Download AI Techniques for Personalized Recommendation Tutorial

AAAI 2002 Tutorial Proposal
AI Techniques for Personalized Recommendation
Anthony Jameson
[email protected]
DFKI, German Research Center for Artificial Intelligence
Joseph A. Konstan and John Riedl
{konstan;riedl}@cs.umn.edu
Department of Computer Science and Engineering
University of Minnesota
Tutorial Goals and Learning Objectives
This tutorial aims to provide a coherent view of the broad range of artificial intelligence techniques that
have been and are currently being developed for the personalized recommendation of products,
documents, or collaborators. Whereas previous tutorials have usually focused on one particular type of
learning or inference technique, this tutorial will cover techniques ranging from all major extensions of
collaborative filtering to case-based reasoning and decision-theoretic approaches. These techniques are
discussed in relation to the wide variety of research and commercial applications that employ AI-based
recommendation. The algorithms, applications, and interfaces presented will be illustrated through a set of
case studies drawn from both research and commercial systems.
The target audience comprises (a) attendees from industry who seek an authoritative update on
recommendation technology; (b) AI researchers who see personalized recommendation as an attractive
application area; and (c) researchers already working on one type of recommendation technology who
would like to learn more about alternative approaches.
After taking this tutorial, attendees will (1) be familiar with a broad set of recommendation techniques and
be able to identify appropriate techniques for specific recommendation challenges; (2) be familiar with the
major recommender interfaces and able to match appropriate interfaces to specific user and designer
goals; and (3) be familiar with a large set of recommendation applications, and able to use these as models
when designing and developing future applications.
This tutorial is novel in terms of (a) the wide range of recommendation techniques covered and (b) the
tight relationships between the technical material covered and the presentation of experience concerning
the practical deployment of recommendation technology. These aspects require a full-day format:
Whereas the presenters have conducted half-day tutorials on subsets of this material, it would not be
possible to fit this broader treatment into a half-day tutorial.
Content
Schedule
The following proposed schedule indicates the topics to be covered and the time to be allotted to each
topic:
Part 1, Introduction, applications, and interfaces:
9:00
9:20
9:35
9:50
10:05
10:20
10:50
11:20
Introduction
Discussion of tutorial objectives, background on systems for personalized
recommendation, gathering background and objectives of participants (to guide
discussion later in tutorial).
Overview of the application space of personalized recommendation, presentation of basic
interfaces
E-commerce applications and interfaces
Information filtering and display applications and interfaces
Off-web applications and interfaces
Case studies on applications and interfaces
In addition to examples presented within each section, these case studies help wrap up the
“applications and interfaces” section.
Coffee Break
Overview of relevant AI techniques
Classification of techniques to be discussed in terms of (a) types of data taken into
account (user attributes, item attributes, ratings or other evaluations, explicit selfdescriptions) and (b) AI methods applied
Part 2, Techniques Based on Collaborative Filtering:
11:30
11:40
12:05
12:20
12:30
Collaborative filtering algorithms
k-nearest-neighbor algorithms
Other machine learning algorithms for collaborative filtering
Dimensionality reduction algorithms
Case studies on collaborative filtering and related algorithms
In addition to the cases presented within each section; these case studies help wrap up the
section on algorithms based on collaborative filtering
1:00
Lunch Break
2
Part 3, Techniques Not Based on Collaborative Filtering:
2:00
2:10
2:30
3:00
3:30
Demographically based models
Algorithms based on case-based reasoning
Methods based on decision-theoretic models of choice:
Common assumptions of decision-theoretic models (basic concepts of multi-attribute
utility theory)
Methods for learning importance weights from users' behavior
Methods for eliciting preferences more explicitly
Case studies of techniques not based on collaborative filtering
Concluding comparative discussion of techniques
3:45
Coffee Break
Part 4, Applying and Extending What Has Been Learned So Far:
4:15
4:45
5:45
6:00
Exercise: Recommender system design
Advanced topics: chosen from among:
Privacy issues in recommender systems, extended uses of data obtained from
recommender systems, anthropomorphism in recommender systems, issues in
mobile/wireless recommender systems, meta-level recommender systems, agent
participation in recommender systems, other new and interesting topics
Conclusions, questions
Adjourn
How the Tutorial will be Conducted
The tutorial is a mix of lecture, interactive discussion, and class exercises.
Jameson, Konstan and Riedl will each teach one-third of the tutorial, with the others present to assist in
discussions, note-taking, etc.
