identify three main interests in this research field: (i) the social
investigation by means of simulation of social groups and societies; (ii) the investigation of the
social metaphors used in the development of multi-agent systems; and (iii) the investigation
of the effects of societies mingling human and non-human agents in the human being.
Simulation can be used to study groups and societies because it allows researchers to
change social parameters and analyze the effect of such a change. Fan and Yen [31]survey
works such as COGNET/BATON and Team-Soar that use agents to simulate teamwork. The
former can simulate different kinds of collaboration, the latter was used to test a theory of
team decision making. McCallum et al. [70] use simulation to analyze the effect of the influ-
ence in organizational change. Zoethout et al. [133] examine, by means of simulation, how a
workgroup modifies task allocation when facing different kinds of task. Glass and Grosz [36]
analyze how social consciousness affects the negotiation process during decision making and
the outcomes of the teamwork.
Social metaphors are used in the development of multi-agent systems. Metaphors such as
agent societies, agent organizations and agent ecologies are used in the domain of agent-ori-
ented software development [131]. In the extended version of the Gaia methodology [132],
the MAS architecture is defined in terms of organizational structure. A role model is used
to organize agents in hierarchies, collectives of peers, market-oriented structures, etc. Van
Aart et al. [116] use notions from the domain of organization design to develop a MAS. They
discuss three MAS architectures that are based in the Mintzberg’s organizational structures:
machine bureaucracy, professional bureaucracy and adhocracy.
We have concentrated our discussion of societies mingling human and non-human agents
to the discussion of communities mingling humans and artificial agents. Although farfetched
in the current technological context, it is possible to envision a world where humans and arti-
ficial agents would constitute communities in which artificial agents would behave almost
123
Auton Agent Multi-Agent Syst
as humans. Such a possibility seems to have inspired different research works that associate
humans and artificial agents.
Barthès [4] proposes an environment called Multi-Agent System with Humans (MASH)
to manage the knowledge of a project team. In the MASH environment, each team member
owns a set of agents composed of a personal agent and an exclusive staff of agents. Each
set of agents acquires on the fly the knowledge of its owner. The acquired knowledge can be
exchanged among team members through their personal agents. If a member quits the team,
his personal agent keeps contributing with the knowledge previously acquired.
Urlings et al. [115] use a video game called “Unreal Tournament” as a virtual world con-
taining humans and agents (implemented with the JACK platform). In this virtual world,
teams associating humans and agents are formed. The idea is to train users in combat situa-
tions using simulations.
Tamb e et al. [112] use an agent integration architecture called Teamcore [82] to group
humans and agents. In the Teamcore, each human or agent is associated to a proxy. The
proxies can work as a team because each has a teamwork model that is used to coordinate
its actions with other proxy actions. To demonstrate how the Teamcore works, the authors
describe two applications: casualty rehearsal evacuation exercise in threatened areas and
coordination of meetings.
Sierhuis et al. [92] describe the mobile agent project based on a research that involves the
formation of teams mingling humans and artificial agents to be employed in space explo-
ration. The idea of the project is to use a system to coordinate the action of the mobile
agents and have them collaborate. Mobile agents are entities such as space suits, cameras,
all-terrain vehicles, and robotic assistants. Each mobile agent and person involved in the
current operation has a personal agent that is used to facilitate the communication between
humans and mobile agents as well as between mobile agents.
Isbell et al. [48] developed an agent called Cobot that became a member of a community
formed inside a virtual world. It has evolved from an inanimate object to a socially adaptive
agent. Cobot is able to adapt by collecting statistical information about users’ interactions.
It can also present such information to other community members. Isbell et al. consider that
Cobot became one of the most “popular” community members and it altered the interactions
within the community.
The research works we just described promote the idea of creating communities mingling
humans and artificial agents. The existence of the research suggests that the possibility of
creating such communities is not negligible. Using a MAS to develop systems to support a
DCoP cannot be characterized as an attempt to create communities mingling humans and
agents, but we consider that the future of such systems involves the creation of this type of
community. We think that the idea of associating multi-agent systems and DCoPs is justi-
fied not only because MAS technical characteristics are appropriate to support DCoPs, but
also because the possibility of mingling humans and artificial agents suggests that such an
association is a first step towards the development of new associations between humans and
technological artifacts.
4 The review study
This review aims at mapping the agents and MAS applications that could support the activ-
ities of a DCoP. The idea is to identify which activities performed by DCoPs are already
supported and the new features that are potentially useful for DCoPs. Such features can be
appropriately explored through the use of agent and MAS approach.
123
Auton Agent Multi-Agent Syst
When we started looking for agents and MAS applications oriented towards DCoPs, we
found out that most of them are aimed at several types of virtual communities. That led us
to consider them all and not just the ones explicitly dedicated to DCoPs. Although several
applications do not meet all generic needs of DCoPs, they present useful features able to
support such communities. We also decided to include both “weak” and “strong” agents.
To guarantee that this study could constitute a representative (but not exhaustive) landscape
of agents and MAS applications for communities, we consulted three of the most relevant sci-
entific/technical databases: Elsevier/Science Direct, IEEE/Computer Science and ACM/Dig-
ital Library. We decided to focus our study mostly on the papers published after 1999,
eventually including some from previous years. We chose 1999 because the concept of CoPs
started to spread after Wenger had published “Communities of Practice: Learning, Meaning
and Identity” [118], despite the fact that the concept of “community of practice” had already
been presented in “Situated Learning” in 1991 by Lave and Wenger [58].
In order to organize our findings, we classify the analyzed systems using the framework
proposed by Wenger et al. [120]. The framework is used to analyze technologies that could
be utilized to help community leaders and their sponsors select tools for supporting commu-
nities without over-emphasizing such tools. Although using the same classification scheme,
in our review we do not intend to help leaders and sponsors choose the most appropriate tools
for their communities.
The framework includes a diagram showing how tools utilized by communities fit
in their activities. The diagram is divided into five regions corresponding to different
types of community activity: Synchronous Interactions, Asynchronous Interactions, Pub-
lishing, Community Cultivation and Individual Participation. The location of the tools
in the diagram is meant to show their intended use and relation to the other tools
(Fig. 1).
Fig. 1 Community tools [120]
123
Auton Agent Multi-Agent Syst
In our review, we locate features of agent and multi-agent applications in the diagram
elaborated by Wenger et al. [120] by considering their potential contribution to the execution
of typical community activities. We also tried to make explicit the relation between applica-
tions. In doing so, we sought to establish a framework picture of how the current agents and
multi-agent systems are inter-related. Such a picture could also be used to suggest which kind
of tool is lacking. We first classified the applications into the five categories of the diagram
(Synchronous Interactions, Asynchronous Interactions, Publishing, Community Cultivation
and Individual Participation). Then, we grouped the studied agents and multi-agent systems
by feature (whenever possible) and positioned each group in the diagram.
Some of the analyzed multi-agent systems and agents present features that could fit into
more than one type of activity. In this case, we selected the features that seemed more impor-
tant or predominant in the system for classifying it.
This classification, using the activities of communities, was adopted instead of a classifi-
cation based on the system features, because the core intention of this review is to identify
the state-of-the-art in multi-agent systems and agents for supporting communities, meaning
that we do not intend to compare systems in a given domain (e.g. information retrieval),
but to look for some insight on how such systems potentially support the activities of a
community.
5 Survey of tools and systems
Again, the aim of this study is to survey how multi-agent systems and agents are used to
support community activities and not to compare systems that support communities. The
multi-agent systems and agents described in this section are not comparable for many rea-
sons. They are at different stages of development, implemented with different technologies
and for different purposes.
5.1 Individual participation
The tools for individual participation are chosen for their potential to support community
members performing tasks for the community. They are not necessarily intended for com-
munities, but can be useful for members performing some tasks in a community.
Basically, all the tools for supporting individual participation are intended to help users
handle problems related to information or explicit knowledge. In order to provide an overview
of the available tools, we classified them according to their major features. In the “Individual
Participation” category, we identified five major features: Individual Profiling, Personalized
Newspaper, Support to Web Navigation, FAQ Retrieval, and Information Filtering/Retrieval
(Table 1).
5.1.1 Descriptions of the systems
One of the tasks that should be well supported in individual participation is the access and use
of multiple distributed sources of information and explicit knowledge. Personalized assis-
tance to improve users’ performance in the execution of everyday tasks could also be useful.
Both the access to information and the assistance should be as personalized as possible, thus
personal profiles might be necessary.
123
Auton Agent Multi-Agent Syst
Tab l e 1 Major features
identified in the Individual
Participation category
Major feature Agents and multi-agent systems
Individual Profiling SUITOR [65]
Personalized Newspaper WebMate [20]
NewsAgent [38]
ETTS [15]
Support to Web Navigation [53]
[96]
FAQ Retrieval [123]
Information Filtering/Retrieval AgentRAIDER [90]
Watson [14]
MAIS [29,30]
MAPIS [81]
[51]
[98]
[88]
[95]
Personal Search [37,38]
Several examples [55]
Klusch [55] reviews agents that possess one or more of the three features: information
acquisition and management, information synthesis and presentation and finally intelligent
user assistance. Most of the agents in his review can be used to support the individual
participation in communities. In his review several systems that could support indi-
vidual activities are mentioned: SIMS/ARIADNE, MIX, ABS, TSIMMIS, OBSERVER,
InfoSleuth II, BIG, PLEIADES, IMPACT, SCOPES, RETSINA [106], Amalthaea [74],
InfoSpider [77], LikeMinds and FireFly. The review also mentions personal information
agents deployed in Virtual Reality environments and other personal agents like: Letizia,
Remembrance, ExpertFinder, Butterfly, Let’s Browse [63], TrIAs, AiA, PAN, WebPersona
and others.
An example of a system that could be used to build a user profile is an attentive information
system developed over an architecture called Simple User Interest Tracker (SUITOR). Such
a system can determine users’ interests in order to provide useful information by following
their behavior. It builds a user’s model by tracking eye gaze, web browsing and application
use [65]. Agents and multi-agent systems to build profiles could be used to provide better
services to community members like handling the increase in the workload caused by the
participation in DCoPs.
Agent for Retrieval and Analysis of Information in Distributed Environments (AgentRA-
IDER) interfaces the user to a system composed of commercially available components that
implement a comprehensive architecture for an information retrieval system [90].
Watson is a system that proactively retrieves information from distributed information
sources based on the context gathered from the document on which the user is currently
working. One can also use Watson to perform explicit requests based on a reference docu-
ment [14]. Instead of using a document to gather context, the system developed by Staab and
Schnurr [98] can look for information related to the task that is currently being performed
by the user. It retrieves information from an organizational memory, either in a reactive or in
a proactive way.
123
Auton Agent Multi-Agent Syst
MAIS [29,30] is a system that performs personalized searches on the web. To achieve this
personalization, the system builds a user profile based on a set of centers of interest. Each
center consists of documents representing the user’s interests. To search a piece of informa-
tion, the user provides some keywords and centers of interest to the system through a Personal
Assistant. A Search Agent retrieves the pages identified by commercial search engines (e.g.
Google, AltaVista) using the keywords. A Filter Agent orders the retrieved pages using the
user’s profile and a machine learning technique.
Multi-Agent Personalized Information System (MAPIS) interacts with users and the dif-
ferent information sources to provide personalized information. The agents in MAPIS play
different roles. Assistant Agents interface users and the system, Search Agents search infor-
mation in the available sources, Profile Agents manage the users’ profile and Solver Agents
coordinate the information retrieval, personalization and integration. The information person-
alization considers the information domain, the search request history and the users’ profile
[81].
SUITOR, AgentRAIDER, Watson, MAIS and MAPIS build users’ profiles using contex-
tual information like documents and tasks, but metadata can also be used to retrieve informa-
tion. Ji and Slavendy [51] developed an agent-like metadata filter that retrieves information
from an intranet portal that works as an organizational memory.
The cited systems favor automatic construction of users’ profiles. Other systems favor
interaction with users to build their profile. Still the goal is the same: to improve the retrieval
of information in distributed sources. Shakshuki et al. [88] developed a system that builds
the user’s profile during interaction. The system uses the profile to customize retrieved infor-
mation for the user. It can also monitor and update modifications in the information sources.
Song et al. [95] developed a system to filter and classify documents, by simply analyzing
their titles. The system has two agents: a Classification Agent (CA) and a Filtering Agent
(FA). The CA uses an information inference model that allows it to discover implicit infor-
mation in the document titles. The information inference model deploys a model of semantic
memory based on a significant corpus of texts. Using this mechanism, the CA can infer the
context of the document from the words in its title. The inference model allows the CA to
establish a set of categories that can include the analyzed document. The FA uses the user’s
profile to define which categories can interest the user.
