Crossing the Rubicon for An Intelligent Advisor

Conference Paper (PDF Available) · January 2005with 62 Reads
Conference: Beyond Personalization 2005, IUI’05
Abstract
Recommender systems (RS) are being used by an increasing number of e-commerce sites to help consumers find prod-ucts to purchase. We define here the features that may char-acterize an "intelligent" RS, based on behavioral science, data mining, and computational intelligence concepts. We present our conclusions from building the WiseUncle Inc. RS, named Rubicon, and give its general description. Rather than being an advisor for a particular application, Rubicon is a generic RS, a platform for generating application specific advisors.
Figures - uploaded by Razvan Andonie
Author content
All content in this area was uploaded by Razvan Andonie
Crossing the Rubicon for An Intelligent Advisor
R˘
azvan Andonie
Computer Science Department
Central Washington University, Ellensburg, USA
andonie@cwu.edu
J. Edward Russo
Johnson Graduate School of Management
Cornell University, Ithaka, USA
jer9@cornell.edu
Rishi Dean
Sloan School of Management
Massachusetts Institute of Technology, USA
rdean@mit.edu
ABSTRACT
Recommender systems (RS) are being used by an increasing
number of e-commerce sites to help consumers find prod-
ucts to purchase. We define here the features that may char-
acterize an ”intelligent” RS, based on behavioral science,
data mining, and computational intelligence concepts. We
present our conclusions from building the WiseUncle Inc.
RS, named Rubicon, and give its general description. Rather
than being an advisor for a particular application, Rubicon is
a generic RS, a platform for generating application specific
advisors.
Keywords
Recommender systems, electronic commerce, user interface,
user modeling
INTRODUCTION
E-commerce sites use RS to guide potential consumers
through the buying process by providing customized infor-
mation and product recommendations. Based on the cus-
tomers’ individual needs, values, and preferences, the goal
of a RS is to find the ”best” possible product from a large
set of complex options. We shall only mention some on-
line recommender systems that have been used, or are being
considered for use: [4, 5, 3, 10]. There are several well-
known e-commerce businesses that use, or have used, RS
technology in their web sites: Amazon, Travelocity, BMW,
MovieFinder, and Dell among them.
Although commercial RS have been available for several
Copyright is held by the author/owner(s).
Workshop: Beyond Personalization 2005
IUI’05, January 9, 2005, San Diego, California, USA
http://www.cs.umn.edu/Research/GroupLens/beyond2005
years now, we are still at the beginning of using RS on a
large scale. In reality, sellers provide an RS to help improve
the (long-term) business relationship. This goal gives rise to
several desiderata that can be difficult to achieve. The RS
should be flexible, scalable, multifunctional, adaptive, and
able to solve complex search and decision problems.
The RS interface with the customer should be based on the
same consumer psychology knowledge and strategies used in
marketing. Behind this ”visible” task, a RS can bring valu-
able information to marketers, making them improve their
offer and products (customer profiling, marketing segmenta-
tion). For instance, RS can help businesses decide to whom
to send a customized offer or promotion.
RS use knowledge to guide consumers through the of-
ten overwhelming task of locating suitable products. This
knowledge may come from experts (e.g. marketing, product
domain) or it can be ”mined” knowledge learned from the
behavior of other consumers. These two types of knowledge
can be used not only during the recommendation process, but
also to adaptively improve the system itself.
When optimizing the recommendation, it is possible that the
system has to search in a huge admissible solution space for
the ”best” product. Solving such an optimization during the
course of an Internet interaction creates a difficult problem,
one that requires devising fast heuristic solutions.
Another knotty problem is related to the formal definition
of the ”best” recommendation. The optimality of the rec-
ommended option generally requires several criteria. The
question is how to quantify the importance of such different
attributes like color, shape, speed, price, etc?