Tutorial Notes
The tutorial notes will include copies of the slides, an annotated bibliography with pointers to about 40-60
important papers, and pointers to introductory material available on-line.
Supplemental Materials
Attendees will receive copies of several key papers for which reprint rights can be obtained. As an
example, one possible list is:
•
Breese, J., Heckerman, D., F Kadie, C. (1998). Empirical analysis of predictive algorithms for
collaborative filtering. In G. F. Cooper & S. Moral (Hrsg.), Uncertainty in Artificial Intelligence:
Proceedings of the Fourteenth Conference (S. 43–52). San Francisco: Morgan Kaufmann.
•
•
•
3
Burke, R. D. (2000). Knowledge-based recommender systems. Encyclopedia of Library and
Information Science.
Herlocker, J. L., Konstan, J. A., Borchers, A., & Riedl, J. (1999). An algorithmic framework for
performing collaborative filtering. Proceedings of the 1999 Conference on Research and
Development in Information Retrieval.
Linden, G., Hanks, S., & Lesh, N. (1997). Interactive assessment of user preference models: The
automated travel assistant. In A. Jameson, C. Paris, & C. Tasso (Hrsg.), User modeling:
Proceedings of the Sixth International Conference, UM97 (S. 67–78). Wien, New York: Springer
Wien New York.
•
•
Pazzani, M. J. (1999). A framework for collaborative, content-based and demographic filtering.
Artificial Intelligence Review, 13, 393–408.
Schafer, J. B., Konstan, J. A., & Riedl, J. (2001). E-commerce recommendation applications.
Journal of Data Mining and Knowledge Discovery, 5(1/2), 115–152.
Sample Slides
The last few pages of this proposal show typical slides from the presenters’ previous relevant tutorials.
Tutorial Description
Personalized recommendation of products, documents, and collaborators has become an extremely
important way of meeting user needs in commerce, information provision, and community services,
whether on the web, through mobile interfaces, or on traditional desktop interfaces. This tutorial first
reviews the types of personalized recommendation that are being used commercially and in research
systems. It then looks at the broad range of artificial intelligence techniques that have been developed for
recommendation systems, exploring both the methods themselves and the ways in which they are applied.
Attendees will learn how to match recommendation techniques to applications, and they will explore
several case studies of actual applications of personalized recommendation technology.
The tutorial presupposes a general knowledge of AI. Some previous familiarity with issues of personalized
recommendation is desirable but not essential.
Information About the Presenters
One of the strengths of this tutorial, particularly for the AAAI audience, is the fact that it brings together
broad experience with AI techniques for user-adaptive systems (Jameson) and extensive research and
industrial experience in the specific personalization subarea of recommender systems based on
collaborative filtering (Konstan and Riedl). The result will be a tutorial with extensive coverage of
recommendation techniques supported by a rich collection of application examples and design experience.
Anthony Jameson
Addresses
Anthony Jameson, Principal Researcher / Adjunct Professor
German Research Institute for Artificial Intelligence (DFKI)
/ International University in Germany
Stuhlsatzenhausweg 3
66123 Saarbrücken Germany
[email protected]
http://dfki.de/~jameson
Phone: (+49)(681) 302-5079
Fax: (+49)(681) 302-5020
Background in the Tutorial Area
Jameson'
s first publication on what is now called personalized recommendation appeared in the
proceedings of IJCAI 1983. Since then he has published widely on personalization and user-adaptive
systems more generally, as is illustrated by the following sample of publications:
4
•
Jameson, A. (1983). Impression monitoring in evaluation-oriented dialog: The role of the
listener'
s assumed expectations and values in the generation of informative statements.
Proceedings of the Eighth International Joint Conference on Artificial Intelligence, Karlsruhe,
pp. 616–620.
•
Jameson, A. (1989). But what will the listener think? Belief ascription and image maintenance in
dialog. In A. Kobsa & W. Wahlster (Eds.), User models in dialog systems (pp. 255–312). Berlin:
Springer.
Jameson, A., Schäfer, R., Simons, J., & Weis, T. (1995). Adaptive provision of evaluationoriented information: Tasks and techniques. In C. S. Mellish (Eds.), Proceedings of the
Fourteenth International Joint Conference on Artificial Intelligence (pp. 1886–1893). San
Mateo, CA: Morgan Kaufmann.