Agent and multi-agent systems filtering and retrieving information as mentioned above,
can be useful to DCoP members. As they use several types of techniques to build profiles, they
can improve information retrievalallowing DCoP members to participate more effectively in
the communal activities. We think that the filtering and retrieving systems described above
could be more efficient if they would consider a community profile as the systems described
in the publishing category section. As DCoP members can participate in different communi-
ties and have private interests, their profiles could present different aspects (e.g. one for each
community). Filtering and retrieving systems would consider such aspects when personaliz-
ing retrieval. To do so, it is necessary to investigate how to “switch” between the different
aspects of the profile or how to mix such aspects in order to obtain a better personalization.
Frequently Asked Questions (FAQ) systems are a considerable source of information.
Yang [ 123] developed a Personal Assistant Agent that interfaces users and FAQ systems.
The goal of such an agent is to provide the most adequate FAQ by “understanding” the users’
intentions. To achieve this goal, the agent uses a domain-specific ontology, an adaptable user’s
model and techniques of linguistic treatment. It is a complex system that can potentially help
integrate newcomers in communities. As newcomers usually need basic information about
the domain and the practice of his new community, a tool to improve the access to FAQs
is potentially useful to accelerate the integration of newcomers. We think that systems to
123
Auton Agent Multi-Agent Syst
retrieve FAQs could be useful, but they should be complemented with systems to extract
FAQs automatically or semi-automatically from sources such as community forums or chats.
WebMate learns the users’ interests continuously in different areas in order to provide
an automatically generated personalized newspaper. The users’ profiles are defined by the
users’ positive feedback when analyzing some documents. Moreover, using such dynamic
profiles helps to refine search, improving the efficiency of the information retrieval [20].
Personal Searcher and News Agent are also agents that employ users’ profiles to retrieve
and organize information from the web. News Agent recommends articles from the user’s
preferred on-line newspapers. Personal Searcher searches the web for documents using key-
words. Both agents use a user’s profile elaborated from the observation of the user’s behavior
while navigating the web. The agents observe actions like the time spent at a page, the way
the page is scrolled and the inclusion of bookmarks, to get an implicit feedback. The pro-
file is also adapted and refined dynamically according to the feedback provided by the user
about the suggestions given by the agents [37,38]. Personal Search was developed using
the WebDCC algorithm for document conceptual clustering to model the users’ preferences
[38]. One of the advantages of these agents is the automatic construction of the profile from
implicit feedback.
Emerging Topic Tracking System (ETTS) is a system that can monitor a region of the
web, detect changes in the information about a specific topic and summarize them to the user.
ETTS has three main components: Area View System, Web Spider and Changes Summarizer.
The Area View System uses keywords provided by the user and commercial search engines
to establish an “information region” that the system will monitor. Web Spider periodically
checks the HTML files in this region to identify new pages and to detect changes in already
checked pages. The Changes Summarizer extracts the changes from the documents (e.g.
phrases included recently) and evaluates them using the Term Frequency-Proportional Doc-
ument Frequency (TF-PDF) algorithm. The most relevant changes are the summary for the
user [15]. Like the Web Mate and The News Agent, ETTS builds a customized newspaper,
but it uses neither an implicit nor an explicit user’s profile.
An agent or a MAS generating personalized newspapers or clippings are similar to sys-
tems to filter and retrieve information. We think that the main difference is their proactive
character. Such a characteristic could help DCoP members to keep track of modifications in
information sources regarding the community domain.
Keeble and Macredie [53] developed an adaptive interface to support web browsing. Their
system can retrieve and cache linked web pages, suggest other web pages, organize book-
marks, perform background searches and repeat periodic page fetching.
Sorensen et al. [96] present an ensemble of agents that support web navigation based on
the users’ bookmarks and navigation history. Support is offered in the form of bookmark
maintenance and automatic verification of updates in web sites.
Systems to improve web navigation could help DCoP members to find information in
the World Wide Web more efficiently. They would be more useful if they could offer the
same kind of services (e.g. bookmark maintenance, verification of updates, suggestion of
web pages) for the community as a whole. Newcomers would benefit from such services
and the DCoP could obtain an information source that can diminish the effect of old-timers’
departures.
There are several approaches, methods and techniques for information filtering and
retrieval from distributed and heterogeneous sources as well as for user profiling. This review
is limited to the studies described above because we considered that they represent the kind
of support agents can give to the individual participation in DCoPs.
123
Auton Agent Multi-Agent Syst
5.1.2 Discussion
We observe that different approaches have been studied for the same problem. An example
is the elaboration of profiles. Some systems try to follow users’ behaviors in order to deter-
mine their preferences and others rely on users’ feedback. Some systems can define the users’
preferences in multiple areas. Several algorithms for information filtering and retrieving have
also been studied. Such research efforts might lead to more appropriate tools for supporting
individuals participating in communities. But, as they are not community oriented, they will
possibly neglect specific needs of DCoPs.
The notion of identity in the DCoPs theoretical framework is rather complex, thus it can-
not be reduced to a personal profile, a user’s model or a set of preferences. However, we can
consider that profiles, models and preferences are “glances” of the community members’
identities. They have been used successfully to personalize services in several systems, but
we consider that it is possible to improve them considering other dimensions of the notion
of identity.
In the DCoPs theoretical framework, since a person belongs to several communities, one
develops different aspects of the same identity. For each community the person belongs to,
one of these aspects is more significant. We think that a profile or a user model should consider
the multiple aspects of an identity. Some of the systems described in the previous section, like
MAIS, deal with the user’s many interests, but it is necessary to make the “current” interest
explicit. Other systems, like Watson, use the current context (e.g. the document the user is
working on) to determine the “current” interest.
The concern that an identity presents multiple aspects has been studied and considered
in the profiles and user models. However, we argue that the multiple aspects of an identity
have been treated in isolation. We consider that a person’s identity is a whole; therefore the
use of the different aspects in isolation does not ideally reflect the notion of identity. But to
consider different aspects of an identity simultaneously implies estimating how much each
aspect is affecting the execution of an activity at a given moment. In such circumstances, it
is possible to envision a MAS in which each agent would represent one aspect of an identity
and would negotiate its importance in a given context.
As identities change in time and describe trajectories inside and through DCoPs, we
believe that the temporal dimension should also be considered. To our knowledge, such a
dimension has been neglected in the profile and user models. Taking into account the tem-
poral dimension exposes some issues. For example, how to evaluate the weight of the past
experiences in a profile? What is the rate of update of a profile? When to “forget” something?
If an identity composed of multiple aspects is considered, would the time affect each aspect
differently?
It is possible to observe that, considering notions such as identity, trajectory and multi-
membership; a set of issues regarding the development of agents and multi-agent systems to
support DCoPs arises. Such issues suggest new subjects to be investigated.
5.2 Synchronous interactions
An agent or a MAS in this category can potentially support synchronous interactions in
a DCoP. Synchronous interactions can happen in face-to-face meetings, videoconferences,
teleconferences, chats or instant messaging. So, we classified the agents related to this kind
of event in this category.
In order to provide an overview of the available tools, we classified them according
to their major features. In the Synchronous Interactions category, we identified five major
123
Auton Agent Multi-Agent Syst
Tab l e 2 Major features
identified in the Synchronous
Interactions category
Major feature Systems
Presence Indicator Gleams of People [127]
Conversational Agent for IM AINI [39]
Meeting Scheduling [22]
[59,60]
RESCH [62]
MSRAC [8]
[89]
Spontaneous Synchronous Encounters Contact Space/Forum [83]
Freewalk [49,75,76]
Smart Meeting Rooms Smart Conference Room [45]
EasyMeeting [21]
MITRE ETR [43]
features: Presence Indicator, Conversational Agent for IM, Meeting Scheduling, Spontaneous
Synchronous Encounters and Smart Meeting Rooms (Table 2).
5.2.1 Descriptions of the systems
A typical system supporting synchronous interactions is a meeting scheduler. As negotiation
is concerned, meeting scheduling is a well explored research topic in the MAS domain (e.g.
[24]). Several systems implement different approaches [22,59]. Although meeting schedulers
were not conceived to support DCoPs, they can help such communities be more efficient.
The RESCH project consists of a MAS to support collaborative work among persons in
a company. In this project, one of the systems helps organize meetings. Each user has an
assistant agent and an agent responsible for managing one’s agenda. There is also a resource
agent. The user assistant promotes meetings (that can be of different types) and negotiates
with other user assistants. It can perform this task considering different factors like the kind
of meeting and the hierarchical position of the participants. The agenda agent controls the
availability of its user and the resource agent controls the availability of the shared resources
like meeting rooms and slide projectors [62].
Lee and Pan [59] propose a system that helps schedule meetings using fuzzy decision
agents. The advantage of such a system is to learn users’ preferences and invitees’ behavior.
Lee et al. [60] refine Lee and Pan’s system by using ontologies. Another system for sched-
uling meetings is proposed by Chun et al. [22]. The system is able to define participants’
preferences by observing the negotiation involved in the meeting organization.
Meeting Scheduling with Reinforcement of Arc Consistency (MSRAC) is also a meeting
scheduling system. The system is more compatible with the needs of actual organizations
because it can relax users’ preferences while keeping the overall consistency. MSRAC is
extensible and experiments demonstrate its efficiency in handling situations with tight con-
straints [8].
Shakshuki et al. [89] also developed a meeting scheduling system. In such a system, users
are represented by a Meeting Scheduling Agent (MSA) that negotiates meeting schedules
with others MSAs. MSAs interact with their users through a GUI and with other MSAs
though email messages or direct communication. To solve a conflict, the system proposes
123
Auton Agent Multi-Agent Syst
three negotiation strategies (First Come First Served, High Rank and Voting) based on the
users’ preferences and scheduling constraints.
Another kind of system that can facilitate synchronous interactions is the Smart Meeting
Room systems. Hellenschimidt and Kirste [46] implemented a smart conference room using
an infrastructure called SodaPop. The smart conference room is able to get a presentation
from a notebook and to control lights, microphones, and loudspeakers.
EasyMeeting is also a smart meeting room system. It has been developed over an archi-
tecture supporting context-aware agents called Context Broker Architecture (CoBrA). Such
an architecture includes context ontologies, context reasoning and control of the users’ con-
textual information. EasyMeeting is able to recognize some vocal commands, to get a pre-
sentation from a specified URL, to play ambient music, to greet the meeting participants and
display a specified web site. The aim of using the context is to allow the system to decide
which services and information should be provided [21].
MITRE Experimental Team Room (ETR) possesses features that are similar to the fea-
tures presented by EasyMeeting. The system can understand vocal commands; control the
lights, the volume of the loudspeakers and the angle of cameras; turn on LCD displays and
start a videoconference connection and navigate a slide presentation. The major feature is an
embodied conversational agent called Electronic MITRE Meeting Assistant (EMMA) that
is used as an interface between the user and the system [43].
Meeting Scheduling and Smart Meeting Room systems are designed to support face-to-
face events. As DCoPs usually organize more on-line events, such systems tend to contribute
peripherally to the support of DCoPs. They would be more useful if they would facilitate the
organization of on-line synchronous events.
Meeting Scheduling systems and Smart Meeting Room systems are more adequate to
facilitate the realization of formal events. To facilitate informal and casual meetings inside a
community, there are more appropriate systems such as Forum and Freewalk.
The Forum system [83] was specifically designed for communities of practice. This web-
based collaborative virtual environment deploys agents to support chance meetings, informal
communication and information sharing among community members. The system has three
components: (i) Contact Space; (ii) Meeting Space; and (iii) Jasper II. Jasper II agents can
build user profiles by observing users’ behaviors and information sharing preferences. The
system can extract information from several sources accessed by users in order to build
their profiles. It can also share information with other people whose profiles indicate similar
interests. Contact Space can move users’ avatars in a collaborative virtual environment in
order to promote informal meetings while Meeting Space supports more formal meetings.
Contact Space utilizes a profile for each interest group to place someone’s avatar in an area
where there are avatars of people with similar profiles. With Jasper II, Contact Space forms
an intelligent environment for information sharing that facilitates information sharing within
a community that is essential for the development of DCoPs. But it is necessary to include
some more cultural and contextual information to allow people to get in touch through the
system more easily [71].
FreeWalk 1 was also developed to promote spontaneous synchronous meetings. Freewalk
1 is part of the social interaction platform called Freewalk that also includes an agent to facil-
itate cross-cultural communication (Freewalk 2) and an environment to a virtual simulator
for conducting virtual evacuation drills (Freewalk 3). But unlike Contact Space, Freewalk
1 allows people to move in a virtual space and to meet other people. Instead of avatars,
Freewalk uses 3D polygons associated with video communication. Freewalk 2 is an agent to
support human-to-human interaction. It animates a conversation and helps finding a common
ground between two persons [49,75,76]. Both tools, FreeWalk 1 and 2, are not community
123
Auton Agent Multi-Agent Syst
oriented, but they can provide a virtual environment that could help finding people with
common interests to form communities. Moreover, a DCoP can use this tool to minimize the
lack of informal meetings.
