Toward an Adaptive Personalized Web

The main goal of the proposed research is to develop novel techniques for adaptive and personalized information searching and navigation guidance on the web. Personalized techniques cater to specific users and adaptive methods learn from the users' (potentially changing) behavior. Our effort will lead to more productive and effective use of the increasing popular web. Rather than requesting the user to provide his/her interests, we propose modeling user interests non-intrusively (no direct user involvement) based on access behavior. We represent user profiles based on access patterns (Web Access Graphs) and interests (Page Interestingness Estimators). Instead of using syntactic and mathematical techniques for page representation, we incorporate semantic information from WordNet. Skewed training class distributions (interesting pages are much fewer than uninteresting ones) and adaptation over time (changing user behavior) are handled using multiple learned models. To efficiently process large amounts of data, we investigate parallelized methods for collecting, collating, and ranking pages. Personalized analyses of results from search engines yield more precise and relevant pages. We also identify similar users based on access behaviors and interests for collaborative filtering.



Related Work

Last modified: Sat Aug 27 18:36:36 EDT 2005