Jameson, A. (1996). Numerical uncertainty management in user and student modeling: An
overview of systems and issues. User Modeling and User-Adapted Interaction, 5, 193–251.
•
•
•
Jameson, A. (in press). Adaptive interfaces and agents. In J. A. Jacko & A. Sears (Eds.),
Handbook of human-computer interaction in interactive systems. Mahwah, NJ: Erlbaum.
The last two publications are broad survey articles.
A complete list of Jameson'
s publications since 1993, most of which are available electronically, can be
accessed via his web homepage (http://dfki.de/ jameson/).
He has recently presented the following tutorials that are closely related to the topic of the proposed
tutorial:
•
Personalization for E-Commerce. UM 2001, Eighth International Conference on User Modeling,
Sonthofen, Germany.
•
Systems That Adapt to Their Users. IJCAI 2001, Seventeenth International Joint Conference on
Artificial Intelligence, Seattle.
•
•
•
Designing Systems That Adapt to Their Users (full-day tutorial). CHI 2001, Conference on
Human Factors in Computing Systems, Seattle. (An updated version will be presented at CHI
2002, Minneapolis.)
Designing User-Adaptive Systems. IUI 2001, International Conference on Intelligent User
Interfaces, Santa Fe, New Mexico.
User-Adaptive Systems: An Integrative Overview. UM 1999, Seventh International Conference
on User Modeling, Banff, Canada and IJCAI 1999, Sixteenth International Joint Conference on
Artificial Intelligence, Stockholm.
Evidence of Teaching Experience
Jameson has taught three university courses on personalization: one in the summer of 2001 at the
International University in Germany (summer of 2001) and two at Saarland University (fall/winter
1998/1999, and fall/winter 2000/2001). More generally, he has taught university courses since 1986 on
various topics in AI, cognitive science, and human-computer interaction at the University of Nijmegen,
Saarland University, and the International University in Germany. Though affiliated primarily with DFKI,
in 2001 he was appointed Adjunct Associate Professor at the International University in Germany in
recognition of teaching excellence.
Evidence of Scholarship in AI or Computer Science
Most of the Jameson'
s papers cited above include a substantial AI component. Two other recent papers
presented at AI conferences are the following:
5
•
Jameson, A., & Wittig, F. (2001). Leveraging data about users in general in the learning of
individual user models. In B. Nebel (Ed.), Proceedings of the Seventeenth International Joint
Conference on Artificial Intelligence (pp. 1185–1192). San Francisco, CA: Morgan Kaufmann.
Wittig, F., & Jameson, A. (2000). Exploiting qualitative knowledge in the learning of conditional
•
probabilities of Bayesian networks. In C. Boutilier& M. Goldszmidt (Eds.), Uncertainty in
Artificial Intelligence: Proceedings of the Sixteenth Conference (S. 644–652). San Francisco:
Morgan Kaufmann.
The following three papers, each of which has an AI component, were awarded best-paper prizes at their
respective conferences:
•
Müller, C., Großmann-Hutter, B., Jameson, A., Rummer, R., & Wittig, F. (2001). Recognizing
time pressure and cognitive load on the basis of speech: An experimental study. In M. Bauer, P.
Gmytrasiewicz, & J. Vassileva (Eds.), UM2001, User modeling: Proceedings of the Eighth
International Conference (pp. 24–33). Berlin: Springer.
•
•
Berthold, A., & Jameson, A. (1999). Interpreting symptoms of cognitive load in speech input. In
J. Kay (Ed.), UM99, User modeling: Proceedings of the Seventh International Conference (pp.
235–244). Wien, New York: Springer Wien New York.
Jameson, A., Schäfer, R., Weis, T., Berthold, A.,& Weyrath, T. (1999). Making systems sensitive
to the user'
s time and working memory constraints. In M. T. Maybury (Ed.), IUI99: International
Conference on Intelligent User Interfaces (pp. 79–86). New York: ACM.
Joe Konstan and John Riedl
Addresses
Joseph A. Konstan, Associate Professor
Department of Computer Science and Engineering
University of Minnesota
4-192 EE/CS Building
200 Union Street SE
Minneapolis, MN 55455
[email protected]
http://www-users.cs.umn.edu/~konstan/
Phone: (+1)(612) 625-1831
John Riedl, Associate Professor
Department of Computer Science and Engineering
University of Minnesota
5-211 EE/CS Building
200 Union Street SE
Minneapolis, MN 55455
[email protected]
http://www-users.cs.umn.edu/~riedl/
Phone: (+1)(612) 624-7372
6
Background in the Tutorial Area
John Riedl co-founded the GroupLens recommender system project with Paul Resnick in 1992, and he
and Joe Konstan have been co-directing the project since 1995. GroupLens was one of the earliest
recommender systems projects, and is now one of the most widely known. Riedl and Konstan also cofounded Net Perceptions, a company that has commercialized the results of their research since 1996.