Contact Space/Forum and Freewalk provide a virtual space that facilitates the occurrence
of informal, spontaneous encounters. Such a space is important to DCoPs, because it can
mimic the socializing effect of “hallway”, “cafeteria” or “water fountain” encounters. For
example, newcomers can feel more comfortable asking questions to an old-timer in informal
encounters. We think that one limitation of systems like Contact Space/Forum and Freewalk is
their relatively closeness. They can promote encounters between people that are “wandering”
in the virtual space created by the system, but not in open spaces like the Internet. Personal
Assistant Agents could be used to indicate when a DCoP member is on-line in a chat room
or navigating in a specific web site.
Another type of tool that can support synchronous interactions among CoP members is
Presence Indicator systems, such as Gleams of People. Gleams of People is a system that
allows users to share presence and humor indicators with other community members. When
a community member chooses a color to represent his humor, the system sends messages to
other members’ Personal Assistant agents to indicate his presence and humor. Each Personal
Assistant responds, sending its user’s presence and humor indicators [127].
Gleams of People can be useful to reinforce the presence of members of distributed com-
munities. It can be used to prompt spontaneous encounters. Presence indicators are current
in several Internet systems such as Gmail, but they are usually restricted to a single system.
We think that there is a lack of more general tools to indicate presence. Such tools should
unify the interface to all Internet systems that require information about presence and status.
Personal Assistant Agents could be an option to implement such a tool.
The instant messaging (IM) is a popular tool to interact synchronously. Goh et al. [39]
developed a conversational agent, called Artificial Intelligence Natural language Identity
(AINI), for this type of system. AINI uses two knowledge sources: a general and a specific.
The system uses the Mindpixel (the training data of the Text Retrieval Conference—TREC)
as a general knowledge source and ALICE Annotated AIML (AAA). As a specific knowledge
source, it uses the knowledge extracted from web sites by Automated Knowledge Extrac-
tion Agent (AKEA). A conversational agent such as AINI could be useful to animate chat
sessions. It could be also used to help integrate newcomers by providing basic information
about the DCoP domain and practice.
5.2.2 Discussion
Although the agent systems for scheduling meetings, setting up and controlling meeting
rooms can be useful for DCoPs, their contribution seems to be still peripheral and oriented
to co-located CoPs. Tools like Contact Space, Freewalk and Gleams of People can play a
more relevant role because they can help form and develop communities. Such systems can
minimize the lack of informal meetings and can help members feel the presence of the com-
munity. Moreover, as they allow people with similar interests to get in touch informally, new
communities can emerge.
Because DCoPs hold few face-to-face meetings, systems for scheduling meetings and
controlling meeting rooms tend to play a secondary role in the support of such communities.
However, since DCoPs meet in virtual environments, a system for scheduling meetings could
be useful to schedule chats, videoconferences or conference calls. Occasionally, DCoPs pro-
mote face-to-face meetings. In such occasions, systems for controlling meeting rooms could
be more useful if they keep a community profile. In this way they could configure the room
123
Auton Agent Multi-Agent Syst
Tab l e 3 Major features
identified in the Asynchronous
Interactions category
Major feature Agents and multi-agent systems
E-mail Intelligent Forward D-mail [64]
Forum of Avatars TelMeA [110,111]
Analysis of Discussion Boards CommunityBoard [44]
Virtualized Ego Interaction EgoChat [57]
for the specific needs of a given DCoP. Systems for controlling meeting rooms would be
more interesting to support DCoPs if they could configure the virtual environment in which
a meeting or a videoconference is going to be held.
DCoPs would benefit from systems scheduling and supporting face-to-face events if such
systems could perform similar tasks in virtual environments. Moreover, DCoPs could be
better supported if such systems could keep a profile of the community.
5.3 Asynchronous interactions
Asynchronous interactions in DCoPs usually occur through emails, discussion boards, wikis
and blogs. So, in this category, we include agents and multi-agent systems that can help users
use such tools more efficiently.
In order to provide an overview of the available tools, we classified them according to
their major features. In the Asynchronous Interactions category, we identified four major
features: E-mail Intelligent Forward, Forum of Avatars, Analysis of Discussion Boards and
Virtualized Ego Interaction (Table 3).
5.3.1 Descriptions of the systems
The Discovery E-mail (D-mail) system can send an email to the “appropriate” recipients
by analyzing their web pages. The system analyzes the similarity between the email topic
and the profile of the possible recipients. Such profiles are defined by an agent that analyzes
the possible recipients’ web pages [64]. Such a system could be used in different ways for
supporting communities. One possibility is the identification of potential members when a
community is being formed. Another possibility is identification of sub-communities inside
a large community. One limitation of the D-mail system is the profile used to define who
should receive the email.
EgoChat was developed to allow the sharing of community tacit knowledge through narra-
tives. Each member of the community is represented by a virtualized-ego (VE) that interacts
with other members’ VEs. A member accesses community tacit knowledge by interacting
with VEs and observing the exchange of messages among them. VEs exchange messages
with voices and gestures. A structured list is used to describe personal tacit knowledge [57].
Newcomers in a community could use this system to learn about the community domain,
accelerating their process of integration. The system can also help newcomers feel more
comfortable interacting with their VE before posting questions to the community discussion
board.
TelMeA is an asynchronous communication system for communities, and works like a
forum. The difference is that messages posted in this system can include information to
express the sender’s affective state (e.g. happy), interpersonal attitude (e.g. acting friendly
or respectfully), reference documents, and comments about such documents. TelMeA has an
123
Auton Agent Multi-Agent Syst
editor that allows users to include this kind of information into the messages. Each user is
represented by an agent that sends both, the textual and non-textual content. As an avatar,
the agent communicates non-textual content by facial expressions (e.g. smiling, nodding).
Moreover, TelMeA can analyze the non-textual content to provide a summary of evaluations
made by other users [110,111]. A system like TelMeA can be useful for CoPs because it
allows users to express themselves in a more comprehensive way. Furthermore, the analysis
of the non-textual content can help users to understand the community dynamics. One limi-
tation of such a system is the need of a specific editor to include non-textual information in
the message.
CommunityBoard is a system that displays the structure of a discussion graphically. It
is formed by a Personal Assistant for each user and a Community Agent. The Community
Agent classifies the messages of the discussion according to criteria such as the subject and
the sender’s reputation. The Personal Assistant presents the structure of the current discus-
sions graphically according to its user’s preferences and some criteria such as the author of
the message, the subject and the relevance of the message. Each message of the discussion
is represented by an icon and its relevance is represented by the type, the position and the
shading of the icon. In this way, users can, for example, decide in which discussions to par-
ticipate, guess who is more knowledgeable on a given topic, visualize the relation among
subjects, and identify the relevance of each topic for the community [44]. CommunityBoard
is a community-oriented system than could help community members comprehend discus-
sions and their context. It can also help a member participate just in discussions in which he
can contribute effectively. Moreover, community coordinators can use the system to monitor
the dynamics of the DCoP. The idea of the CommunityBoard could be applied to other means
that DCoP members use to interact, such as blogs and wikis.
5.3.2 Discussion
In the research for this review, we were not able to find any agents or multi-agent systems
applied to blogs and wikis. Since blogs and wikis can be useful for DCoPs, we believe, they
should be taken into account in the research concerning systems to support DCoPs. Agents
and multi-agent systems could either support users using blogs and wikis or they could use
blogs and wikis to build profiles to personalize the information retrieval. An agent or MAS
similar to CommunityBoard could also analyze blogs and wikis to indicate the relevance of
their contents.
We advocate that systems that, like the CommunityBoard, analyze the asynchronous com-
munication among DCoP members could be useful for three main reasons: they could support
members performing an activity; they could help the DCoP coordination observe what is hap-
pening; and they could be used to study the dynamics of a community. We consider that tools
allowing researchers to study DCoPs and virtual communities are an interesting research
opportunity. Usually, asynchronous communications in DCoPs leave traces (exchanged email
messages, blogs, wikis, etc.) that are a potentially rich material to be analyzed. The amount
of material could be intimidating because virtual communities tend to be large, thus powerful
computational tools to analyze it might be necessary.
It is difficult for DCoPs to be “present” in the daily routine of their members. Asyn-
chronous communications could be used to minimize such lack of “presence.” Agents could
monitor discussion boards, blogs and wikis to generate an indicator of the activity in the
community.
We consider that the content generated by both synchronous and asynchronous communi-
cations represents a considerable research material that should be better explored. The study
123
Auton Agent Multi-Agent Syst
Tab l e 4 Major features
identified in the Publishing
category
Major feature Agents and multi-agent systems
Geographically Co-located
Community Information
Dissemination
LiMe [99]
Elaboration of Newsletters CLELIA [3]
Bookmark and Bibliographic
References Management
CoWing [52]
CommunityItemsTool [56]
Web Pages Collaborative Editing PAIS [113]
Recommender Systems /
Collaborative Filtering and
Retrieval
ACORN [16,40,69,72]
Community Search Assistant [35]
Collaborative Spider [19]
Implicit [10,11]
Information Semi-automatic
Capture
[108]
of virtual communities (including DCoPs) can lead to better tools for communities and a
better understanding of such a social phenomenon.
5.4 Publishing
The Publishing category includes systems that help publish documents and use document
repositories. The category was extended (considering Wenger et al. [120] original concep-
tion) to include tools to handle (create, save, manage, etc.) documents. So, in this paper, we
included other tools like systems to automatically capture information and organizational
memory systems.
In order to provide an overview of the available tools, we classified them according to their
major features. In the Publishing category, we identified six major features: Geographically
Co-allocated Community Information Dissemination, Elaboration of Newsletters, Web Pages
Collaborative Editing, Information Semi-automatic Capture, Recommender Systems/Collab-
orative Filtering and Retrieval, and Bookmark and Bibliographic References Management
(Table 4).
5.4.1 Descriptions of the systems
In the Publishing category we included two singular systems: LiMe and CLELIA.
Living Memory (LiMe) is a system developed to support geographically co-located com-
munities like neighborhoods. The goal of the system is to share information (e.g. local news,
“lamppost announces”) among community members. The system combines specific elec-
tronic devices and a multi-agent system. The employed electronic devices include personal
devices, like home PCs and mobile phones; and shared devices, like Interactive Tables and
Interactive Bus Stops. Interactive Tables are tables with which people sitting around can inter-
act with LiMe, as a collective use multimedia kiosk. The Interactive Bus Stop device allows
people to interact with the system while waiting for the bus. The MAS is composed of three
types of agents: Personal Service Agent (PSA), Location Service Agent (LSA) and Memory
Service Agent (MSA). The PSAs play three main roles: personal profile manager, acquain-
tance model manager and personal history manager. As a personal profile manager, the PSA
manages a community member’s individual profile. As an acquaintance model manager, it
stores the user’s social circle and his/her activities in the different communal locations. As a
123
Auton Agent Multi-Agent Syst
personal history manager, the PSA maintains a log of the user’s interactions with communal
content. LSA manages the model of the location where a static device is situated. It manages
the location profile and history as well as the profile of location neighborhood. MSA plays
the role of information disseminator, relevance monitor, archivist, garbage collector and link
manager [99].
LiMe has been developed for a co-located community in which the participation is more
an alignment instead of the engagement usually found in professional communities. But the
system presents an interesting way to handle individual and communal profiling. Moreover,
LiMe has an acquaintance model that could be useful for DCoP members that participate
in different communities. The acquaintances of a DCoP member could help to define him
and in this sense, they should be part of his profile. A model that maintains information
about the activities a member performs in a given location, could be helpful to personalize
the support to members when they enter a specific “spot” (e.g. a tool or a feature in a sys-
tem) in the community “virtual world” (i.e. the platform the community use to perform its
activities).
CLELIA is a MAS developed to facilitate the publication of information by performing
page editing automatically. It is aimed at overcoming limitations of the current publishing
sequential workflow by allowing information creation and its publication simultaneously. The
system is composed of three main agents: Section Agent, Page Agent and Element Agent.
Section Agents receive all the elements of the section to determine the number of pages that
will be used and to define the elements of a single page. Page Agents place the elements
into a page. Element Agents store information about each element [3]. CLELIA presents
features that can be useful for communities that are performing tasks like the elaboration of
newsletter, manuals or web sites. CLELIA seems to be limited to a restricted task and thus it
would be useful for DCoPs with a very specific profile.