These considerations and others make us define an intelligent
advisor (IA) as a RS having the following features:
a.) During each interaction with a customer, it extracts
knowledge from the customer that is used to build and update
the corresponding customer profile. When the interaction is
concluded, the system makes a valid recommendation.
b.) The IA saves the extracted knowledge and the customer
1
profile can be further ”mined” for marketing-relevant knowl-
edge.
c.) The IA - customer interface is based on the psychology of
the consumer and the purchase decision process. Therefore,
behavioral science techniques should create the fundamen-
tals of an IA, in particular a customer dialog that embraces
what/how people think, rather than forcing consumers to
feed an optimization algorithm. Thus the IA divorces users
from some of the complexity of their decisions.
d.) The IA should be able to improve its functionality by
continually learning from its interactions with consumers.
e.) An IA should be robust in the face of data that are uncer-
tain, noisy, sparse, or missing.
f.) An IA should be scalable and able to work in real-time,
to meet the requirements of an Internet application.
g.) An IA should know how to draw multiobjective compar-
isons among products.
h.) An IA should be largely domain-independent, such that
with minimum modification effort, one should be able to cus-
tomize the same platform for other applications (e.g., selling
computers, cars, financial services).
How far are current RS from an ideal IA? Some of the exist-
ing RS already incorporate some of these requirements. For
instance, TalkMine uses the behavior of its users to adapt
the knowledge stored in information sources [8]. However,
most probably, none of the commercial available RS fulfills
all requirements.
The idea of defining an IA came after several years of build-
ing a commercial RS, called Rubicon. Rather than being an
advisor for a particular application, Rubicon is a generic RS,
a platform for generating application-specific advisors. In
this paper we look at the difficulties we faced when building
Rubicon and the main concepts we used.
DIFFICULTIES IN BUILDING A RS
There are two categories of problems we faced when build-
ing our RS. The first is related to the system design, the sec-
ond to the customer behavior.
Design and Integration
Incorporating complex behavioral data. There are many
types of information that can be collected and used: cus-
tomer knowledge (goals, needs and profile of the user), do-
main knowledge (product information and business rules
specific to a particular vertical application), traffic logs, and
expert knowledge. Using expert knowledge alone we can
recommend ”good” products. Using only customer knowl-
edge, we can recommend products that were sold success-
fully in the past. The first strategy is much better at dealing
with new products, whereas the latter one reflects only the
customer experience.
Scalability and real-time performance. Scalability in rec-
ommender systems includes both very large problem sizes
and real-time latency requirements. For instance, a rec-
ommender system connected to a large web site must pro-
duce each recommendation within a few tens of millisec-
onds while serving thousands of consumers simultaneously
and searching through potentially billions of possible prod-
uct configurations.
Noisy, missing, uncertain, and sparse data. The value of a
RS lies in the fact that most customers have not deeply con-
sidered many of the available products, the product features,
or their personal needs. This means that we must often deal
with extremely sparse data, such as that resulting from a cus-
tomer responding ”I don’t know”, ”I don’t care”, or ”I don’t
want to answer this question”.
Connecting recommenders to marketers. RS should be
connected to the vendor’s product database and the mar-
keter’s reporting systems. Only products that are currently
in stock should be recommended, or products that can be
configured in a feasible manner, both from engineering and
logical perspectives. The highly volatile nature of real-world
products and information systems creates the necessity of ad-
equate database maintenance in the IA.
Domain independence. From the software engineering
point of view, building a domain-independent RS platform
can be done by separating the generic part from the domain
specific knowledge modules.
User Experience
The customer-recommender interface is usually based on a
series of interactive questions presented to the customer by
the RS, accompanied by multiple-choice options for the cus-
tomer to input their answers. In this case, a difficult prob-
lem is what strategy to follow when selecting questions to
present. An intelligent dialog should be personalized. Some
randomness should be used when selecting the questions
[7]. Too much randomness leads to excessive customer ef-
fort, but a small amount of randomness may help to extend
the space over which the recommender understands the cus-
tomer’s interests and ensures that all questions are occasion-
ally presented to customers. A reasonable strategy for se-
lecting information from customers is to minimize customer
effort while still being able to make accurate predictions
[7]. However, this strategy is quite simplistic, and a be-
havioral science-based investigation is necessary here. What
we should measure is not customer effort (measured in the
duration of the dialog), but customer satisfaction. Satisfac-
tion quantification results from longer-term statistics on us-
age and surveying customers.
During the conversation, an IA adopts a five-stage process,
described by [2]: i) Opening; ii) Utilitarian Needs; iii) He-
donic Preferences; iv) Optional Features / Add-ons; and v)
Endgame.
Stage iframes the buyer (e.g., knowledge of the product cat-
egory and extent of product search to date) and the main
product characteristics (e.g., a desktop PC versus a laptop).