They have published broadly in the area of recommender systems, including:
7
•
Schafer, J. B., Konstan, J. A., & Riedl, J. (2001). E-commerce recommendation applications.
Journal of Data Mining and Knowledge Discovery, 5(1/2), 115–152.
•
Sarwar, B. M., Konstan, J. A., and Riedl, J. "Distributed Recommender Systems: New
Opportunities for Internet Commerce". In the book "Internet Commerce and Software Agents:
Cases, Technologies and Opportunities" Rahman, S., and Bignall, R. eds., 2001.
•
Herlocker, J., Konstan, J., and Riedl, J., Explaining Collaborative Filtering Recommendations.
ACM 2000 Conference on Computer Supported Cooperative Work , December 2-6, 2000.
•
Sarwar, B. M., Karypis, G., Konstan, J. A., and Riedl, J. "Analysis of Recommender Algorithms
for E-Commerce". ACM E-Commerce 2000 Conference. Oct. 17-20, 2000.
•
Sarwar, B. M., Karypis, G., Konstan, J. A., and Riedl, J. "Application of Dimensionality
Reduction in Recommender System – A Case Study". Full length paper at ACM WebKDD 2000
Web Mining for E-Commerce Workshop.
•
Herlocker, J. L., Konstan, J. A., Borchers, A., & Riedl, J. (1999). An algorithmic framework for
performing collaborative filtering. Proceedings of the 1999 Conference on Research and
Development in Information Retrieval.
•
Schafer, J.B., Konstan, J., and Riedl, J., Recommender Systems in E-Commerce. Proceedings of
the ACM Conference on Electronic Commerce, November 3-5, 1999.
•
Good, N., Schafer, J.B., Konstan, J., Borchers, A., Sarwar, B., Herlocker, J., and Riedl, J.,
Combining Collaborative Filtering with Personal Agents for Better Recommendations.
Proceedings of the 1999 Conference of the American Association of Artifical Intelligence (AAAI99). pp. 439-446.
•
Herlocker, J., Konstan, J., Borchers, A., Riedl, J.. An Algorithmic Framework for Performing
Collaborative Filtering. Proceedings of the 1999 Conference on Research and Development in
Information Retrieval. Aug. 1999.
•
Sarwar, B., Konstan, J., Borchers, A., Herlocker, J., Miller, B., and Riedl, J..Using Filtering
Agents to Improve Prediction Quality in the GroupLens Research Collaborative Filtering System.
Proceedings of the 1998 Conference on Computer Supported Cooperative Work. Nov. 1998.
•
Konstan, J., Miller, B., Maltz, D., Herlocker, J., Gordon, L., and Riedl, J.GroupLens: Applying
Collaborative Filtering to Usenet News. Communications of the ACM 40, 3 (1997), 77-87.
•
Miller, B., Riedl, J., and Konstan, J. Experiences with GroupLens: Making Usenet useful again.
Proceedings of the 1997 Usenix Winter Technical Conference. Jan 1997.
•
Resnick, P., Iacovou, N., Sushak, M., Bergstrom, P., and Riedl, J. GroupLens: An open
architecture for collaborative filtering of netnews. Proceedings of the 1994 Computer Supported
Collaborative Work Conference.
Joe Konstan and John Riedl have taught three versions of a tutorial on recommender systems, at the 1996
and 2000 ACM conferences on Computer-Supported Cooperative Work, and at the 2000 ACM ECommerce conference. The tutorial has evolved with each offering, but the most recent version includes
an introduction to a variety of algorithms and a set of ten case studies of systems in use. The CSCW 2000
tutorial received the top overall score of tutorials at the conference (evaluations are included in the table
below). Konstan and Riedl are completing a book on recommender systems that will be published some
time in 2002.
