Learning Implicit User Interest Hierarchy for Web Personalization

Learning Implicit User Interest Hierarchy for Web Personalization by Hyoung-rae Kim Dissertation Advisor: Philip K. Chan, Ph.D. Most web search engines are designed to serve all users in a general way, without considering the interests of individual users. In contrast, personalized web search engines incorporate an individual user's interests when choosing relevant web pages to return. In order to provide a more robust context for personalization, a user interest hierarchy (UIH) is presented. The UIH extracts a continuum of general to specific user interests from web pages and generates a uniquely personalized order to search results. This dissertation consists of five main parts. First, a divisive hierarchical clustering (DHC) algorithm is proposed to group words (topics) into a hierarchy where more general interests are represented by a larger set of words. Second, a variable-length phrase-finding (VPF) algorithm that finds meaningful phrases from a web page is introduced. Third, two new desirable properties that a correlation function should satisfy are proposed. These properties will help understand the general characteristics of a correlation function and help choose or devise correct correlation functions for an application domain. Fourth, methods are examined that (re)rank the results from a search engine depending on user interests based on the contents of a web page and the UIH. Fifth, previously studied implicit