2
Stages ii and iii encompass, respectively, the utilitarian and
hedonic or emotional needs. The former include the func-
tional uses of the product, such as an automobile’s seating
capacity or environmental friendliness. Stage iii’s hedonic
needs, like the image of a car’s body style and brand name,
are often harder for a buyer to express. Needless to say,
extracting such knowledge can be a substantial behavioral
challenge in itself. Stage iv captures the remaining, minor
product specifications, like an automobile’s audio speakers
or aspects of its interior. The final stage covers such external
elements as a PC’s warranty or the local availability of reli-
able repair service for an automobile. These five stages are
sufficient to structure the process of a purchase decision for
all complex products.
How can a trusted recommender validate itself to consumers
through a web client? The following factors contribute to the
success of such conversations in Internet-mediated dialogs
[9].
The benefits of the conversation should exceed its costs.
People use information only if it is perceived as adding ben-
efits or as reducing costs. If (expected) costs exceed (ex-
pected) benefits at any point, there is a clear risk of the cus-
tomer terminating the dialog.
Credibility and trust. The information and advice must be
credible, and the source must be trustworthy. An Internet-
delivered RS cannot provide the face-to-face cues of trust-
worthiness that a human can. However, although a RS may
have no initial reputation for trust (based on past experience),
such an image can be built over time by personal usage,
word-of-mouth recommendations, or public endorsements
(e.g., by consumer-oriented magazines’ endorsement of the
system’s knowledge and disinterestedness). One alternative
is to add a confidence metric, and this has the potential to im-
prove user satisfaction and alter user behavior in the RS [6].
A second alternative is to make the RS adaptive. This would
reduce the risk of manipulation: users can detect systems
that manipulate prediction and, this has a negative impact on
their trust [1].
Intelligence and customization. First, the advisor must
know what kinds of information people can validly provide
and how to successfully extract that information from buy-
ers. Consumers can usually say what they need or want the
product to do and can articulate such personal preferences as
style and color. However, they may have difficulty specify-
ing the product features that meet those needs. Second, based
on whatever can be learned from the customer, the problem
of identifying the optimal product must be solved. Thus, the
advisor must first extract the customer’s needs and then build
an inferential bridge from those needs to the most suitable
product.
Control. Customers should be able to request additional or
explanatory information. Or as the conversation proceeds,
the customer may learn something that requires returning to
an earlier point in the dialog and changing a preference stated
there. Buyers who feel impatient should be able to request a
recommendation at any time, even before the advisor would
normally feel comfortable providing one. Finally, the buyer
might even like to suspend the conversation and return later.
More control in any situation is empowering, and more so in
situations where control is expected. Providing satisfactory
conversational control is a special challenge to RS.
Feedback. Specific feedback might include (a) how much
progress has been made toward identifying the best prod-
uct, and (b) how much longer the conversation is expected to
take. Whatever specific feedback options are provided, how-
ever, users do not want to receive feedback only after they
have answered every question (as they must in many static
surveys).
SYSTEM DESCRIPTION
Rubicon is a generic domain-independent advisor, recom-
mending products from an existing set. Each product is con-
figurable, meaning it is comprised of several components,
which may each be described in turn by several attributes.
Building a RS depends largely on the knowledge represen-
tation model, and we chose a computational intelligence
framework. Our RS is a classifier that ”learns” to make good
recommendations. This classifier is an expert system, able to
explicitly expose its acquired knowledge. The main charac-
teristics of Rubicon are:
The inferential process from the customer’s needs to
the best product is constructed in two stages, called
Bridges, one from needs to product attributes, and the
second from attributes to the products themselves.
It can easily be customized for different applications
since the interface to the application-specific knowledge
domain is separated from the main system.
The front-end dialog is dynamic and changes based on
user responses. It elicits information from users to de-
termine a location in a needs space which is then used
to find optimal (sub-optimal) solutions in a products
space.
It accepts imprecise input from users.
It provides a justification for all recommendations.
Reversibility: The system can reverse the decision pro-
cess from effect to cause. This allows forecasting the
adoption of new products or services using real cus-
tomer decision data.
The Rubicon system diagram (Fig. 1) shows the following
main modules.
Conversation Engine (CE)
The CE is responsible for dialog management, presenting
questions to the user and processing the resulting responses
3
Figure 1: Rubicon High Level System Description
sent to it by the user. Questions and their associated re-
sponses are processed to accomplish the following two re-
sults: i) Propagate the knowledge gained from a response to
the subsequent inference mechanisms and ii) Determine the
next question to pose to the user.