Less complex than CLELIA, Proxy Agent-based Information Sharing (PAIS) can be used
to help communities to publish web pages that are being elaborated collaboratively. Its major
component is Wedit. Web browser-based Direct Editing (Wedit) allows users to edit HTML
text using a Web browser. PAIS, as a whole system, enables users who participate in a com-
munity to share information. A community member could edit a web page using Wedit and
other members could see the modifications through a Web browser. The system merges the
modifications with the original page (that is not modified) each time a community mem-
ber accesses it [113]. Communities could use the system to build web pages collectively.
For example, sub-communities could use systems such as PAIS to publish their activities.
Another advantage of the system is that it seems to be simple and inexpensive. PAIS could
be useful for DCoPs, but it seems to us that a wiki could be used to support the execution of
thesamekindoftask.
To share information using PAIS, users should make it available through web pages. But,
there are systems that can semi-automatically capture information and knowledge. Tacla and
Barthès [108] present a system that captures and makes available lessons learned during
the development of R&D projects. It also helps users organize their documents. A Personal
Agent (PA) is the interface between each user and the system. The user invokes services like
organizing documents or sending emails through a PA. The services themselves are imple-
mented by Service Agents (SA). As the SAs work in a structure called “coterie,” there is a
broker agent that can create a communication link between different “coteries.” A “coterie”
is a tightly linked group of agents [5,108]. An agent able to capture lessons learned from the
activities of a community member can be useful to build a “lessons learned” library to a given
community. It is a way to improve the efficiency of the process of experience reification. Such
reified experience, as lessons learned, can be used in the process of negotiation of meaning
123
Auton Agent Multi-Agent Syst
while the community is facing a similar situation. To reify the experience is important, but it
is also important to be able to adapt and modify such an experience. Thus, the system should
also allow community members to question and modify the learned lessons.
DCoPs can also be supported by a system that recommends documents. Agent-Based
Community-Oriented Retrieval Network (ACORN) is a community-oriented MAS that aims
at improving the communication among the members of a community. In this system, each
document or query is an agent. Each agent can introduce itself to users that could find its
information useful. The user can recommend the offered information to other recipients that
could also find it interesting. In doing so, ACORN mixes recommendations generated auto-
matically with those elaborated by humans [16,69,72]. A system like ACORN can be useful
to disseminate information among community members.
Gomez-Sanz and Pavon [40] also present a system to recommend documents. In such a
system, documents are evaluated using annotations made by some community members. If
the document evaluation is positive, the system recommends it to the whole community. To
organize the process of dissemination, the MAS uses a workflow. The process begins when
a community member indicates a document that is evaluated automatically, considering the
community profile. If the document is evaluated positively, it is evaluated by humans who
apply certain social rules. To achieve the ideal composition of the community, the system
can punish users with undesirable behavior (e.g. non evaluation or individual negative eval-
uation of a document with collective positive evaluation). Document recommendation can
be useful if integrated into the CoP practice. Although the idea of using a workflow could be
considered interesting to manage the flow of document dissemination, it seems inadequate
for a community to have a mechanism that tries to achieve an “ideal composition” of the
community. Diverging points of view do not mean inadequacy for a given community.
Another type of document recommendation systems is the one that relies on automatic
collaborative filtering. The Community Search Assistant allows a community to search the
web collaboratively. Each time a community member makes a query, the query is stored as
a graph. The stored queries can be linked by the relatedness of the documents found in the
search. When a community member is about to perform a new search she can follow the
network of queries. In this way, the collective knowledge, embedded in the network of que-
ries, can help a community member to articulate one’s information needs using appropriate
terminology [35].
Collaborative Spider is a MAS for collaborative information retrieval and web mining.
The system supports collaboration by sharing search sessions containing post-retrieval anal-
ysis. In a given group, each user controls a User Agent and the group shares a Collaborator
Agent and a Scheduler Agent. The User Agent can fetch and analyze web pages. It also shows
graphically the search and analysis results, and the annotations elaborated by group members
about the documents. The Scheduler Agent keeps a list with the retrieval and analysis tasks
scheduled by the users and performs such tasks. The Collaborator Agent recommends docu-
ments to users based on their profiles. It also keeps users search sessions, their profiles and the
track of their annotations. The annotations can be attached to the documents and consulted
by other users [19]. Community Search Assistant and Collaborative Spider could be useful
to support some communities not just because they could help members find information but
also because they could help newcomers understand the terminology and access references
adopted by the community.
Implicit [10,11] is a MAS that recommends documents based in the notion of Implicit
Culture. Implicit Culture, a generalization of collaborative filtering, helps a newcomer to a
community behave like an old-timer, but without having explicitly explained the knowledge
needed to do this. When a user sends a request to the system, it suggests specific information
123
Auton Agent Multi-Agent Syst
that was obtained by observing the old-timers’ past behaviors when requesting similar infor-
mation. Implicit combines results of a search engine (Google) with the information obtained
from the analysis of the users’ behavior.
Systems to recommend documents and to filter and retrieve information can facilitate the
dissemination of information inside a DCoP. They can also help accelerate the integration of
newcomers. When newcomers use such systems to search for information, they would use
the experience of the old timers that is embedded in the systems. A possible improvement
for this kind of system should consider that a person participates in several communities and
has her private interests.
Collaborative Web Indexing (CoWing) is a distributed collaborative bookmark system that
enables users to share their bookmarks. The system does not impose extra workload on the
users; it allows users to choose who can access their bookmarks, assists users in classifying
their bookmarks and recommends interesting bookmarks [52].
Like CoWing, CommunityItemsTool also allows community members to share book-
marks. Besides bookmarks, CommunityItemsTool enables users to share bibliographic ref-
erences and annotations about bookmarks and these references. Data can be published and
accessed through a web browser. The system also includes a collaborating filtering feature
and a subscription mechanism. The search can be made based on several attributes like: title,
author, categorization or user ratings. Users can also subscribe for recommendations based on
the same kind of attributes [56]. Systems like CoWing and CommunityItemsTool could help
newcomers in a community to access quickly the references used by the old-timers, therefore
accelerating their integration. Such systems would be improved if they could classify and
search for bookmarks and bibliographic information using information extracted from the
content of the documents themselves.
5.4.2 Discussion
In the publishing category, document recommendation, as well as collaborative filtering and
retrieval have been actively explored, but we consider that a relatively unexplored feature is
the (semi)automatic capture of information from documents and the traces left by the com-
munication between community members. Information extracted (semi)automatically from
such sources could be used to elaborate new documents [109].
Agents could be used to inspect synchronous and asynchronous communication between
members and support the elaboration of documents concerning the main points of the dis-
cussion. For example, agents could inspect the messages of a discussion board; extract some
answers for a question; and save them as a Frequent Asked Question.
5.5 Community cultivation
Communities of practice could be considered as living entities that go through a life cycle:
they start, grow, mature and die. In this context, it is pertinent to talk about cultivating com-
munities. Cultivating communities relates to accepting, supporting as well as understanding
the kind of role communities can play in an organization. Excess of management (and con-
sequent control) can undermine and kill a community as well as the lack of support for
community activities [119]. Following that reasoning we chose tools that could help culti-
vate communities during their lifecycle. The Community Cultivation category includes tools
to form communities by putting together people with similar profiles.
In order to provide an overview of the available tools, we classified them according to
their major features. In the community cultivation category, we identified four major features:
123
Auton Agent Multi-Agent Syst
Tab l e 5 Major features
identified in the “Community
Cultivation” category
Major feature Agents and multi-agent systems
Community Presence Indicator Social Web Cockpit [41]
Agent-Buddy [126]
Community Formation /
Identification
Let’s Browse [63]
[101–104]
Community Organizer
[44,127,128]
MEMOIR [26]
[117]
[125]
[124]
MARS [129,130]
Support to Communities’
Activities
Answer Garden 2 [1]
ICC [107]
eLogbook [34]
KEEx/IVisTo/KARe [94]
Practice Dissemination K-InCA [85]
Community Presence Indicator, Community Formation/Identification, Support to Commu-
nities Activities and Practice Dissemination (Table 5).
5.5.1 Descriptions of the systems
K-InCA is a MAS designed to help users acquire and adopt knowledge sharing practices. The
system assists the user during all the adaptation process, starting by the presentation of the
desired behaviors and ending with the users’ practice within the community. Each user has an
agent that intervenes in a personal and contextualized way. Such a personal agent observes the
user’s behavior and makes suggestions, introduces concepts, and proposes activities in order
to achieve the adoption of sharing practices [85]. We consider that K-InCA could facilitate
the adoption of knowledge sharing practices, but such a contribution could be limited. It may
be difficult to make some practices adopted by communities explicit.
We have identified several tools that could help identify and form CoPs. Although classed
as tools for Synchronous Interactions; Community Space and Freewalk could be classified
as tools for Community Cultivation as well. Another tool that could be classified into two
categories is Let’s Browse [63]. It was mentioned in the individual participation category,
but we think that it could also be considered as a tool for cultivating communities, specifi-
cally helping identify people with similar interests during an event. Let’s Browse is a system
that allows collaborative web browsing. The system can sense when a person approaches its
screen through some special tags. After identifying the people near the screen, the system
uses their personal profiles to build an interest profile of the group. Then, based on such
a collective profile, Let’s Browse recommends web pages and also tells why it chose that
specific page. As mentioned, such a system is capable of working as an icebreaker.
Like Let’s Browse, the system of [101,103,104] was designed to be deployed in events like
conferences or workshops. The system is composed of Personal Digital Assistants (PDAs),
infrared badges, Internet connected kiosks terminals and a website. The PDAs can guide the
123
Auton Agent Multi-Agent Syst
participants in the conference. They can be connected to the kiosks by an infrared interface.
There is also a server with several infrared sensors to detect tags’ IDs. The PDAs can offer
services like browsing the conference program or recommending presentations. The system
uses the infrared badges to follow the user, tracing a touring diary (where he/she had been, the
attended presentations). Also using the badge, the user can mark a presentation that he found
interesting. The information kiosks provide a site map and a semantic map allowing users to
browse the semantic relationships among papers, keywords and participants. One of the most
interesting features for communities is the AgentSalon, a specialized kiosk that facilitates
face-to-face interactions [102]. To stimulate face-to-face interaction, the system promotes
chats among personal agents with touring diaries that present some points in common. In
this chats, agents can recommend each other presentations or can share the evaluation of a
presentation. This system can also be helpful when trying to form communities.
Let’s Browse and the systems designed by Sumi and Mase [101,103,104] could be useful
to form and identify DCoPs, but they should be deployed in a face-to-face event such as a
conference. Such a characteristic limits their possibilities as effective tools to support DCoPs.
SHINE is a framework for the development of communication systems for networked
communities. It was used to develop two systems: Community Organizer and Gleams of
People. Gleams of People was described in the Synchronous Interaction category. Commu-
nity Organizer is a tool that can support the creation of new communities by providing a
virtual place where people sharing the same interests can meet and communicate. Users can
represent their interests in a graphic virtual space through an icon that is associated with a vec-
tor of keywords. If two vectors of keywords are similar, the icons are located near each other.
The center of the space also has a vector of keywords. As the user changes this vector, the
location of the icons changes as well. So users can identify other users with similar interests.
The system also allows users to attach messages to the icons and get in touch [44,127,128].
SHINE is an interesting framework because it facilitates the development of communication
tools for communities, therefore allowing the development of tools to fulfill specific commu-
nity needs. Community Organize can be useful to identify virtual communities in connected
groups.
The Managing Enterprise-scale Multimedia using an Open Framework for Information
Re-use (MEMOIR) system helps users access information and navigate through an intra-
net. It also can be used to find colleagues with similar interests. The system deploys agents
capable of supporting users using trails and links. Trails are sequences of documents used to
accomplish a task. Links could be made by users to associate different types of documents.
The system can analyze the data and suggest interesting documents or people who could
be contacted to achieve a certain task. It also allows users to create trails and links as well
[26]. MEMOIR could be classified in the Individual Participation category, but we consider
it useful for identifying communities, once it can identify people with similar interests by
analyzing their trails.
Another system [117] forms communities observing the behavior of the users. The system
monitors and analyzes the behavior of the users during search operations and then assembles
users (or their personal agents) with similar interests into communities. This way, the system
is able to provide better search results. The system could also help form communities, but it
seems more limited than MEMOIR or SHINE.
Yang et al. [125] developed a system that forms small student communities from a larger
group that uses a web-based educational system. It aims at promoting interaction among
students that share similar interests by grouping them into small communities. A student
profile (called Learning Experience Vector) is defined considering his behavior (e.g. the
time dedicated to study a document, the frequency of access to online resources) and the
123
Auton Agent Multi-Agent Syst
contents (e.g. documents) he accesses. Such a profile is elaborated by the Learner Agent that
follows the student inside the educational system. Group Agents form small communities by
exchanging information with Learner Agents and other Group Agents.
