Content uploaded by Matthias Baldauf
Author content
All content in this area was uploaded by Matthias Baldauf
Content may be subject to copyright.
Int. J. Ad Hoc and Ubiquitous Computing, Vol. 2, No. 4, 2007 263
Copyright © 2007 Inderscience Enterprises Ltd.
A survey on context-aware systems
Matthias Baldauf
V-Research, Industrial Research and Development,
Stadtstrasse 33, 6850 Dornbirn, Austria
E-mail: matthias.baldauf@v-research.at
Schahram Dustdar* and Florian Rosenberg
Distributed Systems Group, Information Systems Institute,
Vienna University of Technology, Argentinierstrasse 8/184-1, 1040 Vienna, Austria
E-mail: dustdar@infosys.tuwien.ac.at E-mail: rosenberg@infosys.tuwien.ac.at
*Corresponding author
Abstract: Context-aware systems offer entirely new opportunities for application developers and
for end users by gathering context data and adapting systems behaviour accordingly. Especially
in combination with mobile devices these mechanisms are of high value and are used to
increase usability tremendously. In this paper, we present common architecture principles of
context-aware systems and derive a layered conceptual design framework to explain the different
elements common to most context-aware architectures. Based on these design principles, we
introduce various existing context-aware systems focusing on context-aware middleware and
frameworks, which ease the development of context-aware applications. We discuss various
approaches and analyse important aspects in context-aware computing on the basis of the
presented systems.
Keywords: context-awareness; context framework; context middleware; sensors; context model;
context ontology; context-aware services.
Reference to this paper should be made as follows: Baldauf, M., Dustdar, S. and Rosenberg, F.
(2007) ‘A survey on context-aware systems’, Int. J. Ad Hoc and Ubiquitous Computing, Vol. 2,
No. 4, pp.263–277.
Biographical notes: Matthias Baldauf is project manager at V-Research, an Austrian
competence center for industrial research and development. In the Department of Technical
Logistics he develops location-aware systems based on GPS, GSM and RFID technology with a
focus on track and trace solutions. His research interests include modern localisation methods and
efficient, flexible localisation architectures.
Schahram Dustdar is a Full Professor of Computer Science with a focus on Internet Technologies
at the Distributed Systems Group, Information Systems Institute, Vienna University of
Technology (TU Wien). In 1999 he co-founded Caramba Labs Software AG (CarambaLabs.com)
in Vienna, a venture capital co-funded software company focused on software for collaborative
processes in teams. Caramba Labs was nominated for several (international and national) awards.
He has published some 100 scientific papers as conference-, journal-, and book contributions.
He has written three academic books, one professional book, and co-edited six
books/proceedings. More information can be found at: http://www.infosys.tuwien.ac.at/Staff/sd.
Florian Rosenberg is research assistant and PhD student at the Distributed Systems Group,
Information Systems Institute, Vienna University of Technology. His research areas include
context-aware and autonomic services, service-oriented architectures and web service
technologies. More information can be found at: http://www.infosys.tuwien.ac.at/Staff/rosenberg.
1 Introduction
With the appearance and penetration of mobile devices such
as notebooks, PDAs, and smart phones, pervasive
(or ubiquitous) systems are becoming increasingly popular
these days. The term ‘pervasive’ introduced first by Weiser
(1991) refers to the seamless integration of devices into
the users everyday life. Appliances should vanish into the
background to make the user and his tasks the central focus
rather than computing devices and technical issues.
One field in the wide range of pervasive computing are
the so-called context-aware (or sentient) systems.
Context-aware systems are able to adapt their operations to
the current context without explicit user intervention and
thus aim at increasing usability and effectiveness by taking
environmental context into account. Particularly when it
264 M. Baldauf, S. Dustdar and F. Rosenberg
comes to using mobile devices, it is desirable that programs
and services react specifically to their current location, time
and other environment attributes and adapt their behaviour
according to the changing circumstances as context data
may change rapidly. The needed context information may
be retrieved in a variety of ways, such as applying sensors,
network information, device status, browsing user profiles
and using other sources. The history of context-aware
systems started when Want et al. (1992) introduced
their Active Badge Location System which is considered to
be one of the first context-aware applications. The infrared
technology based system is able to determine a user’s
current location which was used to forward phone calls to a
telephone close to the user. In the middle of the 1990s,
a couple of location-aware tour guides (Abowd et al., 1997;
Sumi et al., 1998; Cheverst et al., 2000) emerged which
provided information according to the user’s current
location. While location information is by far the most
frequently used attribute of context, attempts to use other
context information as well have grown over the last few
years as the examples in this paper will show. Hence,
it is a challenging task to define the word ‘context’
and many researchers tried to find their own definition for
what context actually includes. In literature the term
context-aware appeared in Schilit and Theimer (1994) the
first time. There the authors describe context as location,
identities of nearby people, objects and changes to those
objects. Such enumerations of context examples were often
used in the beginning of context-aware systems research.
Ryan et al. (1997) referred to context as the user’s location,
environment, identity and time. Dey (1998) defines context
as the user’s emotional state, focus of attention, location and
orientation, date and time, as well as objects and people in
the user’s environment. Another common way of defining
context was the use of synonyms. Hull et al. (1997) describe
context as the aspects of the current situation. These kind of
definitions are often too wide. However, a good one can be
found in Brown (1996). Brown defines context to be the
elements of the user’s environment which the computer
knows about. One of the most accurate definitions is
given by Dey and Abowd (2000b). These authors refer to
context as
“any information that can be used
to characterize the situation of entities
(i.e., whether a person, place or object)
that are considered relevant to the interaction
between a user and an application, including
the user and the application themselves.”
One popular way to classify context instances is the
distinction of different context dimensions. Prekop and
Burnett (2003) and Gustavsen (2002) call these dimensions
external and internal, and Hofer et al. (2002) refer to
xtitphysical and logical context. The external (physical)
dimension refers to context that can be measured by
hardware sensors, i.e., location, light, sound, movement,
touch, temperature or air pressure, whereas the internal
(logical) dimension is mostly specified by the user or
captured by monitoring user interactions, i.e., the user’s
goals, tasks, work context, business processes, the user’s
emotional state. Most context-aware systems make use of
external context factors as they provide useful data, such as
location information. Furthermore, external attributes are
easy to sense by using off-the-shelf sensing technologies.
Virtually all systems presented in this paper apply physical
context information. Examples for the use of logical
data are the Watson Project (Budzik and Hammond, 2000)
and the IntelliZap Project (Finkelstein et al., 2001) which
support the user by providing relevant information due to
information read out of opened web pages, documents, etc.
When dealing with context, three entities can be
distinguished (Dey and Abowd, 2001): places (rooms,
buildings etc.), people (individuals, groups) and things
(physical objects, computer components etc.). Each of these
entities may be described by various attributes which can be
classified into four categories: identity (each entity has a
unique identifier), location (an entity’s position, co-location,
proximity etc.), status (or activity, meaning the intrinsic
properties of an entity, e.g., temperature and lightning for a
room, processes running currently on a device etc.) and time
(used for timestamps to accurately define situation, ordering
events etc.).
This paper is structured as follows. Section 2 introduces
current design principles for context-aware systems and
common context models used in various context-aware
systems. In Section 3, we present a comparison of existent
context-aware systems and explain approaches, varieties
and similarities. In Section 4, we discuss the presented
approaches, and highlight advantages and disadvantages.
Finally, Section 5 draws some concluding remarks and
presents some future work in this area.
2 Design principles
In this section, we describe basic design principles and
introduce a conceptually layered framework, to associate the
functionality implemented in existent frameworks to
various layers. Furthermore, we depict different context
models used for representing, storing and exchanging
contextual information.