In doing so, the dialog management occurs subject to the
following constraints:
Presents the appropriate questions for the system to
confidently determine an intelligent, personalized rec-
ommendation.
Presents questions conforming to users’ expectation of
a real dialog with respect to flow, organization, and co-
herence.
Minimizes the number of questions presented.
Scalable addition/ subtraction/ modification of ques-
tions.
Allows users and administrators to reproduce particular
dialogs.
Uses proper constructs for further data mining.
From a computational standpoint a rule-based expert system
is used to implement the CE’s dialog management process.
Questions and responses are linked by sets of predetermined
rules, and a number of other intermediary constructs. In
this way, the questions, responses, and rules can be speci-
fied, along with goals (i.e., knowledge to be gathered) in-
dependently of knowing the dialog flow in advance. The
system at runtime determines, based on the behavioral and
informational goals, which question to present next to the
user. When all appropriate questions have been presented,
the conversation is determined to be complete. However, the
user may intervene at any time to ask for the system’s cur-
rent best recommendation based on the information provided
thus far.
Inference Engine (IE)
The purpose of the IE is to map the user’s profile of needs
(the output of the CE) to the attributes necessary to comprise
the appropriate recommendation. Given a set of responses
resulting from the dialog, the IE can indicate a set of recom-
mendations, ordered by the degree of their preference. These
recommendations are not concrete (physical) product recom-
mendations yet, but a mapping from the user needs space to
the space of attributes, yielding generic descriptions of the
product, like ”RAM Amount” (e.g., standard, large, maxi-
mum) and ”Network Card type”. Collectively this inference
is called the First Bridge.
The IE is taught by a human expert. However, it can learn
incrementally as well: new teaching examples can be added
without restarting the teaching process from the beginning.
Conditional rules can be extracted to describe the behavior
of the IE and justify recommendations, market research and
performance improvement. The IE is stable under noisy in-
puts and user uncertainty. Such ”noise” may be produced by
”I don’t know” answers, or by contradictory answers in the
dialog.
To implement the IE, a fuzzy neural net architecture is used,
trained to represent the expert knowledge of a particular
product domain. For instance, in the case of a personal com-
puter RS, experts develop training patterns to represent the
varying needs profiles of customers along with their corre-
sponding feature sets for a recommended PC. The inference
process is fast, online.
Product Search Engine (PSE)
The PSE is the Second Bridge, a mapping from the space
of attributes to the space of (physical) products. It is an
optimization module interfacing with the retailer’s product
database to select the best, valid product configurations that
match the criteria specified by the user, such as the mini-
mum cost, the maximum likelihood of success, or a number
of other simultaneous criteria. The inputs to the PSE are the
levels of the attributes (the output of the IE), the configura-
tion constraints (i.e., incompatibilities among components),
and a user’s criteria for optimization (e.g., a desired price
point). These criteria for the algorithm can be set by the IE
and CE and are, therefore, uniquely tailored to a given user.
The PSE can sort though billions of options in real time, al-
lowing searches to be completed online. The products with
the highest degree of fit are passed to the Justification Engine
for further processing.
The response to a question is subsequently used to provide
4
more information that adds to Rubicon’s knowledge base
during a user-experience. This, in turn, leads to a recalcula-
tion and optimal selection of the next most appropriate ques-
tion. This question-response model continues until Rubicon
is either asked, or has sufficient confidence, to make a rec-
ommendation.
The PSE navigates a vast search space, taking into account
different optimization criteria. We used a genetic algorithm
approach for this (NP-complete) optimization problem. In
the initial phases, the PSE operates on abstractions of the
real world, and then through an adaptor layer translates these
abstractions into concrete items. This information is capable
of being read and processed at runtime. The PSE remains
independent of constant updating of the ”real world” items.
This adaptor level is implemented as an XML data bridge.
Justification Engine (JE)
Recommendations are run through the JE to provide a plain
English explanation of why the system has provided a spe-
cific recommendation. This justification is delivered in the
same vernacular as the dialog, personalized to the user, and
is present to facilitate user understanding and adoption of the
recommendation. The JE takes the set of If-Then rules from
the IE and the set of recommended configurations from the
PSE and develops a rationale for selecting each product. The
value of the JE is that it creates confidence in the recommen-
dation.