Yang et al. [124] developed an approach to mine distributed communities in networks.
The approach is not restricted to the analysis of social networks, but could be useful to
identify DCoPs. Their approach can be distinguished from the previous ones because it is
more appropriated for handling distributed and dynamically evolving networks. Each node
of the network is represented by an agent that operates autonomously and asynchronously.
Communities are mined by using the agent’s point of view. If we could associate such an
approach with a Personal Assistant agent, we could have a valuable tool to identify DCoPs
inside social networks.
The aim of the systems described above is to form (or to identify) communities by analyz-
ing similarities among different types of user profiles. But, it is not the only possible approach.
People recommendation can also be used to form communities. Multi-Agent Referral System
(MARS) is a system that recommends people. Each user controls an agent that stores one’s
profile and a model of one’s acquaintances. In order to send a question to acquaintances,
the user receives several recommendations from his agent and he can pick some of them for
whom the question is sent. The selected acquaintance agent analyzes the question and decides
whether to forward it or not to its user. If the acquaintance’s agent considers that the question
is not adequate for its user, it can recommend other people to the agent that generated the
question. The latter integrates such a recommendation to its own model of acquaintances and
decides to send the request or not to its new acquaintances. When the agent sends a question,
it evaluates the answer and updates the model of acquaintances [129,130].
Systems supporting the formation and identification of communities are useful during the
first phases of the community lifecycle. However, as the community evolves, other types of
tools become necessary. One system that could support established communities is the Social
Web Cockpit. Social Web Cockpit is a MAS that supports collaboration among members of
a established community. It can be used with a common browser and with Basic Support
for Cooperative Work (BSCW). When a community member is navigating in a web site, the
system can indicate who the other members doing the same are, and if the web site belongs
to the community repertory. The system also provides information allowing members nav-
igating the same web site to get in touch synchronously (e.g. through a chat). The Cockpit
can facilitate the access to BSCW workspaces with collections of documents or links. Other
features in the tools are: user rating of web sites and documents; collaborative search based
on other member’s searches and creation of a vocabulary for thecommunity [41]. Social Web
Cockpit is a tool that supports users to access and manage several information sources like
the web and BSCW. Beyond the role of facilitating the access to documentation, this system
could be useful to make community members aware of their participation in a community.
As the Cockpit window is always shown on the screen, it could help cultivate the community,
working as a constant reminder of the community activity.
Another tool that can make the user “feel the presence” of the community is Agent-Buddy.
Agent-Buddy could be used with a Computer Support Collaborative Work (CSCW) tool to
provide a sense of “togetherness” but keeping the privacy of each user, as the agent distrib-
utes selectively the information in the community. The aim of the selective distribution is
to keep a “distance” among members. Authors use as analogy the physical distance or the
attitude members keep during meetings in co-located communities [126]. We consider that
awareness is important to keep distributed communities working. As the distance tends to
weaken the links among community members, tools like the Agent-Buddy and Social Web
Cockpit could be useful to partially compensate for it.
123
Auton Agent Multi-Agent Syst
Answer Garden can be considered as a system to cultivate communities because it can
help in the integration of newcomers (by answering questions about the community domain)
and provide old-timers a space for discussion. Answer Garden 2 is a system architecture for
organizational memory and collaborative help support. As its predecessor, Answer Garden,
it helps distributed users through a database of commonly asked questions (FAQ). By using
Answer Garden, a user can look for the answer to common questions in a database and if
he is not satisfied with the answer, he can post his question to an expert indicated by the
system. Then the expert can insert the new answer into the database. Answer Garden 2 is
more flexible and can be applied to any information system. When a user posts a question to
his Agent Garden 2 client, an escalation agent will forward the questions to several instances
(e.g. chat, bulletin board, help desk, human specialist) following previously defined policy
(e.g. first the chat, then the bulletin board, and so on). Yang [123] presents a system that
can also retrieve FAQ, but Answer Garden 2 has a more collective character because it pro-
vides features allowing encounters between newcomers and old-timers. Answer Garden 2
also has features that allow groups of people to collaborate in building answers and infor-
mation repositories [1]. Answer Garden 2 could be very useful for communities because it
has tools for communication and tools allowing the capture and refinement of the knowledge
created during the activities performed with such tools. It also facilitates the access to explicit
knowledge (databases) and to experts. The idea of an escalating agent seems interesting for
communities. Such a mechanism could be applied to distribute questions among members
with different types of participation. For example, a newcomer could post a question and this
question could be asked first to the other newcomers and after to the active members, only
then to the core group.
The eLogbook system aims at supporting the management of tacit and explicit knowl-
edge inside CoPs. It presents features for assets management, activities management and a
platform for discussions. eLogbook is based on the 3A model (Actors, Activities, Assets)
which considers three entities: “actors” that perform “activities” using “assets/artifacts.”
The system offers three types of graphical interface: context-aware interface, content
oriented interface and the two dimensional maps interface. The content-oriented interface
displays the list of entries for each entity (e.g. list of actors, list of activities). The two dimen-
sional interface presents the same lists but in a network format to highlight the relations
among entries. The content-aware interface integrates three regions, each one dedicated to
one of the entities (actor, activities, and assets). When the user chooses an entry (a per-
son, a resource or an activity), the correspondent region is dislocated to the center of the
workspace. The other regions are updated to highlight the entries that have a link with the
chosen entry. Each region presents icons that indicate which operations are available in
the region. With eLogbook, each user can create his own virtual workspace with the neces-
sary entries (e.g. activities that the user is participating in, and people with whom the user is
working) [34].
Knowledge Enhancement and Exchange (KEEx) is an organizational learning environ-
ment that uses a peer-to-peer architecture. It allows users to search and share artifacts, even
without using a common representational language or an ontology. Each user makes public
her knowledge using a hierarchy of concepts (that give a context to an artifact) that contains
the pertinent artifacts. When searching for information in an artifact, the user elaborates the
request for the information using keywords or the context of the information. KEEx examines
other users’ hierarchy of concepts and returns a list of users, contexts, and artifacts. To do it,
the system uses a sophisticated algorithm to match the request and the available contexts. It
also counts on user profiles that are built by means of manned interviews and the observation
of users’ behavior (performed by the system through the personal assistant agent). KEEx is
123
Auton Agent Multi-Agent Syst
the foundation of two other tools: Interactive Visualization Tool (IVisTo) and Knowledge-
able Agent for Recommendations (KARe). IVisTo is a tool allowing the visualization of
the relevance of request results. The relevance of a result is determined by using the users’
social and semantic preferences. The social preferences are represented in the user profiles
by variables such as trust, position, etc. The semantic preferences are determined by KEEx.
The user can define the weight of the social and semantic factors in the computation of the
pertinence. KARe is a MAS designed to recommend both documents and people that can
provide the information requested by the user. It also uses social and semantic preferences
to elaborate the recommendations [94].
Intelligent Conversational Channel (ICC) is a system that can be used to share knowl-
edge about researchers in a given domain. It is based on the extraction of information from
documents, especially from scientific papers, and from the interactions among community
members. ICC allows users to interact with other users and agents. Thus, when a user asks
a question, the answer can come either from a user or from an agent. To answer questions,
agents use information extracted from documents and from the interaction among mem-
bers [107]. ICC could be useful to integrate newcomers to a CoP by helping them identify
knowledgeable community members, the relevance of their contributions, their partners,
etc.
5.5.2 Discussion
One can observe that the issue of systems being able to identify or form virtual communi-
ties has been studied actively. However, we conclude that the development of systems using
agents or multi-agent systems to support activities performed by DCoPs, such as KEEx and
eLogbook, could be further explored. The development of groupware dealing with a similar
set of issues provides a strong foundation for the development of systems to support DCoPs,
but still leaves some open issues. For example, one major concern in the study of DCoPs is
individual and collective learning. Trajectories represent individual learning and the evolu-
tion of community practice represents collective learning. We believe that such a concern is
still neglected in the development of systems supporting DCoPs.
Although potentially useful, the approach used to develop most of the systems to culti-
vate communities has shown some limits. We think that an approach based on the ability of
multi-agent systems to integrate different features (new or provided by legacy systems) by
using agent wrappers in an open environment could give birth to more appropriate systems.
As DCoPs go through a life cycle, they tend to change their needs with time. Systems inte-
grating new features or using agents to encapsulate legacy systems or features might help
deal with such evolving needs.
Agents and MASs could also provide another kind of integration: the integration of differ-
ent interfaces. Different from a typical situation when a DCoP member should use different
systems with different interfaces to accomplish his activities, a user could use a personal
agent to access different systems [109].
We conclude that there is a lack of features that support the coordination of DCoPs among
the systems surveyed in this paper. As DCoPs should be managed with a “light hand” [119],
one way to support DCoPs coordination is to generate information about, for example, the
participation of their members, trajectories or the evolution of the practice [87].
123
Auton Agent Multi-Agent Syst
Fig. 2 MAS applications to support communities
5.6 Applying Agents and MAS to support communities: a general vision
We presented several multi-agent systems and agents able to support distributed CoPs and
classified them into five categories: Individual Participation, Synchronous Interactions, Asyn-
chronous Interactions, Publishing and Community Cultivation. In each category, we grouped
a set of systems by their features. In order to provide a representative landscape of such sys-
tems, we graphically indicated the major features found within the ensemble of the analyzed
tools (Fig. 2) in the diagram proposed by Wenger et al. [120].
In order to link each system to its category, we built a table establishing the relations
between each system and one of the five categories through its features (Table 6).
5.7 Current and potential contributions to the support of DCoPs
The idea of this survey was to review a number of MAS applications and analyze in which
way they could support activities performed by DCoPs as well as to expose potentially useful
new features. In this sub-section, we summarize both issues: the contribution of the current
features (Table 7); and potentially useful new features (Table 8).
6 Concluding remarks
In this paper, we presented a review of agent and multi-agent systems containing features
that could be used to support communities. Such a review is not intended to be exhaustive
but rather comprehensive enough to present a landscape of the systems that are available or
could be developed for supporting of DCoPs.
123
Auton Agent Multi-Agent Syst
Tab l e 6 A landscape of MAS applications to support communities
Category Major feature Agents and multi-agent systems
Individual Participation Individual Profiling SUITOR [65]
Personalized Newspaper WebMate [20]
NewsAgent [38]
ETTS [15]
Support to Web Navigation [53]
[96]
FAQ Retrieval Yang [123]
Information Filtering/Retrieval AgentRAIDER [90]
Watson [14]
MAIS [29,30]
MAPIS [81]
[51]
[98]
[88]
[95]
Personal Search [37,38]
Several examples [55]
Synchronous Interactions Presence Indicator Gleams of People [127]
Conversational Agent for IM AINI [39]
Meeting Scheduling [22]
[59]
RESCH [62]
MSRAC [8]
[89]
Spontaneous Synchronous Encounters Contact Space [83]
Freewalk [49,75,76]
Smart Meeting Rooms Smart Conference Room [45]
EasyMeeting [21]
MITRE ETR [43]
Asynchronous Interactions E-mail Intelligent Forward D-mail [64]
Forum of Avatars TelMeA [111,110]
Analysis of Discussion Boards CommunityBoard [44]
Virtualized Ego Interaction EgoChat [57]
Publishing Geographically Co-located
Community Information
Dissemination
LiMe [99]
Elaboration of Newsletters CLELIA [3]
Bookmark and Bibliographic CoWing [52]
References Management CommunityItemsTool [56]
Recommender ACORN [69]
Systems/Collaborative [40]
and Retrieval Community Search Assistant [35]
123
Auton Agent Multi-Agent Syst
Tab l e 6 Continued
Category Major feature Agents and multi-agent systems
Collaborative Spider [19]
Implicit [10,11]
Web Pages Collaborative Editing PAIS [113]
Information Semi-automatic Capture [108]
Community Cultivation Community Presence Indicator Social Web Cockpit [41]
Agent-Buddy [126]
Community Formation / Identification Let’s Browse [63]
[101–104]
Community Organizer [44,127,128]
MEMOIR [26]
[117]
[125]
[124]
MARS [129,130]
Support to Communities’ Activities Answer Garden 2 [1]
ICC [107]
eLogbook [34]
KEEx/IVisTo/KARe [94]
Practice Dissemination K-InCA [85]
In order to organize the study, we adopted the categorization elaborated by Wenger et al.
[120] who established categories based on the usual kinds of activities that are performed
by communities: Synchronous Interactions, Asynchronous Interactions, Publishing, Com-
munity Cultivation, and Individual participation.
