Tutorial 1
Recommender Systems: Advanced Concepts in Research and Practice Joseph Konstan, University of Minnesota, USA
After a decade of increasingly common usage, researchers and practitioners
have learned a great deal about the design and implementation of recommender
systems. This half-day tutorial reviews lessons from research and practice
focusing on three key themes: (a) emergence of more sophisticated data-rich
and knowledge-rich algorithms, (b) expansion of evaluation criteria and
metrics beyond coverage and accuracy, and (c) engineering the user
experience. The tutorial will combine brief lecture segments and reviews of
recent research results with case studies and collaborative exercises. At
the end of the tutorial, researchers and advanced practitioners should both
have greater awareness of the state of research and practice in recommender
systems, and greater confidence in their ability to use such systems in
their own work.
Content
About the Presenter
 Joseph A. Konstan is Professor of Computer Science and Engineering at the
University of Minnesota and co-Director of the GroupLens Research Group.
His research on recommender systems ranges from user interface design, to
algorithm development and evaluation, to studies of user behavior. In 1996,
he co-founded Net Perceptions, Inc., a company that commercialized
recommender systems technology. He is co-author of Word of Mouse: The
Marketing Power of Collaborative Filtering, a book that reviews three dozen
good and poor examples of personalization in research and deployed systems.
He has taught over a dozen tutorials on recommender systems and
human-computer interaction. In addition to his recommender systems work,
Dr. Konstan has been active in research on online applications to assess and
promote public health.
Prof. Konstan received his Ph.D. from the University of California,
Berkeley, in 1993. He is President of ACM SIGCHI, an ACM Distinguished
Lecturer, and a member of he Editorial Board of the journal User Modeling
and User-Adapted Interaction.
For more details see homepage of Joseph A. Konstan.
Schedule
20 minutes: Recommender systems: historical overview and context
20 minutes: Case analysis: where do recommenders fail
15 minutes: Evaluation criteria: matching recommenders to user experience
20 minutes: Overview of attributes of recommenders and algorithm metrics
15 minutes: Modern recommender algorithms I: content, collaborative
<break>
20 minutes: Modern recommender algorithms II: collaborative (cont.),
knowledge-based, and hybrids
20 minutes: Design exercise: recommender algorithm design
20 minutes: Embedding recommenders in context: the user experience
20 minutes: Case analysis: application critiques
20 minutes: Summary and discussion
As indicated above, I like to mix some lecture content (never
more than about 30 minutes at a time) with case studies and small-group
exercises. Exercises include analyses of existing applications and
designing applications (e.g., defining requirements and selecting
appropriate data and algorithms to meet them). I encourage discussion
during the tutorial and also leave some time at the end for summative
discussion.
Description
After a decade of increasingly common usage, researchers and practitioners
have learned a great deal about the design and implementation of recommender
systems. This half-day tutorial reviews lessons from research and practice
focusing on three key themes: (a) emergence of more sophisticated data-rich
and knowledge-rich algorithms, (b) expansion of evaluation criteria and
metrics beyond coverage and accuracy, and (c) engineering the user
experience. The tutorial will combine brief lecture segments and reviews of
recent research results with case studies and collaborative exercises.
Target Audience
- Researchers interested in personalization and recommender systems
(including students)
- "Advanced practitioners" looking to move to the next step in
recommender applications
Prerequisite Knowledge and Preparation
While it is not necessary to have
any prior research or practice experience with recommender systems, it would
be helpful to have user experience with such systems. I recommend trying
the following sites (see notes for features) prior to the tutorial:
Expected Results
Higher awareness of current research results and practice
trends in recommender systems; greater competence in using such systems in
research or applications.
Other tutorials offered at AH2006:
|