Rubicon is implemented using a complimentary modular
software approach that encapsulates the individual compu-
tational blocks, as well as the necessary software architec-
ture emphasizing a stable and reusable model that is com-
pliant with the J2EE technology standard. The user inter-
face is HTML 4.0 compliant, utilizing DHTML, and com-
bining client-side scripting and styles. It is implemented us-
ing XML/XSLT, built for 4th generation web browsers, and
rendered via the adapter layer using JSP/servlets.
PRELIMINARY TESTS
Although still under development, Rubicon was sufficiently
developed to be submitted to usability testing by two major
PC manufacturers. Each test involved about a dozen users
and compared three RS. One was the manufacturer’s current
online RS, one was an attractive competitor, while the third
was Rubicon. The results made available revealed that Rubi-
con was judged clearly superior in both tests. For instance,
in one test, when asked which of the three RS the user would
”be most likely to use again”, nine of eleven respondents
chose Rubicon.
Rubicon was tested online by a webhosting services
provider. Of 2200 online users who began a conversation,
83% completed it to the point of receiving a recommenda-
tion (which was the only result made available to us).
CONCLUSIONS
Is Rubicon ”intelligent”? According to our IA criteria from
the Introduction, yes. Furthermore, we believe that these
criteria may be a base for classifying RS. We have not pre-
sented here other modules of Rubicon, used for prediction,
customer profiling, and marketing segmentation, since we
have tried to focus on the core system. It was a challenging
task to build Rubicon, especially because of its generic char-
acter. Making the system largely independent of a specific
e-commerce application required greater complexity and ab-
straction. But do we really need a generic RS? From a user
perspective this may be a non-issue. However, for the RS
designer and software engineer this is a critical requirement.
We should think not only in terms of how to use a RS, but
also how to build it and how to adapt it fast for very different
application areas.
REFERENCES
1. D. Cosley, S. Lam, I. Albert, J. Konstan, and J. Riedl. Is
seeing believing? how recommender systems influence
users’ opinions. In Proceedings of CHI 2003 Conference
on Human Factors in Computing Systems, pages 585–
592, Fort Lauderdale, FL, 2003.
2. A. Horowitz and J. Russo. Modeling new car customer-
salesperson interaction for a knowledge-based system.
Advances in Consumer Research, 16:392–398, 1989.
3. S.-Y. Hwang, W.-C. Hsiung, and W.-S. Yang. A pro-
totype www literature recommendation system for dig-
ital libraries. Online Information Review, 27:169–182,
2003.
4. W. Lin, S. Alvarez, and C. Ruiz. Efficient adaptive-
support association rule mining for recommender sys-
tems. Data Mining and Knowledge Discovery, 6:83–
105, 2002.
5. J. Liu and J. You. Smart shopper: An agent-based web-
mining approach to Internet shopping. IEEE Transac-
tions on Fuzzy Systems, 11:226–237, 2003.
6. S. McNee, S. Lam, C. Guetzlaff, J. Konstan, and
J. Riedl. Confidence displays and training in recom-
mender systems. In Proceedings of INTERACT ’03 IFIP
TC13 International Conference on Human-Computer
Interaction, pages 176–183, 2003.
7. A. M. Rashid, I. Albert, D. Cosley, S. Lam, S. McNee,
J. Konstan, and J. Riedl. Getting to know you: Learning
new user preferences in recommender systems. In Pro-
ceedings of the 2002 International Conference on Intel-
ligent User Interfaces, pages 127–134, San Francisco,
2002.
8. L. M. Rocha. Talkmine: a soft computing approach
to adaptive knowledge recommendation. In V. Loia
and S. Sessa, editors, Soft Computing Agents: New
Trends for Designing Autonomous Systems - Studies in
5
Fuzziness and Soft Computing, pages 89–116. Physica-
Verlag, Springer, 2001.
9. J. E. Russo. Aiding purchase decisions on the Internet.
In Proceedings of the Winter 2002 SSGRR (Scuola Su-
periore Guglielmo Reiss Romoli) International Confer-
ence on Advances in Infrastructure for Electronic Busi-
ness, Education, Science, and Medicine on the Internet,
L’Aquila, Italy, 2002.