2.1 Architecture
Context-aware systems can be implemented in many ways.
The approach depends on special requirements and
conditions such as the location of sensors (local or remote),
the amount of possible users (one user or many), the
available resources of the used devices (high-end-PCs or
small mobile devices) or the facility of a further extension
of the system. Furthermore, the method of context-data
acquisition is very important when designing context-aware
systems because it predefines the architectural style of the
system at least to some extent. Chen (2004) presents three
different approaches on how to acquire contextual
information.
A survey on context-aware systems 265
• Direct sensor access. This approach is often used in
devices with sensors locally built in. The client
software gathers the desired information directly from
these sensors, i.e., there is no additional layer for
gaining and processing sensor data. Drivers for the
sensors are hardwired into the application, so this
tightly coupled method is usable only in rare cases.
Therefore, it is not suited for distributed systems due to
its direct access nature which lacks a component
capable of managing multiple concurrent sensor
accesses.
• Middleware infrastructure. Modern software design
uses methods of encapsulation to separate e.g., business
logic and graphical user interfaces. The middleware
based approach introduces a layered architecture to
context-aware systems with the intention of hiding
low-level sensing details. Compared to direct sensor
access this technique eases extensibility since the client
code has not be modified anymore and it simplifies the
reusability of hardware dependent sensing code due to
the strict encapsulation.
• Context server. The next logical step is to permit
multiple clients access to remote data sources. This
distributed approach extends the middleware based
architecture by introducing an access managing remote
component. Gathering sensor data is moved to this
so-called context server to facilitate concurrent multiple
access. Besides the reuse of sensors, the usage of a
context server has the advantage of relieving clients of
resource intensive operations. As probably the majority
of end devices used in context-aware systems are
mobile gadgets with limitations in computation power,
disk space etc., this is an important aspect. In return one
has to consider about appropriate protocols, network
performance, quality of service parameters etc., when
designing a context-aware system based on
client-server architecture.
In a similar manner, Winograd (2001) describes three
different context management models for coordinating
multiple processes and components:
• Widgets. Derived from the homonymous GUI elements
a widget is a software component that provides a public
interface for a hardware sensor (Dey and Abowd,
2000a, 2001). Widgets hide low-level details of
sensing and ease application development due to their
reusability. Because of the encapsulation in widgets it is
possible to exchange widgets which provide the same
kind of context data (e.g., exchange a radio frequency
widget by a camera widget to collect location data).
Widgets are usually controlled by some kind of a
widget manager. The tightly coupled widget approach
increases efficiency but is not robust to component
failures.
• Networked services. This more flexible approach,
argued for example in Hong and Landay (2001),
resembles the context server architecture. Instead of a
global widget manager discovery techniques are used to
find networked services. This service based approach is
not as efficient as a widget architecture due to complex
network based components but provides robustness.
• Blackboard model. In contrast to the process-centric
view of the widget and the service-oriented model, the
blackboard model represents a data-centric view. In this
asymmetric approach processes post messages to a
shared media, the so-called blackboard, and subscribe
to it to be notified when some specified event occurs.
Advantages of this model are the simplicity of adding
new context sources and the easy configuration. One
drawback is the need of a centralised server to host the
blackboard and the lack in communication efficiency as
two hops per communication are needed.
In this paper we will focus on middleware based and
context-server based systems with regards to their usability
in distributed systems. Many layered context-aware systems
and frameworks have evolved during the last years. Most of
them differ in functional range, location and naming of
layers, the use of optional agents or other architectural
concerns. Besides these adaptations and modifications, a
common architecture in modern context-aware applications
is identifiable when analysing the various design
approaches.
As mentioned above, a separation of detecting and
using context is necessary to improve extensibility and
reusability of systems. The following layered conceptual
architecture, as depicted in Figure 1, augments layers for
detecting and using context by adding interpreting and
reasoning functionality (Ailisto et al., 2002; Dey and
Abowd, 2001).
Figure 1 Layered conceptual framework for context-aware
systems
The first layer consists of a collection of different sensors.
It is notable that the word ‘sensor’ not only refers to
sensing hardware but also to every data source which may
provide usable context information. Concerning the way
data is captured; sensors can be classified in three groups
(Indulska and Sutton, 2003).
266 M. Baldauf, S. Dustdar and F. Rosenberg
• Physical sensors. The most frequently used type of
sensors are physical sensors. Many hardware sensors
are available nowadays which are capable of capturing
almost any physical data. Table 1 shows some
examples of physical sensors (Schmidt and van
Laerhoven, 2001).
• Virtual sensors. Virtual sensors source context data
from software applications or services. For example, it
is possible to determine an employee’s location not
only by using tracking systems (physical sensors) but
also by a virtual sensor, e.g., by browsing an electronic
calendar, a travel-booking system, emails etc., for
location information. Other context attributes that can
be sensed by virtual sensors include, e.g., the user’s
activity by checking for mouse-movement and
keyboard input.
• Logical sensors. These sensors make use of a couple of
information sources, and combine physical and virtual
sensors with additional information from databases or
various other sources in order to solve higher tasks.
For example, a logical sensor can be constructed to
detect an employee’s current position by analysing
logins at desktop PCs and a database mapping of
devices to location information.
Table 1 Commonly used physical sensor types
Type of
context Available sensors
Light Photodiodes, colour sensors, IR and UV-
sensors etc.
Visual
context
Various cameras
Audio Microphones
Motion,
acceleration
Mercury switches, angular sensors,
accelerometers, motion detectors, magnetic
fields
Location Outdoor: Global Positioning System (GPS),
Global System for Mobile Communications
(GSM); Indoor: Active Badge system, etc.
Touch Touch sensors implemented in mobile devices
Temperature Thermometers
Physical
attributes
Biosensors to measure skin resistance, blood
pressure
The second layer is responsible for the retrieval of raw
context data. It makes use of appropriate drivers for
physical sensors and APIs for virtual and logical sensors.
The query functionality is often implemented in reusable
software components which make low-level details of
hardware access transparent by providing more abstract
methods such as getPosition(). By using interfaces for
components responsible for equal types of context these
components become exchangeable. Therefore, it is possible,
for instance, to replace a RFID system by a GPS system
without any major modification in the current and upper
layers.
The Preprocessing layer is not implemented in every
context-aware system but may offer useful information if
the raw data are too coarse grained. The preprocessing layer
is responsible for reasoning and interpreting contextual
information. The sensors queried in the underlying layer
most often return technical data that are not appropriate
to use by application designers. Hence this layer raises
the results of layer two to a higher abstraction level.
The transformations include extraction and quantisation
operations. For example, the exact GPS position of a person
might not be of value for an application but the name of the
room the person is in, may be.
In context-aware systems consisting of several different
context data sources, the single context atoms can be
combined to high-level information in this layer.
This process is also called ‘aggregation’ or ‘composition’.
A single sensor value is often not important to an
application, whereas combined information might be more
precious and accurate. In this vein, a system is able to
determine, e.g., whether a client is situated indoor or
outdoor by analysing various physical data like temperature
and light or whether a person is currently attending a
meeting by capturing noise level and location. To make this
analysis work correctly a multitude of statistical methods
are involved and often some kind of training phase is
required.
Obviously, this abstraction functionality could also be
implemented directly by the application. But due to a couple
of reasons this task should better be encapsulated and
moved to the context server. The encapsulation advances
the reusability and, hence, eases the development of client
applications. And by making such aggregators remotely
accessible the network performance increases (as clients
have to send only one request to gain high-level data instead
of connecting to various sensors) and limited client
resources are saved.