For each category, we grouped the identified systems by features (e.g. Personalized News-
letter, Community Formation/Identification, etc.). Then, we located each feature in the dia-
gram elaborated by Wenger et al. [120]. The idea is to facilitate the visualization of the
relations among categories and features. Finally, to link categories, features and systems we
built a table that summarizes the results of this review.
Despite the fact that most systems in the Individual Participation category do not consider
participation in groups, they could be useful as tools to perform individual activities while
participating in communities. User profiles are necessary to adapt a tool to the users and,
at the same time, they can also be used to identify and form communities. A personalized
newsletter or a system retrieving information from distributed sources could improve users’
efficiency when accessing information on personal interests or community domain. Even a
system oriented toward communities should consider the individual dimension. An individ-
ual can participate in several communities, and develop a different aspect of one’s identity in
each one of them. When he is engaged in the activity of a specific community, he “wears” the
correspondent aspect of his identity. Thus, even a system for communities should consider
individual participation.
The features identified in the Synchronous Interaction and Asynchronous Interaction cat-
egories could be useful for face-to-face interactions and technology mediated interactions,
respectively. Meeting scheduling and smart meeting rooms are systems dedicated to the
123
Auton Agent Multi-Agent Syst
Tab l e 7 A summary of contribution of the current features to the support of DCoPs
Major feature Current contribution
Individual Participation
Individual Profiling Profile builders can help provide personalized services to
community members
Personalized Newspaper Personalized newspapers and clippings could help DCoP
members keep track of modifications in information sources
regarding the community domain
Support to Web Navigation Systems to support web navigation could help DCoP members
find information in the World Wide Web more efficiently
FAQ Retrieval FAQ retrieval could be used to accelerate the integration of
newcomers in DCoPs
Information Filtering/Retrieval Information filtering and retrieval using different types of
profile can help community members find, in an easier way,
information required to perform communal activities
Synchronous Interactions
Presence Indicator Presence Indicators can be useful to reinforce the presence of
members of distributed communities. It can be used to prompt
spontaneous encounters
Conversational Agent for IM A conversational agent could be useful to animate chat
sessions. It could be also used to help integrate newcomers
Meeting Scheduling Meeting scheduling systems are useful to organize face-to-face
events organized by the DCoP
Spontaneous Synchronous Encounters Spontaneous Synchronous Encounter systems can mimic the
socializing effect of spontaneous encounters
Smart Meeting Rooms Smart Meeting Room systems are useful to organize
face-to-face events organized by the DCoP
Asynchronous Interactions
E-mail Intelligent Forward An E-mail Intelligent Forward system can be used to identify
potential members when a community is being formed. It can
also be used to identify sub-communities inside a large
community
Forum of Avatars The use of avatars can allow DCoP members to express
themselves in a more comprehensive way. The analysis of the
non-textual content (e.g. emotions) expressed by the avatars
can help users to understand the community dynamics
Analysis of Discussion Boards Analysis of Discussion Boards system could help DCoP
members to comprehend discussions and their context. It can
help them participate just in discussions in which they can
contribute effectively. Community coordinators can use the
system to monitor the dynamics of a DCoP
Virtualized Ego Interaction Virtualized Ego Interaction systems could be used by
newcomers trying to learn about the DCoP domain,
accelerating their process of integration. It can also help
newcomers feel more comfortable because they interact with
a virtualized ego before interact with community old timers
Publishing
Geographically Co-located
Community Information
Dissemination
Elaboration of Newsletters Systems to elaborate newsletters can be useful for communities
that are performing tasks like the elaboration of newsletter,
manuals or web sites
Bookmark and
Bibliographic References
Management
Bookmark and Bibliographic References Management systems
could help newcomers in a DCoP access quickly the
references used by the old-timers, therefore accelerating their
integration
123
Auton Agent Multi-Agent Syst
Tab l e 7 Continued
Major feature Current contribution
Web Pages Collaborative Editing Systems to build web pages collectively can be useful for
DCoPs. Sub-communities could use such systems to publish
their activities
Recommender
Systems/Collaborative
Filtering and Retrieval
Systems to recommend documents and to filter and retrieve
information can facilitate the dissemination of information
inside a DCoP. They can also help accelerate the integration
of newcomers. When newcomers use such systems to search
for information, they could use the experience of the old
timers that is embedded in the systems
Information Semi-automatic Capture A system to capture information semi-automatically can be
used to capture information about the activities that DCoP
members are currently performing. It could be used to create
a library of lessons learned or best practices
Community Cultivation
Community Presence Indicator Systems that indicate the presence of the community could be
useful to make members aware of their participation in a
community. They could help cultivate the community,
working as a constant reminder of the community activity
Community Formation/Identification Systems supporting the formation and identification of
communities are useful during the first phases of the
community lifecycle
Support to Communities’ Activities Systems to support communal activities offer features that
allow DCoP to perform its activities
Practice Dissemination A practice dissemination system could facilitate the adoption of
knowledge sharing practices
face-to-face interactions, while forums of avatars and analysis of forums are features for
computer mediated interactions. Community oriented systems should consider integrating
tools to enable both kinds of interaction since communities promote both local and computer
mediated events. In other words, community oriented systems should consider the constraints
imposed by temporal and spatial dimensions.
Although the systems that compose the Publishing category were not conceived specifi-
cally for CoPs, they can be useful to facilitate the negotiation of meaning inside a community.
The negotiation of meaning is composed of two processes: the reification and the participa-
tion. Systems in the Publishing category could support the reification process by capturing
information semi-automatically or by supporting the collaborative elaboration of a news-
letter. They could also facilitate the participation by recommending documents and people.
Although reification is not limited to the creation of documents and participation is not just
retrieving them, systems in the publishing category could still contribute to the negotiation
of meaning process.
Some of the systems in the Cultivating Communities category, such as Answer Garden 2
and Social Web Cockpit, are not specifically designed for communities, but they were con-
ceived as groupware to support collaborative work. Although KEEx and eLogbook also use
notions from the CSCW domain, they are more specific for communities. They emphasize the
relations among people, artifacts and activities. KEEx presents the relations among people
considering the similarity of interests and the trust. eLogbook constantly presents the relation
among people, artifacts and activities. We consider that the emphasis on such relations might
foster the development of more adequate systems for communities.
123
Auton Agent Multi-Agent Syst
Tab l e 8 A summary of the new features that could contribute to the support of DCoPs
Major feature Possibly useful new features
Individual Participation
Individual Profiling To improve users’ profiles, models and preferences
considering other dimensions of the notion of identity
such as multi-membership and trajectories
Personalized Newspaper To use profiles that better represent other dimensions of
the notion of identity such as multi-membership and
trajectories
Support to Web Navigation Systems to support web navigation could be more useful
if they could offer the same kind of services (e.g.
bookmark maintenance, verification of updates,
suggestion of web pages) for the community as a
whole
FAQ Retrieval Systems to extract FAQs automatically or
semi-automatically from sources such as community
forums or chats
Information Filtering/Retrieval To use profiles that better represent other dimensions of
the notion of identity such as multi-membership and
trajectories
Synchronous Interactions
Presence Indicator Personal Assistant Agents could be an option to
implement tools to unify the interface to all Internet
systems that require information about presence and
status
Conversational Agent for IM
Meeting Scheduling Meeting scheduling systems could be more useful if
adapted to facilitate the organization of on-line events
(e.g. chats) organized by the DCoP
Spontaneous Synchronous Encounters Spontaneous Synchronous Encounter systems could be
extended to promote such kind of encounter in open
systems like the Internet
Smart Meeting Rooms Smart Meeting Room systems could be more useful if
adapted to facilitate the organization of on-line events
(e.g. chats) organized by the DCoP
Asynchronous Interactions
E-mail Intelligent Forward To use profiles that better represent other dimensions of
the notion of identity such as multi-membership and
trajectories
Forum of Avatars The forum of avatars should include an integrated editor
for textual and non-textual information
Analysis of Discussion Boards Other means that DCoP members use to interact such as
blogs and wikis could be analyzed
Virtualized Ego Interaction
Publishing
Geographically Co-located Community
Information Dissemination The system that presents such a feature has an
acquaintance model for each user. Such a model
represents the social circle and user’s activities in the
different communal locations. An acquaintances
model could be useful for DCoP members that
participate in different communities
Elaboration of Newsletters The system with this feature seems to be limited to a
restricted task and thus it would be useful for DCoPs
with a very specific profile
123
Auton Agent Multi-Agent Syst
Tab l e 8 Continued
Major feature Possibly useful new features
Bookmark and Bibliographic References
Management
Bookmark and Bibliographic References
Management systems would be improved
if they could classify and search for
bookmarks and bibliographic information
using information extracted from the
content of the documents themselves
Web Pages Collaborative Editing It seems to us that a wiki could be used to
support the execution of the same kind of
task
Recommender Systems/Collaborative
Filtering and Retrieval
Recommender systems and collaborative
filtering and retrieval systems could be an
improvement in considering that a person
participates in several communities and
has its private interests
Information Semi-automatic Capture A system to capture information
semi-automatically should be also able to
allow community members to question and
modify such information
Community Cultivation
Community Presence Indicator
Community Formation/Identification Systems supporting the formation and
identification of communities could use
profiles that better represent other
dimensions of the notion of identity such
as multi-membership and trajectories
Support to Communities’ Activities Systems to support communal activities
should be able to integrate systems and
features in an open environment. They
should be designed using abstractions like
trajectories and negotiation of meaning
that are presented in the CoPs’ theoretical
framework
Practice Dissemination A practice dissemination system is limited
by the kind of practices that could be made
explicit in the system
It is possible to observe a significant number of agents or multi-agent systems belonging
to the Individual Participation and to the Publishing categories. However, the Synchronous
Interactions and Asynchronous Interactions categories are not well covered by the MAS
approach, especially for distributed communities.
In the Community Cultivation category we find interesting systems, especially for com-
munity formation/identification. We consider that, in this category, there are some important
types of community activity that are still unsupported such as community coordination, com-
munity reflection, and gathering of community indicators. In addition, in this survey we did
not find tools that enable researchers to study distributed communities.
When analyzing the map developed in this review, we consider that agents or multi-agent
systems are good candidates for supporting DCoPs, for example through personalized infor-
mation retrieval, meeting scheduling, discussion board analysis, or community formation or
identification. Clearly most such features are also proposed by traditional (non agent) prod-
ucts. However, there does not seem to be any reasonable way to combine such traditional
123
Auton Agent Multi-Agent Syst
products easily. On the contrary, an agent approach can enable this integration, while adding
agent wrappers to encapsulate features or systems, and personal agents to unify interfaces.
Moreover, agents and multi-agent systems could be used to provide new features. We
consider that users’ profiles better reflect the notion of identity if a MAS is used to represent
the multiple aspects of the identity. Indeed, it is possible to envision a MAS in which each
agent would represent one aspect of an identity and would negotiate the importance of this
aspect in a given context. Identities also change in time, thus profiles or user models should
consider their temporal dimension. Profiles should be updated with current information about
their user and obsolete information should be eliminated.
Agents could also be used to follow DCoPs members in order to gather data to represent
some aspects of a member’s trajectory in a community. Data collected by agents could be
also used to partially describe the evolution of the practice of a community. As DCoPs usu-
ally interact through technological tools, their members leave traces of their activities. Data
can be gathered from forums, discussion boards, blogs, wikis, chats, etc. We think that the
availability of such information constitute a valuable research material. Moreover, such data
could be used to support members performing an activity, or help the DCoP coordination
observe what is happening, or build users’ profiles or models.
Since DCoPs meet in virtual environments, systems for scheduling meetings and control-
ling meeting rooms would be more useful if they could be applied to on-line events such as
chats, web seminars or videoconferences.
Agents and multi-agent systems should also be applied to capture information automat-
ically or semi-automatically. For example, they could extract some answers to a question
posted in a discussion board or in a chat and save them as Frequently Asked Questions.
We consider that several aspects of the development of agents and multi-agent systems to
cultivate DCoPs should be investigated. In particular: the development of an integrated open
MAS to support DCoP activities; and features in such a system for managing its coordination.
Such examples illustrate the possibilities opened by MAS applications and agent technol-
ogy.
In the context presented above, we consider that the association between agent and multi-
agent technologies and DCoPs have the potential to bear fruits that could affect, in the long
term, the way humans and artificial agents interact.
Acknowledgment Gilson Yukio Sato was supported by the Program AlBan, the European Union Program
of High Level Scholarships for Latin America, scholarship no.E04D032545BR and by the Federal University
of Technology—Paraná (Brazil). The authors are also grateful for the reviewers’ comments that contributed
to greatly improve this paper.