10. H.-W. Tung and V.-W. Soo. A personalized restau-
rant recommender agent for mobile e-service. In Pro-
ceedings of the 2004 IEEE International Conference
on e-Technology, e-Commerce and e-Service (EEE’04),
pages 259–262, 2004.
APPENDIX: THE USER EXPERIENCE WITH THE
RUBICON COMPUTER ADVISOR
Instantiated as a computer advisor, Rubicon shows (Fig. 2)
its first screen to a customer. Buyer control is offered, inter
alia, by the ”Why Ask” button, which pops up a justification
for the current question. This feedback also exhibits respect
for the buyer’s right to know why their time and effort are
being spent answering this particular question.
Figure 2: Getting acquainted
Ten questions into the conversation, just as it moves to Stage
2, the screen from Fig. 3 appears.
The Progress box on the right now shows a current recom-
mendation in the form of the best type of computer for the
particular buyer and the two most suitable alternative types.
For each of the three types, a measure of fit to the buyer’s
needs is displayed. This measure should increase as more
needs are elicited through further questioning and the recom-
mended product is further customized to those needs. Below
the recommendation is an announcement about the number
of facts known, either directly or inferred. And below that is
an indicator of how far through the conversation the system
expects the buyer is. The measure is not time, but the num-
ber of questions already asked and the expected number to
be asked before the conversation is finished.
After this screen the next question is ”Do you have any
problems with your current computer?” If the buyer answers
Figure 3: Extract the customer’s needs
”Yes” and identifies ”Too Slow” as the only problem, then
the screen shown in Fig. 4 appears.
Figure 4: Continue extracting the needs
Note that the fit of the Hobbyist PC to the buyer’s needs has
moved from 40% to 53% and the number of known facts
from 14 to 21. Both changes provide feedback and show the
benefit of participating in the dialog, as it progresses toward
the best recommendation. If the conversation reaches its nat-
ural conclusion (i.e., is not terminated prematurely by the
buyer), three products are recommended, in order. This is
another illustration of buyer control. People prefer to make
the final choice themselves from several options, although
they also want to know the advisor’s ranking. A demo ver-
sion of Rubicon can be found at www.wiseuncle.com.
6
  • ... Researches in this area focused on developing algorithms for efficiently producing recommendations to users. But effective explanation of these recommendations is also an important issue to increase the adoption and satisfaction level of users [9]. In this paper a constrained based mechanism is discussed which allow the user to input the constraint to discover the desired patterns. ...
    Article
    Full-text available
    Data mining process may uncover thousands of patterns from a given data set; most of them may be unrelated to the users' interest. Also these rules occupy more memory space, take more time also require more efforts of the decision maker in analysis. To confine the search space users have the good sense of which direction of mining may lead to related or interested patterns they would like to find. Therefore, a good heuristic is to have the users specify such intuition or expectation as constraints to limit the search space. In this paper efforts are made to discover valuable patterns using the user input constraints.
  • ... Sabe-se que a construção do conhecimento ocorre individualmente mas que a colaboração é capaz de levar a uma construção de conhecimento efetiva e significativa para todos os envolvidos (Piaget, 1973, 2005), filmes (Schafer et al, 1999), músicas, livros, software ou produtos de consumo em geral. Através de técnicas de análise de dados usadas pelos sistemas de recomendação, estes podem ajudar uma pessoa a encontrar os itens que melhor atendam suas necessidades (Adonie, 2005). Entre os trabalhos relacionados estão os seguintes: Na educação,Reategui et al (2006)propuseram o emprego dos sistemas de recomendação tanto para identificar conteúdo relevante para cada estudante, quanto para promover a interação entre estes. ...
    Article
    Full-text available
    Recomendação baseado em Filtragem Colaborativa (FC), também conhecida como Filtragem Social (FS), para ambientes móveis. O modelo proposto foi implementado em um protótipo visando a recomendação de artigos científicos. A avaliação das recomendações recebidas foi realizada pelos usuários utilizando uma escala likert de 5 pontos. No final deste artigo são descritos alguns experimentos realizados e os resultados obtidos com a aplicação do modelo proposto são discutidos. Palavras-chaves:filtragem colaborativa, sistemas de recomendação, computação móvel.
  • ... The paper concludes with the results of the preliminary tests of WiseUncle performed by three e-commerce businesses. A preliminary paper describing our results may be found in [2]. ...