The problem of sensing conflicts that might occur when
using several data sources has to be solved in this layer as
well. For example, when a system is notified about a
person’s location by the coordinates of her mobile phone
and by a camera spotting this person, it might be difficult to
decide what information to use. Often this conflict is
approached by using additional data like time stamps and
resolution information.
The fourth layer, Storage and Management, organises
the gathered data and offers them via a public interface to
the client. Clients may gain access in two different ways,
synchronous and asynchronous. In the synchronous manner
the client is polling the server for changes via remote
method calls. Therefore, it sends a message requesting some
kind of offered data and pauses until it receives the server’s
answer. The asynchronous mode works via subscriptions.
Each client subscribes to specific events it is interested in.
On occurrence of one of these events, the client is either
simply notified or a client’s method is directly involved
using a call back. In the majority of cases the asynchronous
approach is more suitable due to rapid changes in the
underlying context. The polling technique is more resource
A survey on context-aware systems 267
intensive as context data has to be requested quite often and
the application has to prove for changes itself, using some
kind of context history.
The client is realised in the fifth layer, the
Application layer. The actual reaction on different events
and context-instances is implemented here. Sometimes
information retrieval and application specific context
management and reasoning is encapsulated in form of
agents, which communicate with the context server and act
as an additional layer between the preprocessing and the
application layer (Chen, 2004). An example for context
logic at the client side is the display on mobile devices: as a
light sensor detects bad illumination, text may be displayed
in higher colour contrast.
All the systems, we analysed in this paper implement
most of the layers of the conceptual framework presented
above.
2.2 Context models
A context model is needed to define and store context data
in a machine processable form. To develop flexible and
useable context ontologies that cover the wide range
of possible contexts is a challenging task. Strang and
Linnhoff-Popien (2004) summarised the most relevant
context modelling approaches, which are based on the data
structures used for representing and exchanging contextual
information in the respective system.
• Key-Value models. These models represent the simplest
data structure for context modelling. They are
frequently used in various service frameworks, where
the key-value pairs are used to describe the capabilities
of a service. Service discovery is then applied by using
matching algorithms which use these key-value pairs.
• Markup scheme models. All markup based models use a
hierarchical data structure consisting of markup tags
with attributes and content. Profiles represent typical
markup-scheme models. Typical examples for such
profiles are the Composite Capabilities/Preference
Profile (CC/PP) (W3C, 2004a) and User Agent Profile
(UAProf) (Wapforum, 2001), which are encoded in
RDF/S. Various other examples can be found in Strang
and Linnhoff-Popien (2004).
• Graphical models. The Unified Modelling Language
(UML) is also suitable for modelling context. Various
approaches exist where contextual aspects are modelled
in by using UML, e.g., Sheng and Benatallah (2005).
Another modelling approach includes an extension to
the Object-Role Modelling (ORM) by context
information presented in Hendricksen et al. (2003).
• Object oriented models. Modelling context by using
object-oriented techniques offers to use the full power
of object orientation (e.g., encapsulation, reusability,
inheritance). Existing approaches use various objects to
represent different context types (such as temperature,
location, etc.), and encapsulate the details of context
processing and representation. Access the context and
the context processing logic is provided by well-defined
interfaces. Hydrogen (Hofer et al., 2002) uses such an
object-oriented example. We explain the system in
more detail in Section 3.
• Logic based models. Logic-based models have a high
degree of formality. Typically, facts, expressions and
rules are used to define a context model. A logic based
system is then used to manage the aforementioned
terms and allows to add, update or remove new facts.
The inference (also called reasoning) process can be
used to derive new facts based on existing rules in the
systems. The contextual information needs to be
represented in a formal way as facts. One of the first
approaches was published by McCarthy and Buvac
(1997).
• Ontology based models. Ontologies represent a
description of the concepts and relationships.
Therefore, ontologies are a very promising instrument
for modelling contextual information due to their high
and formal expressiveness and the possibilities for
applying ontology reasoning techniques. Various
context-aware frameworks use ontologies as underlying
context models. We describe some of them in
Section 3.
The conclusion of the evaluation presented in Strang and
Linnhoff-Popien (2004), based on six requirements, show
that ontologies are the most expressive models and fulfil
most of their requirements. Korpipää et al. (2003) present
some requirements and goals having designed a context
ontology:
• simplicity: the used expressions and relations should be
as simple as possible to simplify the work of
applications developers
• flexibility and extensibility: the ontology should support
the simple addition of new context elements and
relations
• genericity: the ontology should not be limited to special
kind of context atoms but rather support different types
of context
• expressiveness: the ontology should allow to describe
as much context states as possible in arbitrary detail.
268 M. Baldauf, S. Dustdar and F. Rosenberg
Numerous tools are available to define declarative
representations and to publish and share ontologies
developed by the World Wide Web Consortium, e.g., the
Resource Description Language (RDF) (W3C, 2000) and
the Web Ontology Language (OWL) (W3C, 2004b).
A single context atom can be described with a couple of
attributes. The two most obvious are:
• Context type. The context type refers to the category of
context such temperature, time, speed, etc. This type
information may be used as a parameter for a context
query or a subscription, e.g., subscribeToChanges
(‘temperature’). It is important to use
meaningful type names, hence, as the system grows,
some names might not be unique anymore. For example
the type position may belong to a mobile device or a
user. One solution for creating well-structured type
names is the use of cascaded names (Korpipää and
Mäntyjärvi, 2003) as shown in Table 2.
• Context value. Context value means the raw data
gathered by a sensor. The unit depends on the context
type and the applied sensor, e.g., degree Celsius, miles
per hour, etc.
In most cases, context type and context value are not
enough information to build a working context-aware
system. Additional attributes that might be useful include:
• Time stamp. This attribute contains a date/time-value
describing when the context was sensed. It is needed
e.g., to create a context history and deal with sensing
conflicts.
• Source. A field containing information how the
information was gathered. In case of a hardware sensor
it might hold the ID of the sensor and allow an
application to prefer data from this sensor.
• Confidence. The confidence attribute describes the
uncertainty of this context type. Not every data source
delivers accurate information, e.g., location data suffers
inaccuracy depending on the used tracking tool.
Table 2 Example context vocabulary
Context type Context
Environment:Temperature Cold
Environment:Temperature Normal
Environment:Temperature Hot
Environment:Light:Source 50 Hz
Environment:Light:Source 60 Hz
Environment:Light:Source NotAvailable
Device:Activity:Placement AtHand
Device:Activity:Placement NotAtHand
Part of a flexible context model is an extendable context
vocabulary to deal with abstract descriptions rather than
technical data. It simplifies the description of various
context atoms and context instances. Table 2 shows a
small part of an example vocabulary from Korpipää and
Mäntyjärvi (2003). Notice that not all contexts have to be
available at a time. In contrast to temperature, a light source
is not always measurable.
3 Existent systems and frameworks
In this section, we discuss three types of existing
context-aware systems. Firstly, we briefly describe a special
case of context-aware systems, the so-called location-aware
systems. Secondly, we describe an existing context-aware
system assisting hospital information management. Thirdly,
we draw our main focus on context-aware frameworks, as
basic building block for context-aware systems and
applications.
3.1 Location-aware systems
Context-aware systems dealing with location information
are widespread and the demand for them is growing due to
the increasing spread of mobile devices. Examples for
location-aware systems are various tourist guide projects
where information dependent to the current location is
displayed. Other examples can be found in Espinoza et al.
(2001), Priyantha et al. (2000), Burrell and Gay (2002) and
Kerer et al. (2004). A couple of different location aware
infrastructures are available in order to collect position data.