References
1. Ackerman, M. S., & McDonald, D. W. (1996). Answer Garden 2: Merging organizational memory with
collaborative help. In Proceedings of the 1996 ACM conference on computer supported cooperative
work (pp. 97–105). Boston, MA, USA: ACM Press.
2. Agogino, A., & Tumer, K. (2010). A multiagent approach to managing air traffic flow. Autonomous
Agents and Multi-Agent Systems, 1–25.
3. Arranz Matia, A. L., & Sanz-Bobi, M. A. (2005). CLELIA: A multi-agent system for publishing
printed and electronic media. Expert Systems with Applications, 28(4), 725–734.
4. Barthès, J.-P. (2003). MASH environments for corporate KM. In Proceedings of the international
joint conference on artificial intelligence 2003 KM-OM workshop. Acapulco, Mexico.
5. Barthès, J.-P. A., & Tacla, C. A. (2002). Agent-supported portals and knowledge management in
complex R&D projects. Computers in Industry, 48(1), 3–16.
123
Auton Agent Multi-Agent Syst
6. Barton, D. & Tusting, K. (Eds.). (2005). Beyond communities of practice. Learning in doing: Social,
cognitive and computational perspectives. Cambridge: Cambridge University Press.
7. Bellifemine, F., Caire, G., et al.(2007). Developing multi-agent systems with JADE. Chichester: Wiley.
8. BenHassine, A., & Ho, T. B. (2007). An agent-based approach to solve dynamic meeting scheduling
problems with preferences. Engineering Applications of Artificial Intelligence, 20(6), 857–873.
9. Berners-Lee, T., Hendler, J., et al. (2001). The Semantic Web: A new form of Web content that is
meaningful to computers will unleash a revolution of new possibilities. Scientific American Magazine,
34–43.
10. Birukou, A., Blanzieri, E., et al. (2005). Implicit: An agent-based recommendation system for
web search. In Proceedings of the fourth international joint conference on autonomous agents and
multiagent systems (pp. 618–624). The Netherlands: ACM.
11. Birukou, A., Blanzieri, E., et al. (2010). Implicit: A multi-agent recommendation system for web
search. Autonomous Agents and Multi-Agent Systems, 1–34.
12. Bradshaw, P., Powell, S., et al. (2004). Building a community of practice: Technological and social
implications for a distributed team. In P. Hildreth & C. Kimble (Eds.), Knowledge networks: Innovation
through communities of practice (Vol. 1, pp. 184–201). Hershey: Idea Group Publishing.
13. Brown, J. S., & Duguid, P. (2000). Organizational learning and communities of practice: Toward
a unified view of working, learning, and innovation. In E. L. Lesser, M. A. Fontaine, &
J. A. Slusher (Eds.), Knowledge and communities (pp. 99–118). Boston: Butterworth Heinemann.
14. Budzik, J., Hammond, K. J., et al. (2001). Information access in context. Knowledge-Based
Systems, 14(1–2), 37–53.
15. Bun, K. K., & Ishizuka, M. (2006). Emerging topic tracking system in WWW. Knowledge-Based
Systems, 19(3), 164–171.
16. Carter, J., Ghorbani, A. A., et al. (2002). Just-in-time information sharing architectures in multi-
agent systems. In Proceedings of the first international joint conference on autonomous agents and
multiagent systems: Part 2 (pp. 647–654). Bologna, Italy: ACM Press.
17. Case, S., Azarmi, N., et al. (2001). Enhancing e-communities with agent-based systems. Com-
puter, 34(7), 64–69.
18. Chaib-Draa, B., Jarras, I., et al. (2001). Systèmes multi-agents : principes généraux et applications. In
J.-P. Briot & Y. Demazeau, Principes et Architecture des Systèmes Multi-Agent. (pp. 27–70). Paris: Her-
mès Science Publications.
19. Chau, M., Zeng, D., et al. (2003). Design and evaluation of a multi-agent collaborative web mining
system. Decision Support Systems, 35(1), 167–183.
20. Chen, L., & Sycara, K. (1998). WebMate: A personal agent for browsing and searching. Proceedings
of the second international conference on autonomous agents (pp. 132–139). Minneapolis, MN,
USA: ACM Press.
21. Chen, H., Perich, F., et al. (2004). Intelligent agents meet semantic web in a smart meeting room.
In Proceedings of the third international joint conference on autonomous agents and multiagent
systems—Vol. 2 (pp. 854–861). New York, NY: IEEE Computer Society.
22. Chun, A., Wai, H., et al. (2003). Optimizing agent-based meeting scheduling through preference
estimation. Engineering Applications of Artificial Intelligence, 16(7–8), 727–743.
23. Cox, A. (2005). What are communities of practice? A comparative reviewof four seminal works. Journal
of Information Science, 31(6), 527–540.
24. Crawford, E., & Veloso, M. (2007). An experts approach to strategy selection in multiagent meeting
scheduling. Autonomous Agents and Multi-Agent Systems, 15(1), 5–28.
25. d’Inverno, M., & Luck, M. (2001). Understanding agent systems. Berlin: Springer.
26. De Roure, D., Hall, W., et al. (2001). MEMOIR—an open framework for enhanced navigation of
distributed information. Information Processing & Management, 37(1), 53–74.
27. Durfee, E. H., & Lesser, V. R. (1989). Negotiating task decomposition and allocation using partial
global planning. In L. Gasser & M. Huhns (Eds.), Distributed artificial intelligence (pp. 229–
244). London: Pitman Publishing.
28. Ellis, C. A., Gibbs, S. J., et al. (1991). Groupware: Some issues and experiences. Communications
of the ACM, 34(1), 39–58.
29. Enembreck, F., & Barthès, J. P. (2004). MAIS - Un système multi-agents pour la recherche d’infor-
mation sur le web. In B. Rumpler & J.-M. Pinon (Eds.), Fouille de Textes et Organisation de
Documents (Vol. 8, pp. 83–106). Paris: Lavoisier.
30. Enembreck, F., & Barthès, J.-P. (2005). ELA—a new approach for learning agents. Autonomous
Agents and Multi-Agent Systems, 10(3), 215–248.
31. Fan, X., & Yen, J. (2004). Modeling and simulating human teamwork behaviors using intelligent
agents. Physics of Life Reviews, 1(3), 173–201.
123
Auton Agent Multi-Agent Syst
32. Gasser, L. (1991). Social conceptions of knowledge and action: DAI foundations and open systems
semantics. Artificial Intelligence, 47(1–3), 107–138.
33. Gherardi, S. (2005). Organizational knowledge: The texture of workplace learning. Oxford: Blackwell.
34. Gillet, D., El Helou, S., et al. (2008). Turning Web 2.0 social software into versatile collaborative
learning solutions. In First international conference on advances in computer-human interaction.
35. Glance, N. S. (2001). Community search assistant. In Proceedings of the 6th international conference
on intelligent user interfaces (pp. 91–96). Santa Fe, NM, USA: ACM Press
36. Glass, A., & Grosz, B. J. (2003). Socially conscious decision-making. Autonomous Agents and
Multi-Agent Systems, 6(3), 317–339.
37. Godoy, D., & Amandi, A. (2006). Modeling user interests by conceptual clustering. Information
Systems, 31(4–5), 247–265.
38. Godoy, D., Schiaffino, S., et al. (2004). Interface agents personalizing Web-based tasks. Cognitive
Systems Research, 5(3), 207–222.
39. Goh, O. S., Fung, C. C., et al. (2008). Domain knowledge query conversation bots in IM. Knowl-
edge-Based Systems, 21(7), 681–691.
40. Gomez-Sanz, J. J., & Pavon, J. (2003). Personalized information dissemination using agents orga-
nizations. In Proceedings of the ninth IEEE workshop on future trends of distributed computing
systems. FTDCS 2003.
41. Gräther, W., & Prinz, W. (2001). The social web cockpit: Support for virtual communities. In:
Proceedings of the 2001 international ACM SIGGROUP conference on supporting group work (pp.
252–259). Boulder, CO, USA: ACM Press.
42. Grudin, J. (1994). Computer-supported cooperative work: History and focus. Computer, 27(5), 19–26.
43. Harper, L. D., Gertner, A. S., et al. (2004). Perceptive assistive agents in team spaces. In Proceedings
of the 9th international conference on intelligent user interface (pp. 253–255). Funchal, Madeira,
Portugal: ACM Press.
44. Hattori, F., Ohguro, T., et al. (1999). Socialware: Multiagent systems for supporting network
communities. Communications of the ACM, 42(3), 55.
45. Hellenschmidt, M., & Kirste, T. (2004). Software solutions for self-organizing multimedia-appli-
ances. Computers & Graphics, 28(5), 643–655.
46. Hildreth, P., & Kimble, C. (2004). Knowledge networks: Innovation through communities of prac-
tice. Hershey: Idea Group Publishing.
47. Hughes, J., Jewson, N., et al. (Eds.). (2007). Communities of practice: Critical perspectives. London:
Routledge.
48. Isbell, C., Kearns, M., et al. (2006). Cobot in LambdaMOO: An adaptive social statistics agent. Auton-
omous Agents and Multi-Agent Systems, 13(3), 327–354.
49. Isbister, K., Nakanishi, H., et al. (2000). Helper agent: Designing an assistant for human-human
interaction in a virtual meeting space. In Proceedings of the SIGCHI conference on human factors
in computing systems (pp. 57–64). The Hague, The Netherlands: ACM Press.
50. Jennings, N. R., Sycara, K., et al. (1998). A roadmap of agent research and development. Journal
of Autonomous Agents and Multi-Agent Systems, 1(1), 7–38.
51. Ji, Y. G., & Salvendy, G. (2002). A metadata filter for intranet portal organizational memory
information systems. International Journal of Human-Computer Studies, 56(5), 525–537.
52. Kanawati, R., & Malek, M. (2002). A multi-agent system for collaborative bookmarking. In Pro-
ceedings of the first international joint conference on autonomous agents and multiagent systems:
Par t 3 (pp. 1137–1138). Bologna, Italy: ACM Press.
53. Keeble, R. J., & Macredie, R. D. (2000). Assistant agents for the world wide web intelligent interface
design challenges. Interacting with Computers, 12(4), 357–381.
54. Kimble, C. (2006). Communities of practice: Never knowingly undersold. In Innovative approaches
for learning and knowledge sharing: EC-TEL 2006 workshops proceedings, Crete, Grece.
55. Klusch, M. (2001). Information agent technology for the Internet: A survey. Data & Knowledge
Engineering, 36(3), 337–372.
56. Koch, M., Lacher, M., et al. (2001). The CommunityItemsTool-interoperable community support
in practice. In Proceedings of the tenth IEEE international workshops on enabling technologies:
Infrastructure for collaborative enterprises. WET ICE 2001.
57. Kubota, H., Nishida, T., et al. (2000). Exchanging tacit community knowledge by talking-virtual-
ized-egos. In Proceedings of the fourth international conference on autonomous agents. Barcelona,
Spain: ACM Press.
58. Lave, J., & Wenger, E. (1991). Situated learning: Legitimate peripheral participation. Cambridge: Cam-
bridge University Press.
123
Auton Agent Multi-Agent Syst
59. Lee, C.-S., & Pan, C.-Y. (2004). An intelligent fuzzy agent for meeting scheduling decision support
system. Fuzzy Sets and Systems, 142(3), 467–488.
60. Lee, C.-S., Jiang, C.-C., et al. (2006). A genetic fuzzy agent using ontology model for meeting
scheduling system. Information Sciences: Softcomputing Meets Agents, 176(9), 1131–1155.
61. Lee Lock, L., & Neff, M. (2004). How information technologies can help build and sustain an
organisation’s CoP: Spanning the socio-technical divide?. In P. Hildreth & C. Kimble (Eds.), Knowl-
edge networks: Innovation through communities of practice (pp. 165–183). Hershey: Idea Group
Publishing.
62. Lemaître, C., & Excelente, C. B. (1998). Multi-agent network for cooperative work. Expert Systems
with Applications, 14(1–2), 117–127.
63. Lieberman, H., van Dyke, N., et al. (1999). Let’s browse: A collaborative browsing agent. Knowl-
edge-Based Systems, 12(8), 427–431.
64. Loia, V., Senatore, S., et al. (2004). Combining agent technology and similarity-based reasoning for
targeted e-mail services. Fuzzy Sets and Systems, 145(1), 29–56.
65. Maglio, P. P., Campbell, C. S., et al. (2001). An architecture for developing attentive information
systems. Knowledge-Based Systems, 14(1–2), 103–110.
66. Malsch, T. (2001). Naming the unnamable: Socionics or the sociological turn of/to distributed artificial
intelligence. Autonomous Agents and Multi-Agent Systems, 4(3), 155–186.