CSCW 2000 evaluation results for Tutorial 10
(n=12)
Means
Item
This
tutorial
All
tutorials
1. Overall, this was an excellent tutorial
4.58
4.12
2. The tutorial notes will be a valuable 4.17
resource
4.16
3. I learned much from this tutorial
4.58
3.81
4. The slides were of high quality
4.08
3.87
5. The tutorial description was accurate
4.08
3.95
6. The tutorial notes supported the 4.50
presentation
4.39
7. The tutorial addressed social and value 4.42
implications of CSCW systems
3.64
8. I would enthusiastically recommend 4.58
this tutorial
3.89
9. I am principally a system designer and 3.25
developer
3.18
10. I am a regular tutorial participant at 3.50
conferences that I attend
3.09
11. The tutorial addressed technical 3.83
implications of CSCW systems
3.47
8
Evidence of Teaching Experience
Joe Konstan has been on the faculty of the Computer Science and Engineering Department of the
University of Minnesota since 1992. He has taught many University Computer Science and Software
Engineering courses in the areas of human-computer interaction, programming, and computer graphics.
He has consistently earned among the top student evaluations in the department, and was awarded the
George Taylor award as the best young teacher in the college in 1999. He has also taught many industry
short courses.
John Riedl has been on the faculty of the Computer Science and Engineering Department of the
University of Minnesota since 1990. He has taught over twenty University courses in the areas of objectoriented programming, operating systems, and related areas. He has earned the award as the department’s
best teacher three times, and was awarded the George Taylor award as best young teacher in the college in
1996.
Evidence of Scholarship in AI or Computer Science
Konstan and Riedl are both tenured associate professors in the Department of Computer Science and
Engineering at the University of Minnesota. In addition to their work together on recommender systems
listed above, Konstan has published extensively in human-computer interaction and multimedia, and Riedl
has published extensively in collaborative systems, scientific databases, and visualization. Three example
articles from each, apart from the collaborative filtering articles listed above, include:
Konstan
• J. Carlis and J. Konstan. "Interactive Visualization of Serial Periodic Data," Proceedings of ACM
UIST 98, San Francisco, November 1998, pp. 29-38.
•
•
T. Gustafson, J.B. Schafer, and J. Konstan. "Agents in Their Midst: Evaluating User Adaptation to
Agent-Assisted Interfaces," Proceedings of the 1998 International Conference on Intelligent User
Interfaces, San Francisco: ACM, January 1998, pp. 163-170.
J. Konstan, L. Lichtblau, P. Sturm, and J. McLeod. "ISAP: A Web-Based Pharmacology Course
and Study Environment," Computers and Education special issue on Multimedia, December 1997,
pp. 63-71.
Riedl
• Vahid Mashayekhi, Janet Drake, Wei-Tek Tsai, and John Riedl. Distributed, collaborative
software inspection. IEEE Software, 10(5), September 1993.
• M. Claypool, J. Riedl, J. Carlis, G. Wilcox, R. Elde, E. Retzel, A. Georgeopolous, J. Pardo, K.
Ugurbil, B. Miller, and C. Honda. Network requirements for 3d flying in a zoomable brain
database. IEEE JSAC Special Issue on Gigabit Networking, 13(5):816-827, June 1995.
• Ed H. Chi, John Riedl, Phillip Barry, and Joseph Konstan. Principles for information
visualization spreadsheets. IEEE Computer Graphics and Applications (Special Issue on
Visualization), pp. 30-38, July 1998.
9
Note: The graphic quality of the slide reproductions within this document is lower than the original quality
of the slides. Also, the slides for the AAAI tutorial will be stylistically more consistent.
Slides from Jameson's UM 2001 Tutorial
10
Slides from Jameson's IJCAI 2001 Tutorial
11
New Slide Illustrating Part of Integrative Framework for AAAI 2002
Tutorial
12
Slides from Konstan and Riedl's CSCW 2000 Tutorial
ratings
C.F. Engine
Ratings
Joe
John
Susan
Pat
Jean
Ben
Nathan
13
request
predictions
recommendations
Correlations
Hoop
Dreams
D
A
A
D
A
F
D
Star
Wars
A
F
A
A
C
A
Neighborhood
Pretty Titanic
Woman
B
D
A
A
A
D
A
C
C
Blimp
?
F
A
Rocky
XV
?
A
A
F
A
14
$
(
)
(
*
$'
"#
"$
%
"&
%
!
$
(
)
(
*
$'
" +%
,
"+
'
,
-
!
15
%,
/
'
0
0
1 ,,
,
/
%
'
'
-
2 , %
/3
0'
/ 0'
%
/
/ &4
/
%
/
/
.
.!
16