These include GPS satellites, mobile phone towers, badge
proximity detectors, cameras, magnetic card readers,
barcode readers, etc. These sensors can provide either
position or proximity information. In addition, they differ in
price and accuracy. Some need a clear line of sight, other
signals can travel through walls, etc. As a detailed example,
we introduce an indoor location sensing system from Harter
et al. (2002), who describes a location-aware system using
ultrasonic technique. To each entity (person or equipment),
which should be detectable, a small sending unit called bat
is attached. These bats have globally unique identifiers and
contain ultrasonic transducers. To monitor the signals sent
by the bats, receivers are installed at the rooms ceilings and
connected by a wired network. The third hardware type
needed is a base station. It periodically sends radio
messages with specific bat ids and resets the receivers.
A corresponding bat reacts by emitting an ultrasonic
impulse which is caught by the receivers. By recording the
time of arrival of signals the distance between the bat and
the receiver can be calculated. The bat’s exact position is
then determined by using multilateration (an extension of
trilateration). A challenge the authors where confronted
with due to the use of ultrasonic technique was the
incorrect measurement because of unwanted reflections of
the signals. The problem was solved by using a statistical
outlier rejection algorithm to improve the accuracy of the
calculated positions.
A survey on context-aware systems 269
3.2 Context-aware systems
The systems named in the prior chapter use only one aspect
of context, namely location information. The use of
different types of context atoms such as noise, light and
location allows the combination to high-level context
objects. These elements are necessary to build more
adaptive, useful and user-friendly systems. As an example
for this kind of context-aware infrastructures serves the
system presented by Munõz et al. (2003), which extends the
instant messaging paradigm by adding context-awareness to
support information management within a hospital setting.
All users (in this case physicians, nurses etc.) are equipped
with mobile devices to write messages that are sent when a
specified set of circumstances is satisfied. For example, a
user can formulate a message that should be delivered to the
first doctor that enters room number 108 after 8 am. The
contextual elements this system is aware of include location,
time, roles and device state. Its context functionality is
moved to agents which include three modules (layers).
The perception module gathers raw context information
from data sources (sensors, users, other agents, the server).
The reasoning module governs the agents’ actions and
finally the action module triggers a user-specified event.
All messages between agents are XML encoded.
3.3 Context-aware frameworks
Context-aware systems capable of dealing with special
types of context are well-suited for specific conditions,
e.g., in hospital scenarios. These systems can be optimised
for the situations they are used in. They do not have to be
flexible and extensible. To actually simplify the
development of context-aware applications an abstract
framework is needed. Such a generic infrastructure not
only provides a client with access to retrieve context data,
it also permits the simple registration of new distributed
heterogeneous data sources. In this section different
context-aware frameworks are introduced and compared
based on various design criteria (architecture, resource
discovery, sensing, context model, context processing
historical context data, security and privacy).
3.3.1 Architectures
The most common design approach for distributed
context-aware frameworks is a classical hierarchical
infrastructure with one or many centralised components
using a layered architecture as presented in Section 2.
This approach is useful to overcome memory and processor
constraints of small mobile devices but provides one
single point of failure and thereby lacks robustness. The
architecture of the Context Managing Framework
presented by Korpipää et al. (2003) is depicted in
Figure 2. Four main functional entities comprise this context
framework: the context manager, the resource
servers, the context recognition services
and the application.
Figure 2 Context managing framework architecture
Whereas the resource servers and the context recognition
services are distributed components, the so-called context
manager represents a centralised server managing a
blackboard. It stores context data and provides this
information to the client applications.
The Service-Oriented Context-Aware Middleware
(SOCAM) project introduced by Gu et al. (2004a, 2004b) is
another architecture for the building and the rapid
prototyping of context-aware mobile services. It uses a
central server as well, here called context interpreter, which
gains context data through distributed context providers and
offers it in mostly processed form to the clients. The
context-aware mobile services are located on top of the
architecture, thus, they make use of the different levels of
context and adapt their behaviour according to the current
context.
One further extensible centralised middleware approach
designed for context-aware mobile applications is a project
called Context-Awareness Sub-Structure (CASS) presented
in Fahy and Clarke (2004). In Figure 3, the system
architecture of CASS is presented. The middleware
contains an Interpreter, a ContextRetriever,
aRule Engine and a SensorListener. The
SensorListener listens for updates from sensors,
which are located on distributed computers called
sensor nodes. Then the gathered information is stored
in the database by the SensorListener. The
ContextRetriever is responsible for retrieving stored
context data. Both of these classes may use the services of
an interpreter. The ChangeListener is a component
with communication capabilities that allows a mobile
computer to listen for notification of context change events.
Sensor and LocationFinder classes also have built-in
communications capabilities. Mobile clients connect to the
server over wireless networks. To reduce the impact of
intermittent connections local caching is supported on the
client side.
270 M. Baldauf, S. Dustdar and F. Rosenberg
Figure 3 Architecture of the CASS system
Context Broker Architecture (CoBrA) (Chen et al., 2003) is
an agent based architecture for supporting context-aware
computing in so-called intelligent spaces. Intelligent spaces
are physical spaces (e.g., living rooms, vehicles, corporate
offices and meeting rooms) that are populated with
intelligent systems that provide pervasive computing
services to users. Central to CoBrA is the presence of an
intelligent context broker that maintains and manages a
shared contextual model on the behalf of a community of
agents. These agents can be applications hosted by mobile
devices that a user carries or wears (e.g., cell phones, PDAs
and headphones), services that are provided by devices in a
room (e.g., projector service, light controller and room
temperature controller) and web services that provide a web
presence for people, places and things in the physical world
(e.g., services keeping track of peoples and object
whereabouts). The context broker consists of four functional
main components: the Context Knowledge Base, the
Context Inference Engine, the Context Acquisition Module
and the Privacy Management Module. To avoid the bottle
neck problem CoBrA offers the possibility of creating
broker federations.
The Context Toolkit (Salber et al., 1999; Dey and
Abowd, 2000a; Dey, 2001), another context-aware
framework, takes a step towards a peer-to-peer architecture
but it still needs a centralised discoverer where distributed
sensor units (called widgets), interpreters and aggregators
are registered in order to be found by client applications.
The toolkits object-oriented API provides a superclass
called BaseObject which offers generic communication
abilities to ease the creation of own components.
Another framework based on a layered architecture is
built in the Hydrogen project (Hofer et al., 2002). Its context
acquisition approach is specialised for mobile devices.
While the availability of a centralised component is essential
in the majority of existent distributed content-aware systems,
the Hydrogen system tries to avoid this dependency.
It distinguishes between a remote and a local context.
The remote context is information another device knows
about, the local context is knowledge our own device is
aware of. When the devices are in physical proximity they
are able to exchange these contexts in a peer-to-peer manner
via WLAN, Bluetooth, etc. This exchange of context
information among client devices is called context sharing.
Figure 4 shows the management of a device’s context
which consists of its own local context and a set of remote
contexts gathered from other devices. Both local and remote
context are made up of context objects. The superclass
ContextObject is extended by different context types,
e.g., LocationContext,DeviceContext, etc. This
approach allows the simple addition of new context types by
specialising ContextObject. A context type has to
implement the methods toXML() and fromXML() from
the ContextObject class in order to convert the data
from and to a XML stream.
Figure 4 Hydrogen’s object-oriented approach
The architecture consists of three layers which are all
located on the same device (Figure 5). The Adaptor layer is
responsible for retrieving raw context data by querying
sensors. This layer permits a sensor’s concurrent use by
different applications. The second layer, the Management
layer, makes use of the Adaptor layer to gain sensor data
and is responsible for providing and retrieving contexts.
The so-called Context server offers the stored information
via synchronous and asynchronous methods to the client
applications. On top of the architecture is the Application
layer, where the appliance code is implemented to react on
specific context changes reported by the context manager.