67. Malsch, T., & Schulz-Schaeffer, I. (2007). Socionics: Sociological concepts for social systems of
artificial (and human) agents. Journal of Artificial Societies and Social Simulation, 10(1), 11.
68. Manvi, S. S., & Venkataram, P. (2004). Applications of agent technology in communications: A
review. Computer Communications, 27(15), 1493–1508.
69. Marsh, S., & Masrour, Y. (1997). Agent augmented community-information: The ACORN architec-
ture. In Proceedings of the 1997 conference of the Centre for Advanced Studies on Collaborative
Research Toronto (p. 17). Ontario: Canada IBM Press.
70. McCallum, M., Vasconcelos, W., et al. (2008). Organisational change through influence. Autonomous
Agents and Multi-Agent Systems, 17(2), 157–189.
71. Merali, Y., & Davies, J. (2001). Knowledge capture and utilization in virtual communities. In
Proceedings of the 1st international conference on knowledge capture (pp. 92–99). Victoria, BC,
Canada: ACM Press.
72. Miralaei, S., & Ghorbani, A. A. (2005). Category-based similarity algorithm for semantic similarity in
multi-agent information sharing systems. In International conference on intelligent agent technology.
IEEE/WIC/ACM.
73. Miyoshi, A., Yagawa, G., et al. (1999). An interface agent that actively supports CAE beginner
users in performing analyses. Advances in Engineering Software, 30(8), 575–579.
74. Moukas, A., & Maes, P. (1998). Amalthaea: An evolving multi-agent information filtering and
discovery system for the WWW. Autonomous Agents and Multi-Agent Systems, 1(1), 59–88.
75. Nakanishi, H., Nakazawa, S., et al. (2003). Can software agents influence human relations?: Balance
theory in agent-mediated communities. In Proceedings of the second international joint conference
on autonomous agents and multiagent systems (pp. 717–724). Melbourne, Australia: ACM Press.
76. Nakanishi, H. (2004). FreeWalk: A social interaction platform for group behavior in a virtual
space. International Journal of Human-Computer Studies, 60(4), 421–454.
77. Pant, G., & Menczer, F. (2002). MySpiders: Evolve your own intelligent web crawlers. Autonomous
Agents and Multi-Agent Systems, 5(2), 221–229.
78. Parunak, H. V. D. (2000). A practitioners’ review of industrial agent applications. Autonomous Agents
and Multi-Agent Systems, 3(4), 389–407.
79. Parunak, H. V. D., Sauter, J., et al. (1999). The RAPPID project: Symbiosis between industrial
requirements and MAS research. Autonomous Agents and Multi-Agent Systems, 2(2), 111–140.
80. Pechoucek, M., & Maík, V. (2008). Industrial deployment of multi-agent technologies: Review and
selected case studies. Autonomous Agents and Multi-Agent Systems, 17(3), 397–431.
81. Petit-Rozé, C., & Grislin-Le Strugeon, E. (2006). MAPIS, a multi-agent system for information
personalization. Information and Software Technology, 48(2), 107–120.
82. Pynadath, D. V., & Tambe, M. (2003). An automated teamwork infrastructure for heterogeneous
software agents and humans. Autonomous Agents and Multi-Agent Systems, 7(1), 71–100.
83. Raybourn, E. M., Kings, N., et al. (2003). Adding cultural signposts in adaptive community-based
virtual environments. Interacting with Computers, 15(1), 91–107.
84. Roberts, J. (2006). Limits to communities of practice. Journal of Management Studies, 43(3), 623–639.
85. Roda, C., Angehrn, A., et al. (2003). Using conversational agents to support the adoption of
knowledge sharing practices. Interacting with Computers, 15(1), 57–89.
123
Auton Agent Multi-Agent Syst
86. Sandholm, T. (2000). Agents in electronic commerce: Component technologies for automated
negotiation and coalition formation. Autonomous Agents and Multi-Agent Systems, 3(1), 73–96.
87. Sato, G. Y., & Barthès, J. P. (2010). CoPBoard: A catalyst for distributed communities of practice. Inter-
national Journal of Software Science and Computational Intelligence, 2(1), 52–71.
88. Shakshuki, E., Ghenniwa, H., et al. (2003). An architecture for cooperative information sys-
tems. Knowledge-Based Systems, 16(1), 17–27.
89. Shakshuki, E., Koo, H.-H., et al. (2008). A distributed multi-agent meeting scheduler. Journal of
Computer and System Sciences, 74(2), 279–296.
90. Shaw, N. G., Mian, A., et al. (2002). A comprehensive agent-based architecture for intelligent informa-
tion retrieval in a distributed heterogeneous environment. Decision Support Systems, 32(4), 401–415.
91. Shen, W., Norrie, D. H., et al. (2001). Multi-agent systems for concurrent intelligent design and
manufacturing. London: Taylor and Francis.
92. Sierhuis, M., Bradshaw, J. M., et al. (2003). Human-agent teamwork and adjustable autonomy in
practice. In Proceeding of the 7th international symposium on artificial intelligence. Robotics and
Automation in Space: i-SAIRAS 2003. Nara, Japan.
93. Sierra, C. (2004). Agent-mediated electronic commerce. Autonomous Agents and Multi-Agent Sys-
tems, 9(3),
285–301.
94. Soller, A., Guizzardi, R., et al. (2004). SCALE: Supporting community awareness, learning, and
evolvement in an organizational learning environment. In Proceedings of the 6th international
conference on learning sciences. Santa Monica, CA: International Society of the Learning Sciences.
95. Song, D., Lau, R. Y. K., et al. (2007). An intelligent information agent for document title classification
and filtering in document-intensive domains. Decision Support Systems, 44(1), 251–265.
96. Sorensen, C., Macklin, D., et al. (2001). Navigating the World Wide Web: Bookmark maintenance
architectures. Interacting with Computers, 13(3), 375–400.
97. Soulier, E. (2004). Les Communautés de Pratique au Coeur de l’Organisation Réelle des Entreprises.
Systèmes d’Information Et Management, 9(1), 3–23.
98. Staab, S., & Schnurr, H.-P. (2000). Smart task support through proactive access to organizational
memory. Knowledge-Based Systems, 13(5), 251–260.
99. Stathis, K., de Bruijn, O., et al. (2002). Living memory: Agent-based information management for
connected local communities. Interacting with Computers, 14(6), 663–688.
100. L. E., Storck, & Storck, S. J. (2004). Trusting the knowledge of large online communities : strategies
for leanding from behind. In P. Hildreth & C. Kimble (Eds.), Knowledge networks: Innovation
through communities of practice (Vol. 1, pp. 243–255). Hershey: Idea Group Publishing.
101. Sumi, Y., & Mase, K. (2000). Supporting awareness of shared interests and experiences in commu-
nity. SIGGROUP Bulletin, 21(3), 35–42.
102. Sumi, Y., & Mase, K. (2001). AgentSalon: Facilitating face-to-face knowledge exchange through con-
versations among personal agents. In Proceedings of the fifth international conference on autonomous
agents (pp. 393–400). Montreal, QC, Canada: ACM Press.
103. Sumi, Y., & Mase, K. (2002). Conference assistant system for supporting knowledge sharing in
academic communities. Interacting with Computers, 14(6), 713–737.
104. Sumi, Y., & Mase, K. (2002). Supporting the awareness of shared interests and experiences in
communities. International Journal of Human-Computer Studies, 56(1), 127–146.
105. Sycara, K. (1998). Multiagent systems. AI Magazine, 79–92.
106. Sycara, K., Paolucci, M., et al. (2003). The RETSINA MAS infrastructure. Autonomous Agents and
Multi-Agent Systems, 7(1), 29–48.
107. Syed Mustapha, S. M. F. D. (2005). Extracting social knowledge in the intelligent conversational chan-
nel for agent communication. In 19th international conference on advanced information networking
and applications. AINA 2005.
108. Tacla, C. A., & Barthès, J.-P. (2003). A multi-agent system for acquiring and sharing lessons
learned. Computers in Industry, 52(1), 5–16.
109. Tacla, C. A., Freddo, A. R., et al. (2011). Supporting small teams in cooperatively building application
domain models. Expert Systems with Applications, 38(2), 1160–1170.
110. Takahashi, T., & Katagiri, Y. (2003). TelMeA2003: Social summarization in online communities.
CHI ‘03 extended abstracts on Human factors in computing systems (pp. 928–929). Ft. Lauderdale,
FL, USA: ACM.
111. Takahashi, T., Bartneck, C., et al. (2005). TelMeA—expressive avatars in asynchronous communi-
cations. International Journal of Human-Computer Studies, 62(2), 193–209.
112. Tambe, M., Pynadath, D. V., et al. (2000). Adaptive agent architectures for heterogeneous team
members. In Proceedings of the international conference on multi-agent systems (ICMAS).
123
Auton Agent Multi-Agent Syst
113. Tashiro, N., Hattori, H., et al. (2004). Implementing a proxy agent based writable web for a dynamic
information sharing system. In Proceedings of the 13th international conference on World Wide
Web: Alternate track papers & posters (pp. 256–257). New York, NY, USA: ACM Press.
114. Tweedale, J., Ichalkaranje, N., et al. (2007). Innovations in multi-agent systems. Journal of Network
and Computer Applications, 30(3), 1089–1115.
115. Urlings, P., Sioutis, C., et al. (2006). A future framework for interfacing BDI agents in a real-time
teaming environment. Journal of Network and Computer Applications, 29(2–3), 105–123.
116. Van Aart, C.J., Wielinga, B., et al. (2004). Organizational building blocks for design of distributed
intelligent system. International Journal of Human-Computer Studies, 61(5), 567–599.
117. Wang, F. (2002). Self-organising communities formed by middle agents. In Proceedings of the first
international joint conference on autonomous agents and multiagent systems: Part 3 (pp. 1333–1339).
Bologna, Italy: ACM Press.
118. Wenger, E. (1998). Communities of practice: Learning, meaning and identity. Cambridge: Cambridge
University Press.
119. Wenger, E., McDermott, R., et al. (2002). Cultivating communities of practice: A guide to managing
knowledge. Boston: Harvard Business School Press.
120. Wenger, E., White, N., et al. (2005). Outiller sa communauté de pratique. In L. Langelier (Ed.)
Travailler, Apprendre et Collaborer en Réseau: guide de mise en place et d’animation de communautés
de pratique intentionelles (pp. 47–66). Quebec: CEFRIO.
121. Wooldridge, M. (2002). An introduction to multiagent system. Chichester: Wiley.
122. Wooldridge, M., & Jennings, N. R. (1995). Intelligent agents: Theory and practice. Knowledge
Engineering Review, 10(2), 115–152.
123. Yang, S.-Y. (2009). Developing of an ontological interface agent with template-based linguistic
processing technique for FAQ services. Expert Systems with Applications, 36(2, Part 2), 4049–4060.
124. Yang, B., Liu, J., et al. (2010). An autonomy-oriented computing approach to community mining in
distributed and dynamic networks. Autonomous Agents and Multi-Agent Systems, 20(2), 123–157.
125. Yang, F., Wang, M., et al. (2007). Community-organizing agent: An artificial intelligent system for
building learning communities among large numbers of learners. Computers & Education, 49(2), 131–
147.
126. Ye, Y., Boies, S., et al. (2001). Smart distance and WWW aware: A multi-agent approach. In
Proceedings of the fifth international conference on autonomous agents (pp. 176–177). Montreal,
QC, Canada: ACM Press.
127. Yoshida, S., Kamei, K., et al. (2003). Shine: A peer-to-peer based framework of network community
support systems. Computer Communications, 26(11), 1199–1209.
128. Yoshida, S., Ohguro, T., et al. (2000). A platform for making network community support systems in
a cooperative distributed architecture. In Seventh international conference on parallel and distributed
systems 2000.
129. Yu, B., & Singh M. P. (2002). An agent-based approach to knowledge management. In Proceedings
of the eleventh international conference on information and knowledge management (pp. 642–644).
McLean, VA, USA: ACM.
130. Yu, B., & Singh, M. P. (2003). Searching social networks. Proceedings of the second international
joint conference on autonomous agents and multiagent systems (pp. 65–72). Melbourne, Australia:
ACM.
131. Zambonelli, F., & Omicini, A. (2004). Challenges and research directions in agent-oriented software
engineering. Autonomous Agents and Multi-Agent Systems, 9(3), 253–283.
132. Zambonelli, F., Jennings, N. R., et al. (2003). Developing multiagent systems: The Gaia methodol-
ogy. ACM Transactions Software Engineering Methodologies, 12(3), 317–370.
133. Zoethout, K., Jager, W., et al. (2008). Task dynamics in self-organising task groups: Expertise,
motivational, and performance differences of specialists and generalists. Autonomous Agents and
Multi-Agent Systems, 16(1), 75–94.
123