    Article
    Full-text available
    Recommender systems (RS) are being used by an increasing number of e-commerce sites to help consumers find the personally best products. We define here the criteria that a RS should satisfy, drawing on concepts from behavioral sci-ence, computational intelligence, and data mining. We present our conclusions from building the WiseUncle RS and give its general description. Rather than being an advisor for a particular application, WiseUncle is a generic RS, a platform for gener-ating application-specific advisors.
  • ... Através de técnicas de análise de dados usadas pelos sistemas de recomendação, estes podem ajudar uma pessoa a encontrar os itens que melhor atendam suas necessidades (Adonie, 2005). Entre os trabalhos relacionados estão os seguintes: Na educação, Reategui et al (2006) propuseram o emprego dos sistemas de recomendação tanto para identificar conteúdo relevante para cada estudante, quanto para promover a interação entre estes. ...
    Article
    Full-text available
    Recommender Systems (SR) are widely used nowadays specially in e-commerce sites, for example, Amazon.com and Submarino.com. In this project, we propose the use of such systems in the area of education, specifically for the recommendation of contents in mobile devices. The advantage of using Recommender Systems in mobile devices is that it is an easy way to deliver recommendations to the users. Based on this scenario, this paper presents a model of a Recommender System based on Collaborative filtering (CF) - known as Social Filtering (SF) tool for mobile environments. The proposed model was implemented in a prototype aimed to recommend academic papers through mobile devices. The evaluation of the received recommendations by users was conducted using a likert scale of 5 points. At the end of this paper some experiments are described and the results obtained with the implementation of the proposed model are discussed.
  • ... The paper concludes with the results of the preliminary tests of WiseUncle performed by three e-commerce businesses. A preliminary paper describing our results may be found in [2] ...
  • Conference Paper
    In this paper, the design of an asynchronized messaging system where a user may proactively seek for answers or advice by depositing a question on the messaging system is presented. The messaging system will automatically disseminate the question which is related to a specific location, to a group of users who are either within the proximity currently or have just recently been there. The users are supposed to help each other; karma or some sort of point/score system can be added to distinguish the most helpful users in the community. A name is suggested for this system called “Lemmings” which stands for Location-based Mutual and Mobile Information Navigation System; it is based on a classical video game where a group of creatures have to work and win through the puzzle game together. The design presented is modular and it comprised of the logics, the data flows, and the database schema.
  • Conference Paper
    Full-text available
    Recommender systems help users sort through vast quantities of information. Sometimes, however, users do not know if they can trust the recommendations they receive. Adding a confidence metric has the potential to improve user satisfaction and alter user behavior in a recommender system. We performed an experiment to measure the effects of a confidence display as a component of an existing collaborative filtering- based recommender system. Minimal training improved use of the confidence display compared to no training. Novice users were less likely to notice, understand, and use the confidence display than experienced users of the system. Providing training about a confidence display to experienced users greatly reduced user satisfaction in the recommender system. These results raise interesting issues and demonstrate subtle effects about how and when to train users when adding features to a system.
  • Article
    Full-text available
    Collaborative recommender systems allow personalization for e-commerce by exploiting similarities and dissimilarities among customers' preferences. We investigate the use of association rule mining as an underlying technology for collaborative recommender systems. Association rules have been used with success in other domains. However, most currently existing association rule mining algorithms were designed with market basket analysis in mind. Such algorithms are inefficient for collaborative recommendation because they mine many rules that are not relevant to a given user. Also, it is necessary to specify the minimum support of the mined rules in advance, often leading to either too many or too few rules; this negatively impacts the performance of the overall system. We describe a collaborative recommendation technique based on a new algorithm specifically designed to mine association rules for this purpose. Our algorithm does not require the minimum support to be specified in advance. Rather, a target range is given for the number of rules, and the algorithm adjusts the minimum support for each user in order to obtain a ruleset whose size is in the desired range. Rules are mined for a specific target user, reducing the time required for the mining process. We employ associations between users as well as associations between items in making recommendations. Experimental evaluation of a system based on our algorithm reveals performance that is significantly better than that of traditional correlation-based approaches.