Due to platform and language independency, all inter-layer
communication is based on a XML-protocol.
Figure 5 Architecture of the hydrogen project
A survey on context-aware systems 271
The CORTEX system is an example for a context-aware
middleware approach. The architecture is based on the
Sentient Object Model (Biegel and Cahill, 2004) which was
designed for the development of context-aware applications
in an ad-hoc mobile environment. The model’s special
suitability for mobile applications depends on the use of
STEAM, a location-aware event-based middleware service
designed specifically for ad-hoc wireless networking
environments.
A sentient object is an encapsulated entity consisting of
three main parts, as depicted in Figure 6, Sensory capture,
Context hierarchy and Inference engine. Via interfaces a
sentient object communicates with sensors which produce
software events and actuators which consume software
events. As Figure 6 shows, sentient objects can be both
producer and consumer of another sentient object. Own
sensors and actuators are programmed using STEAM.
For building sentient objects, a graphical development tool
is available which allows developers to specify relevant
sensors and actuators, define fusion networks, specify
context hierarchies and production rules, without the need to
write any code.
The Gaia project (Roman et al., 2002; Gaia Project,
2005), another middleware infrastructure, extends typical
operating system concepts to include context-awareness.
It aims at supporting the development and execution of
portable applications for active spaces. Gaia exports
services to query and utilise existing resources,
to access and use current context, and it provides a
framework to develop user-centric, resource-aware,
multi-device, context-sensitive and mobile applications.
The current system consists of the Gaia kernel and the
application framework as depicted in Figure 7.
Figure 6 The sentient object model
Figure 7 Architecture of the Gaia system
In this paper, we focus on Gaia’s parts concerning
context-awareness, namely the Event Manager, the Context
Service and the Context File System. The Event Manager
service is responsible for event distribution in the active
space and implements a decoupled communication model
based on suppliers, consumers, and channels. Each channel
has one or more suppliers that provide information
to the channel and one or more consumers that receive the
information. The reliability is increased as suppliers are
exchangeable. With the help of the Context Service,
applications may query and register for particular context
information and higher level context objects. Finally the
Context File System makes personal storage automatically
available in the users present location. It constructs a
virtual directory hierarchy to represent context as
directories, where path components represent context types
and values. For example, to determine which files have the
context of location == RM2401 && situation
== meeting associated with them, one may enter
the /location:/RM2401/situation:/meeting
directory.
3.3.2 Resource discovery
As sensors in a distributed network may fail or new ones
may be added, a discovery mechanism to search for and find
appropriate sensors at runtime is important. For these
purposes the Context Toolkit offers the already mentioned
discoverer. The discoverer works as registry component
which interpreters, aggregators and widgets have to notify
about their presence and their contact possibilities. After
registration the components are pinged to ensure that they
are operating. If a component does not respond to a
specified number of consecutive pings, the discoverer
determines that the component is unavailable and removes it
from its registry list. Customers may find appropriate
components querying the discoverer either via a white page
lookup (a search for the components name) or a yellow page
lookup (a search for specific attributes). In case the lookup
was successful the discoverer returns a handle to contact the
context component.
SOCAM offers a discovery mechanism as well called
Service Locating Service. In Gaia different context
providers are stored in a registry component. A pure
peer-to-peer context-aware system such as Hydrogen only
uses local built-in sensors and does not connect to distributed
sensors, therefore, no discovery mechanism is involved.
3.3.3 Sensing
The Context Toolkit authors presented a new approach to
handle different data sources. Derived from the use of
widgets in GUI development, they introduced so-called
context widgets to separate applications from context
acquisition concerns. In these widgets the complexity of
sensing is hidden. Furthermore, they abstract the gained
272 M. Baldauf, S. Dustdar and F. Rosenberg
context information (e.g., the accurate position of a
person might not be of value but the application should be
notified when this person enters another room) and as
widgets are encapsulated software components, they are
reusable. Each widget owns some attributes that can be
queried by applications, e.g., the IdentityPresence
widget, implemented by the authors, offers attributes
such as its location, the last time a presence was
detected and the identity of the last user detected.
Beside the polling mechanism an asynchronous way
of data retrieval is possible too: if an application
subscribes to a widget, it is notified when the widgets
context changes. The IdentityPresence provides the
callbacks PersonArrives(location,identity,
timestamp) and PersonLeaves (location,
identity, timestamp) which are triggered when a
person either arrives or leaves a room. The separation of
acquisition and use of context permits a simple exchange of
widgets since e.g., identity may be sensed in various ways
like Active Badges, video recognition, etc. This manner of
building reusable sensor units that make the action of
sensing transparent to the customer (whether it is a
centralised server or a distributed client component),
became widely accepted in distributed context-aware
systems: CASS applies sensor nodes,SOCAM uses context
providers, the Context Managing Framework refers to
resource servers and CoBrA makes use of context
acquisition components.
3.3.4 Context model
An efficient model for handling, sharing and storing
context data is essential for a working context-aware
system. The Context Toolkit handles context in simple
attribute-value-tuples, which are encoded using XML for
transmission. As already described above Hydrogen uses an
object-oriented context model approach with a superclass
called ContextObject which offers abstract methods to
convert data streams from XML representations to context
objects and vice versa. More advanced ways of dealing with
context data based on ontologies are found in SOCAM,
CoBrA and the Context Managing Framework. The SOCAM
authors divide a pervasive computing domain into several
sub-domains, e.g., home domain, office domain etc., and
define individual low-level ontologies in each sub-domain
to reduce the complexity of context processing. Each of
these ontologies implemented in OWL provides a special
vocabulary used for representing and sharing context
knowledge. CoBrA also uses an own OWL-based ontology
approach, namely COBRA-Ont (Chen et al., 2003, 2004).
Listing 1 shows parts of an COBRA-Ont example.
The ontology structure and vocabulary applied in the
Context Managing Toolkit are described in RDF. Parts of its
vocabulary are used as an example in Table 2 in Section 2.
Listing 1 CORBA-ONT example
In Gaia context is represented in a special manner,
namely through 4-ary predicates in the way Context
(<ContextType>,<Subject>, <Relater>,
<Object>) written in DAML + OIL. The <Context
Type> refers to the type of context the predicate is
describing, the <Subject> is the person, place or thing
with which the context is concerned, and the <Object> is
a value associated with the <Subject>. The <Relater>
relates the <Subject> and the <Object> such as a
comparison operator (=, >, or <), a verb, or preposition.
An example for a context instance is Context(temperature,
room 3231, is, 98 F). This syntax is used for both,
representing context and for forming inference rules.
3.3.5 Context processing
As soon as the raw context data is sensed by a data source, it
has to be processed as its customers mostly are rather
interested in already interpreted and aggregated information
than in raw, fine-grained data. Whereas context aggregation
refers to the composition of context atoms either to collect
all context data concerning a specific entity or to build
higher-level context objects, context interpretation refers to
the transformation of context data including special
knowledge. These forms of context data abstraction
ease the application designer’s work tremendously.
The Context Toolkit offers facilities for both context
aggregation and context interpretation. The context
aggregators (former called context servers) are responsible
for composing context of particular entities by subscribing
to relevant widgets, context interpreters provide the
possibility of transforming context, e.g., in a simple case
returning the corresponding e-mail address to a passed
name. Like widgets aggregators and interpreters inherit
communication methods from the upperclass BaseObject
and have to be registered at the discoverer in order to be
found. The Context Managing Framework presented by
Korpip et al. (Figure 2) offers various processing facilities
as well. The resource servers’ tasks are complex. First they
gather raw context information by connecting to various
data sources. After the preprocessing and feature abstraction
crip limits and fuzzy sets are used for quantisation. But now
A survey on context-aware systems 273
the data are delivered by posting it to the context manager’s
blackboard. The context recognition services are used
by the context manager to create higher-level context object
out of context atoms. In this vein new recognition services
are easy to add.
