TITLE: A Study of Factors and Approaches in the Design of Hyperlink Recommender Systems Based on Implicit User Modeling

WHEN: Wednesday Dec 10, 1PM

WHERE: Large Commons Room, 5th Fl. IS Bldg.

WHO: Denis L. Nkweteyim

Committee:

1) Prof. Stephen Hirtle, DIST, Committee Chair
2) Prof. Jerrold May, Katz Graduate School of Business
3) Prof. Michael Spring, DIST
4) Prof. Paul Munro, DIST
5) Prof. Peter Brusilovosky, DIST

ABSTRACT: The current information glut has led to the search of ways to personalize the information people receive so that they receive the information they need and little or no information that is not relevant to their current information needs.  Personalization systems make use of a user profile to tailor the information users receive to their individual needs. Recommender systems do this filtering during a live interaction with the system.  Most of the research on recommender systems have been in the domain of E-commerce, and have been based largely on user profiles that are explicitly provided by the user.  This dissertation research considers the problem of recommending relevant hyperlinks to a user navigating a Web site, with the aid of user profiles that are implicitly determined by observing the user's navigation behavior, and comparing it to the navigation behaviors of past users of the Web site.  The project borrows from current techniques in recommender system design, and proposes improvements as well as new approaches in the following areas: user modeling for hyperlink recommendation; useful implicit indicators of Web page relevance; and Web usage mining. Evaluation criteria are also proposed for effectiveness of the user modeling approaches and implicit indicators of page relevance, and for efficiency of a new data mining approach proposed in the study.