  • Article
    Full-text available
    This article describes a service for providing literature recommendations, which is part of a networked digital library project whose principal goal is to develop technologies for supporting digital services. The proposed literature recommendation system makes use of the Web usage logs of a literature digital library. The recommendation framework consists of three sequential steps: data preparation of the Web usage log, discovery of article associations, and article recommendations. We discuss several design alternatives for conducting these steps. These alternatives are evaluated using the Web logs of our university’s electronic thesis and dissertation (ETD) system. The proposed literature recommendation system has been incorporated into our university’s ETD system, and is currently operational.
  • Conference Paper
    Full-text available
    Recommender systems use people's opinions about items in an information domain to help people choose other items. These systems have succeeded in domains as diverse as movies, news articles, Web pages, and wines. The psychological literature on conformity suggests that in the course of helping people make choices, these systems probably affect users' opinions of the items. If opinions are influenced by recommendations, they might be less valuable for making recommendations for other users. Further, manipulators who seek to make the system generate artificially high or low recommendations might benefit if their efforts influence users to change the opinions they contribute to the recommender. We study two aspects of recommender system interfaces that may affect users' opinions: the rating scale and the display of predictions at the time users rate items. We find that users rate fairly consistently across rating scales. Users can be manipulated, though, tending to rate toward the prediction the system shows, whether the prediction is accurate or not. However, users can detect systems that manipulate predictions. We discuss how designers of recommender systems might react to these findings.
  • Article
    This paper presents an agent-based Web-mining approach to Internet shopping. We propose a fuzzy neural network to tackle the uncertainties in practical shopping activities, such as consumer preferences, product specification, product selection, price negotiation, purchase, delivery, after-sales service and evaluation. The fuzzy neural network provides an automatic and autonomous product classification and selection scheme to support fuzzy decision making by integrating fuzzy logic technology and the backpropagation feed forward neural network. In addition, a new visual data model is introduced to overcome the limitations of the current Web browsers that lack flexibility for customers to view products from different perspectives. Such a model also extends the conventional data warehouse schema to deal with intensive data volumes and complex transformations with a high degree of flexibility for multiperspective visualization and morphing capability in an interactive environment. Furthermore, an agent development tool named "Aglet" is used as a programming framework for system implementation. The integration of dynamic object visualization, interactive user interface and data mining decision support provides an effective technique to close the gap between the "real world" and the "cyber world" from a business perspective. The experimental results demonstrate the feasibility of the proposed approach for Web-based business transactions.
  • Article
    Full-text available
    Recommender systems have become valuable resources for users seeking intelligent ways to search through the enormous volume of information available to them. One crucial unsolved problem for recommender systems is how best to learn about a new user. In this paper we study six techniques that collaborative filtering recommender systems can use to learn about new users. These techniques select a sequence of items for the collaborative filtering system to present to each new user for rating. The techniques include the use of information theory to select the items that will give the most value to the recommender system, aggregate statistics to select the items the user is most likely to have an opinion about, balanced techniques that seek to maximize the expected number of bits learned per presented item, and personalized techniques that predict which items a user will have an opinion about. We study the techniques thru offline experiments with a large preexisting user data set, and thru a live experiment with over 300 users. We show that the choice of learning technique significantly affects the user experience, in both the user effort and the accuracy of the resulting predictions.
  • Article
    Full-text available
    We present a soft computing recommendation system named Talk Mine,to advance adaptive web and digital library technology. Talk Mine leads different databases or websites to learn new and adapt existing keywords to the categories recognized by its communities of users. It uses distributed artificial intelligence algorithms and soft computing technology. Talk Mine is currently being implemented for the research library of the Los Alamos National Laboratory under the Active Recommendation Project (http:// arp. lanl. gov ). Talk Mine is based on the integration of distributed knowledge networks using Evidence Sets, an extension of fuzzy sets. The identification of the interests of users relies on a process of combining several fuzzy sets into evidence sets, which models an ambiguous “and/or” linguistic expression. The interest of users is further finetuned by a human-machine conversation algorithm used for uncertainty reduction. Documents are retrieved according to the inferred user interests. Finally, the retrieval behavior of all users of the system is employed to adapt the knowledge bases of queried information resources. This adaptation allows information resources to respond well to the evolving expectations of users. In this article the distributed architecture of Talk Mine is presented together with a description of its implementation in the Active Recommendation Project. In particular, the characterization of information resources as interacting distributed memory banks is presented. Evidence sets and the operations to produce them from several fuzzy sets are detailed. The conversation and adaptation algorithms used by TalkMine to interact automatically with users is described.