In SOCAM, the Context Reasoning Engine reasons over
the knowledge base, its tasks include inferring deduced
contexts, resolving context conflicts and maintaining the
consistency of the context knowledge base. Different
inference rules used by the reasoning engine can be
specified. The interpreter is implemented by using Jena2
(Jena, 2005), a semantic web toolkit.
In CoBrA the so-called Inference Engine processes
context data. The engine contains the Context Reasoning
Module responsible for aggregating context information.
It reasons over the Context Knowledge Base and deduces
additional knowledge from information acquired from
external sources.
In CASS deriving of high-level context is also based on
an inference engine and a knowledge base. The knowledge
base contains rules queried by the inference engine to find
goals using the so-called forward chaining technique.
As these rules are stored in a database separated from the
interpreter neither recompiling nor restarting of components
is necessary when rules change. Table 3 shows an example
rule database entry containing criteria to display rather
indoor than outdoor activities in a CASS based tour-guide
application.
Table 3 Example rule database entry
Rain Brightness Temperature Goal
Wet Dull Cold Indoor
In CORTEX, the whole context processing is encapsulated
in Sentient Objects. The sensory capture unit performs
sensor fusion to manage uncertainty of sensor data
(sensing conflicts) and to build higher-level context objects.
Different contexts are represented in a so-called context
hierarchy together with specific actions to be undertaken in
each context. Since only one context is active at any point
in time (concept of the active context) the number of rules
that have to be evaluated are limited. Thus efficiency of the
inference process is increased. The inference engine
component is based on C Language Integrated Production
System (CLIPS). It is responsible for changing application
behaviour according to the current context by using
conditional rules. Gaia’s context processing is hidden in the
Context Service Module allowing the creation of high-level
context objects by performing first order logic operations
such as quantification, implication, conjunction, disjunction,
and negation of context predicates. One example of a rule is
Context(Number of people, Room 2401, >, 4)
AND Context(Application, Powerpoint, is,
Running) Context(Social Activity, Room
2401, Is, Presentation). Almost all current
context-aware frameworks permit the aggregation and
interpretation of raw context data. Only in a few cases the
higher-level abstractions are handled by the application
layer, such as in Hydrogen.
3.3.6 Historical context data
Sometimes it might be necessary to have access to historical
context data. Such context histories may be used to establish
trends and predict future context values. As most data
sources constantly provide context data, the maintenance of
a context history is mainly a memory concern, so a
centralised high-resource storage component is needed.
Since in a server-based architecture the context data
provided by sensors has anyway to be stored at the
server-side to offer it to customers, the majority of these
systems has the facility to query historical context data.
The Context Toolkit,CoBrA,CASS,SOCAM and CORTEX
save sensed context data persistently in a database. A further
advantage of using a database is the use of the Structured
Query Language (SQL) which enables to read and to
manipulate operations at a high abstraction level. In the
CoBrA and the CASS architecture the persistent storage is
called Context Knowledge Base. Additionally a set of APIs
is offered to assert, delete, modify and query the stored
knowledge. CASS uses its database not only to save context
data but also to store domain knowledge and inference rules
needed for creating high-level context. Due to limited memory
resources a peer-to-peer network of mobile devices like
Hydrogen is not able to offer persistent storage possibilities.
3.3.7 Security and privacy
As context may include sensitive information on people,
e.g., their location and their activity, it is necessary to have
the opportunity to protect privacy. For these purposes the
Context Toolkit introduces the concept of context
ownership. Users are assigned to sensed context data as
their respective owners. They are allowed to control the
other users’ access. New components involved in this access
control are the Mediated Widgets, Owner Permissions,
a modified BaseObject and Authenticators.
The MediatedWidget class is an extension of a basic
widget which contains a so-called widget developer
specifying who owns the data being sensed. The Owner
Permission is the component that receives permission
queries and determines to grant or to deny access based on
stored situations. These situations include authorised
users, time of access etc. The modified BaseObject contains
all the original methods augmented with identification
mechanisms. Now applications and components have to
provide their identity along with the usual request for
information. Finally the authenticator is responsible for
proofing the identity by using a public-key infrastructure.
CoBrA includes an own flexible policy language to control
context access, called Rei (Kagal et al., 2003). This policy
language is modelled on deontic concepts of rights,
prohibitions, obligations and dispensations and controls data
access through dynamically modifiable domain dependent
policy rules.
274 M. Baldauf, S. Dustdar and F. Rosenberg
4 Discussion of approaches
In Table 4 we have summarised the main aspects
of the discussed approaches. The architectural style of a
context-aware system is mainly driven by the context
acquisition method. The main criteria for a reasonable
architectural approach is the separation of concerns between
the context acquisition and the user components as proposed
by Dey (2000). All the frameworks presented in this paper
support this separation of concerns.
The sensing technology is implemented differently in
every framework. It is important, that the concrete
sensing mechanism is encapsulated in separate components,
to support the aforementioned separation of concerns.
Furthermore, it encapsulates sensing and allows to access
the contextual data via defined interfaces. Currently, there is
no standard description language or ontology for sensing
contextual information from various sources to enable reuse
across various middleware systems and frameworks.
Therefore, proprietary solutions as used by the different
frameworks, have emerged. SOCAM use the most
sophisticated approach for sensing context information.
External virtual sensors are consumed via web services
(by using SOAP). Internal providers for querying sensors
are consumed by using context events represented in OWL
based on a predefined ontology.
The context model and the context processing logic
supported by different frameworks is a major criteria for
providing intelligent and adaptable context-aware services
or applications. As mentioned before, ontologies provide a
rich formalism for specifying contextual information. Based
on such ontological models, highly sophisticated ontology
reasoning engines can derive new concepts to adapt the
service behaviour accordingly. The major drawback of the
Context Toolkit is, therefore, its context model, a set of
attribute-value tuples. The development of intelligent context
processing and aggregation is limited due to the fact that
such attributes do not have a meaning. Furthermore, using
non-ontology based models requires a lot of programming
effort and tightly couples the context model to the rest of
the system. Moreover, the lack of declarative semantics of
programs does not allow reasoning and knowledge sharing
amount systems. For example, SOCAM uses a general upper
ontology to specify basic common contextual properties and
to refine this general ontology. Domain-specific ontologies
can be defined to provide very fine grained possibilities for
specifying and formalising context.
Table 4 Summary of discussed approaches
Architecture Sensing
Context
model
Context
processing Resource discovery
Historical
context data
Security
and privacy
CASS Centralised
middleware
Sensor nodes Relational
data model
Inference engine
and knowledge base
n.a. Available n.a.
CoBra Agent based Context
acquisition
module
Ontologies
(OWL)
Inference engine
and knowledge base
n.a. Available Rei policy
language
Context
Management
Framework
Blackboard
based
Resource
servers
Ontologies
(RDF)
Context recognition
service
Resource servers
+ subscription
mechanism
n.a. n.a.
Context
Toolkit
Widget based Context
widgets
Attribute-value
tuples
Context
interpretation and
aggregation
Discoverer
component
Available Context
ownership
CORTEX Sentient object
model
Context
component
framework
Relational
data model
Service discovery
framework
Resource
management
component
framework
Available n.a.
Gaia MVC
(extended)
Context
providers
4-ary predicates
(DAML + OIL)
Context-service
module (first-order
logic)
Discovery service Available Supported
(e.g., secure
tracking, location
privacy, access
control)
Hydrogen Three layered
architecture
Adapters
for various
context types
Object-oriented Interpretation and
aggregation of raw
data only
n.a. n.a. n.a.
