CS 6604
Recommender Systems
Instructor
-
Naren Ramakrishnan
McBryde 629, 231-8451
naren@cs.vt.edu
Office Hours: MW 2-4pm
Class Meeting Times and Contact Info:
-
MWF McBryde 322
12:20-1:10pm
Listserv: CS6604_15091@listserv.vt.edu
Pre-requisites
-
Graduate Student Standing.
One or more of CS5114 (algorithms), CS5485 (numerical
analysis), CS5604 (information retrieval), CS5614 (database systems), CS5724 (HCI), and CS 5714 (usability engineering).
My goal is to organize a diverse and rich student audience, to have a good
mix of backgrounds and interests. It is expected that students have a primary
motivation to study recommender systems, as (i) a field in its own right,
(ii) a methodology to personalize information in a domain of interest, and/or (iii)
an excellent example of integrated and amalgamated research in computer science.
Course Format
The course will be centered around presentations and discussions led by the instructor, with a semester-long student project. The instructor will attempt to obtain industry involvement for most student projects (already two companies have agreed to participate by providing access to data and other resources).There will be *no* homeworks/assignments. Student participation in discussions is a must and will constitute 25% of the grade. The remaining 75% of the grade will go to your project. Students are encouraged to work in groups of 2-3, depending on the size of the project.
Every project is expected to lead to a quality publication, in a premier conference, or a journal (The significance and originality of the research will hence directly affect the grade). There will be no exceptions. The work need not have appeared in print/circulation at the end of the semester, but should be in a form and of such quality that is clearly publish-worthy in the top-tier places. Incremental development and/or work not at the graduate student level will result in a failing grade.
One of the main goals of this course is to foster and encourage research productivity among students in our department (click here for an example of such course-work in a different university that resulted in three award winning papers.) I expect VT to steal the show at the forthcoming AAAI, CHI, or SIGIR conference(s).
Work Load
O(2-3) sleepless nights per week.Recommended and Required Books
None. All readings will be from current literature, available either on the web or via handouts.Topics and Areas
(in approximate chronological order; it might perhaps appear unsettling that a course devoted to reducing information overload should have so much content, but the instructor expects to personalize the delivery too! :-))- History, Motivation, and Overview
Greek Mythology (and the earliest examples of recommender systems), Onset of Information Overload, Filtering, Customization Dichotomies, CS Content Areas, Basic Principles, Integrated Frameworks, Application Areas (Books, Music, Movies, Web Sites) and their associated idiosyncracies.
- Models of Recommender Systems
Information Filtering, Collaborative Filtering, Ratings, User Evaluations, Diffusion, Effusivity & Sparsity in Recommender Systems, Social Networks, Random Graphs, "Expert Sites" (Askme.com, Epinions.com etc.), Experimental Algorithmics.
- Search Engines (a.k.a. "Public Transportation")
Indexing (Traditional Search Engines and why they don't index more than 15% of the web), Link-Based Approaches (Google, Clever), Semantic Corpora Analysis, Graph Structure in the Web (Bow-Tie Analysis), Diameter of the Web, Scaling Properties of the Web.
- Algorithmics
Linear Algebra (Bellcore's LSI etc.), Graph Theoretical Approaches (e.g. Kleinberg's HITS Framework), Jumping Connections, "Hammock Theory" (courtesy Ben Keller), Statistical Co-occurence Models.
- The Small World Phenomenon
Web as a Small World (why two web pages are at most 19 mouse clicks apart), Watts-Strogatz models, Heavy-Tailed Distributions on the Web, Random Graphs with Pre-Specified Degree Distributions, Implications of Small-World Behavior for making Recommendations
- Integrated Aproaches ("Hot Rods")
Non-Destructive Personalization (e.g. footprints), Destructive Personalization (e.g. PIPE), Classification Hierarchies, Navigation vs. Personalization, Taxonomies and Meta-Data, Adaptive Web Sites, Brief Mention of Applications in Non-Conventional Areas (e.g. Scientific Computing, Wireless System Design) and in Non-Traditional Environments (News on Demand, Handheld Devices)
- Commercial and Broadening Aspects of Personalization (Brief Mention Only)
Economics, Information Rules, Privacy Concerns, Community Standards, Improving Web Site Organization by Personalization, Mass Customization, Information Foraging, Ethical Aspects of Data Mining.