SOCAM Distributed with
centralised
server
Context
providers
Ontologies
(OWL)
Context reasoning
engine
Service
locating service
Available n.a.
A survey on context-aware systems 275
Resource discovery mechanisms are currently rarely used in
the presented frameworks. Such dynamic mechanisms are
important, especially in a pervasive environment, where
available sensors, the context sources, change rapidly
(new ones are added or removed). SOCAM, for example,
which is the only system that is based on a service-oriented
architecture, offers a service locating service, which
dynamically binds to available context providers.
These providers can also be changed at runtime. The lack of
resource discovery support is a major drawback of many
frameworks, because it implies that the used context sources
are stable and permanently available, which is not always
the case in real-world applications. Erroneous behaviour of
one or more context sources may lead to a decreased
availability of the context-aware service or application.
Managing historical context data provides the ability
to implement intelligent learning algorithms which allow
to provide highly adaptable context-aware services.
Furthermore, based on learning algorithms, contextual
information can be predicted to proactively provide a certain
set of services to the user. Many of the systems store
contextual information but none of them do not use learning
techniques to provide context-aware service proactively.
Another important aspect is security and privacy.
Contextual information mostly considers user profile
information and other sensitive information. Therefore,
concepts are needed to express policies and to define
ownership of context information. CoBra uses the Rei
policy language, to express (security) policies about
contextual information. Gaia uses several mechanisms to
define privacy restrictions and secure communication
for tracking locations of people. The Context Toolkit
implements the concept of context ownership, which is used
to allow access to the context to the owner only. The other
frameworks do not use security concepts, which is one of
their major drawback. When dealing with sensitive data,
secure connections for context acquisition as well as privacy
of different user specific contextual information is very
important.
5 Conclusions and future work
In this survey paper we described different design principles
and context models for context-aware systems and
presented various existent middleware and server-based
approaches to ease the development of context-aware
applications. The direct comparison of the named systems
and frameworks shows their similarity concerning the
layered structure. Especially remarkable is the strict division
of the context data acquisition and use. Thus context sources
become reusable and are able to serve a multitude of context
clients. Although most authors refer to abstract context
sources, the currently mainly used and tested sources are
physical sensors. Virtual and logical sensors are capable of
providing useful context data as well and should be more
incorporated in ongoing research.
Other often disregarded aspects are security and
privacy issues. These facets belong to the most important
components of a context-aware system as the protection of
sensitive context data must be guaranteed. Many systems
totally lack security modules, others provide basic security
mechanisms and only a few systems offer advanced and
sufficient security options. Probably the main problem in
the presented approaches is the variety of used context
encodings and ways to find and access context sources.
Every system and framework uses its own format to
describe context and its own communications mechanisms.
We believe that standardised formats and protocols are
important for the enhancements of context-aware systems to
make the development of context services the focus rather
than the communication between context sources and users.
In our opinion web service technologies seem to be an
appropriate solution to achieve that aim as they provide
standardised methods for service description and access.
Our future work in this area will investigate the use of
service-oriented and autonomic computing concepts for
building context-aware service frameworks. We believe
that standardised technologies and protocols, such as
WSDL and SOAP, could help to build more interoperable
context aware services. Furthermore, the use of ontologies
for building a context model is an important approach
(as can be seen from existing approaches) to build
more sophisticated algorithms, which derive new contextual
knowledge or patterns to proactively aggregate new
context-aware services. Autonomously orchestrating atomic
context-aware services into higher level services based
on context information and available QoS parameters
provides high potential for offering more accurate services
to the user.
References
Abowd, G.D., Atkeson, C.G., Hong, J., Long, S., Kooper, R.
and Pinkerton, M. (1997) ‘Cyberguide: a mobile
context-aware tour guide’, Wirless Networks, Vol. 3, No. 5,
pp.421–433.
Ailisto, H., Alahuhta, P., Haataja, V., Kylloenen, V. and
Lindholm, M. (2002) ‘Structuring context aware applications:
Five-layer model and example case’, Proceedings of the
Workshop on Concepts and Models for Ubiquitous
Computing, Goteborg, Sweden, available online: http://
www.comp.lancs.ac.uk/computing/users/dixa/conf/ubicomp2
002-models/pdf/Ailisto-Ubicomp%20Workshop8.pdf.
Biegel, G. and Cahill, V. (2004) ‘A framework for developing
mobile, context-aware applications’, Proceedings of the 2nd
IEEE Conference on Pervasive Computing and
Communication, pp.361–365.
Brown, P.J. (1996) ‘The stick-e document: a framework for
creating context-aware applications’, Proceedings of the
Electronic Publishing, Palo Alto, pp.259–272.
Budzik, J. and Hammond, K.J. (2000) ‘User interactions with
everyday applications as context for just-in-time information
access’, Proceedings of the 5th international Conference on
Intelligent User Interfaces, New Orleans, Louisiana, USA,
pp.44–51.
276 M. Baldauf, S. Dustdar and F. Rosenberg
Burrell, J. and Gay, G. (2002) ‘E-graffiti: evaluating real-world
use of a context-aware system’, Interacting with
Computers – Special Issue on Universal Usability, Vol. 14,
No. 4, pp.301–312.
Chen, H. (2004) An Intelligent Broker Architecture for Pervasive
Context-Aware Systems, PhD Thesis, University of Maryland,
Baltimore County.
Chen, H., Finin, T. and Joshi, A. (2003) ‘An ontology for
context-aware pervasive computing environments’, The
Knowledge Engineering Review, Cambridge University Press,
Vol. 18, pp.197–207.
Chen, H., Finin, T. and Joshi, A. (2004) ‘An ontology for
context-aware pervasive computing environments’, Special
Issue on Ontologies for Distributed Systems, Knowledge
Engineering Review, Vol. 18, No. 3, pp.197–207.
Cheverst, K., Davies, N., Mitchell, K., Friday, A. and
Efs-tratiou, C. (2000) ‘Developing a context-aware electronic
tourist guide: some issues and experiences’, Proceedings of
the SIGCHI conference on Human Factors in Computing
Systems, ACM Press, New York, USA, pp.17–24.
Dey, A.K. (1998) ‘Context-aware computing: the CyberDesk
project’, Proceedings of the AAAI, Spring Symposium on
Intelligent Environments, Menlo Park, CA, pp.51–54.
Dey, A.K. (2000) Providing Architectural Support for Building
Context-Aware Applications, PhD Thesis, Georgia Institute of
Technology, Georgia Institute of Technolgy, USA.
Dey, A.K. (2001) The Context Toolkit – A Toolkit for
Context-aware Applications, http://www.cs.berkeley.edu/
~dey/context.html.
Dey, A.K. and Abowd, G.D. (2000a) ‘The context toolkit: aiding
the development of context-aware applications’, Workshop on
Software Engineering for Wearable and Pervasive
Computing, Limerick, Ireland.
Dey, A.K. and Abowd, G.D. (2000b) ‘Towards a better
understanding of context and context-awareness’, Proceedings
of the Workshop on the What, Who, Where, When and How of
Context-Awareness, ACM Press, New York.
Dey, A.K. and Abowd, G.D. (2001) ‘A conceptual framework and
a toolkit for supporting rapid prototyping of context-aware
applications’, Human-Computer Interactions (HCI) Journal,
Vol. 16, Nos. 2–4, pp.7–166.
Espinoza, F., Persson, P., Sandin, A., Nystrom, H., Cac-ciatore, E.
and Bylund, M. (2001) ‘GeoNotes: social and navigational
aspects of location-based information systems’, Proceedings
of the 3rd International Conference on Ubiquitous
Computing, Atlanta, Georgia, USA, pp.2–17.
Fahy, P. and Clarke, S. (2004) ‘CASS – a middleware for
mobile context-aware applications’, Workshop on Context
Awareness, MobiSys 2004.
Finkelstein, L., Gabriolovic, E., Matias, Y., Rivilin, E., Solan, Z.,
Wolfman, G. and Ruppin, E. (2001) ‘Placing search in context:
the concept revisited’, Proceedings of the 10th International
World Wide Web Conference (WWW 10), Hong Kong.
Gaia Project (2005) http://gaia.cs.uiuc.edu/.
Gu, T., Pung, H.K. and Zhang, D.Q. (2004a) ‘A middleware
for building context-aware mobile services’, Proceedings
of IEEE Vehicular Technology Conference (VTC), Milan,
Italy.
Gu, T., Pung, H.K. and Zhang, D.Q. (2004b) ‘A middleware for
building context-aware mobile services’, Proceedings of
IEEE Vehicular Technology Conference (VTC 2004), Milan,
Italy.
Gustavsen, R.M. (2002) ‘Condor – an application framework for
mobility-based context-aware applications’, Proceedings of
the Workshop on Concepts and Models for Ubiquitous
Computing, Goeteborg, Sweden.
Harter, A., Hopper, A., Steggles, P., Ward, A. and Webster, P.
(2002) ‘The anatomy of a context-aware application’, Wirless
Networks, Vol. 8, Nos. 2–3, pp.187–197.
Hendricksen, K., Indulska, J. and Rakotonirainy, A. (2003)
‘Generating context management infrastructure from
high-level context models’, Proceedings of the 4th
International Conference on Mobile Data Management
(MDM'03), pp.1–6.
Hofer, T., Schwinger, W., Pichler, M., Leonhartsberger, G.
and Altmann, J. (2002) ‘Context-awareness on mobile
devices – the hydrogen approach’, Proceedings of the 36th
Annual Hawaii International Conference on System Sciences,
pp.292–302.
Hong, J.I. and Landay, J.A. (2001) ‘An infrastructure for
context-aware computing’, Human-Computer Interaction,
Vol. 16, pp.287–303.
Hull, R., Neaves, P. and Bedford-Roberts, J. (1997) ‘Towards
situated computing’, Proceedings of the First International
Symposium on Wearable Computers (ISWC ‘97), p.146.
Indulska, J. and Sutton, P. (2003) ‘Location management in
pervasive systems’, CRPITS’03: Proceedings of the
Australasian Information Security Workshop, pp.143–151.
Jena (2005) A Semantic Web Framework for Java, http://jena.
sourceforge.net/.
Kagal, L., Finin, T. and Joshi, A. (2003) ‘A policy language for a
pervasive computing environment’, Proceedings of the 4th
IEEE International Workshop on Policies for Distributed
Systems and Networks (POLICY), pp.63–74.
Kerer, C., Dustdar, S., Jazayeri, M., Gomes, D., Szego, A. and
Caja, J.A.B. (2004) ‘Presence-aware infrastructure using web
services and RFID technologies’, Proceedings of the 2nd
European Workshop on Object Orientation and Web Services,
Oslo, Norway.
Korpipää, P. and Mäntyjärvi, J. (2003) ‘An ontology for mobile
device sensor-based context awareness’, Proceedings of
CONTEXT, 2003, Vol. 2680 of Lecture Notes in Computer
Science, pp.451–458.
Korpipää, P., Mantyjarvi, J., Kela, J., Keranen, H. and Malm, E-J.
(2003) ‘Managing context information in mobile devices’,
IEEE Pervasive Computing, Vol. 2, No. 3, July–September,
pp.42–51.
McCarthy, J. and Buvac, S. (1997) ‘Formalizing context
(expanded notes)’, in Buvac, S. and Iwahska, L. (Eds.):
Working Papers of the AAAI Fall Symposium on Context in
Knowledge Representation and Natural Language, California.
American Association for Artificial Intelligence, Menlo Park,
pp.99–135.
Munõz, M.A., Gonzalez, V.M., Rodriguez, M. and
Fa vela, J. (2003) ‘Supporting context-aware collaboration
in a hospital: an ethnographic informed design’, Proceedings
of Workshop on Artificial Intelligence, Information
Access, and Mobile Computing 9th International
Workshop on Groupware, CRIWG 2003, Grenoble, France,
pp.330–334.
Prekop, P. and Burnett, M. (2003) ‘Activities, context and
ubiquitous computing’, Special Issue on Ubiquitous
Computing Computer Communications, Vol. 26, No. 11,
pp.1168–1176.
A survey on context-aware systems 277
Priyantha, N.B., Chakraborty, A. and Balakrishnan, H. (2000)
‘The cricket location-support system’, Proceedings of the 6th
Annual International Conference on Mobile Computing and
Networking, ACM Press, pp.32–43.
Roman, M., Hess, C., Cerqueira, R., Ranganathan, A.,
Campbell, R.H. and Nahrstedt, K. (2002) ‘A middleware
infrastructure for active spaces’, IEEE Pervasive Computing,
Vol. 1, No. 4, pp.74–83.
Ryan, N., Pascoe, J. and Morse, D. (1997) ‘Enhanced reality
fieldwork: the context-aware archaeological assistent’,
Proceeding of the 25th Anniversary Computer Applications in
Archaeology, http://www.caaconference.org/.
Salber, D., Dey, A.K. and Abowd, G.D. (1999) ‘The context
toolkit: aiding the development of context-aware
applications’, Proceedings of the ACM CHI, Pittsburgh, PA,
pp.434–441.
Schilit, B. and Theimer, M. (1994) ‘Disseminating active map
information to mobile hosts’, IEEE Network, Vol. 8, No. 5,
pp.22–32.
Schmidt, A. and van Laerhoven, K. (2001) ‘How to build smart
applications?’, IEEE Personal Communications, Vol. 8,
No. 4, pp.66–71.
Sheng, Q.Z. and Benatallah, B. (2005) ‘ContextUML:
a UML-based modeling language for model-driven
development of context-aware web services’, Proceedings of
the International Conference on Mobile Business (ICMB’05),
pp.206–212.
Strang, T. and Linnhoff-Popien, C. (2004) A Context Modeling
Survey, First International Workshop on Advanced Context
Modelling, Reasoning and Management, UbiComp.
Sumi, Y., Etani, T., Fels, S., Simonet, N., Kobayashi, K. and Mase,
K. (1998) ‘C-map: Building a context-aware mobile assistant
for exhibition tours’, Community Computing and Support
Systems, Social Interaction in Networked Communities
[the book is based on the Kyoto Meeting on Social Interaction
and Communityware, held in Kyoto, Japan, in June 1998],
Springer-Verlag, London, UK, pp.137–154.
W3C (2000) Resource Description Framework (RDF),
http://www.w3.org/RDF.
W3C (2004a) Composite Capability/ Preference Profiles (CC/PP),
http://www.w3.org/TR/2004/ REC-CCPP-struct-vocab-
20040115/.
W3C (2004b) OWL Web Ontology Language Overview, W3C
Recommendation 10 February, http://www.w3.org/TR/
owl-features/.
Want, R., Hopper, A., Falcao, V. and Gibbons, J. (1992)
‘The active badge location system’, ACM Transactions on
Information Systems, Vol. 10, No. 1, pp.91–102.
Wapforum (2001) User Agent Profile (UAProf) Specification,
http://www.wapforum.org.
Weiser, M. (1991) ‘The computer for the 21st century’, Scientific
American, pp.94–104.
Winograd, T. (2001) ‘Architectures for context’, Human-Computer
Interaction (HCI) Journal, Vol. 16, No. 2, pp.401–419.