PROS: A Personalized Ranking Platform for Web Search

Current search engines rely on centralized page ranking algorithms which compute page rank values as single (global) values for each Web page. Recent work on topic-sensitive PageRank [6] and personalized PageRank [8] has explored how to extend PageRank values with personalization aspects. To achieve personalization, these algorithms need specific input: [8] for example needs a set of personalized hub pages with high PageRank to drive the computation. In this paper we show how to automate this hub selection process and build upon the latter algorithm to implement a platform for personalized ranking. We start from the set of bookmarks collected by a user and extend it to contain a set of hubs with high PageRank related to them. To get additional input about the user, we implemented a proxy server which tracks and analyzes user’s surfing behavior and outputs a set of pages preferred by the user. This set is then enrichened using our HubFinder algorithm, which finds related pages, and used as extended input for the [8] algorithm. All algorithms are integrated into a prototype of a personalized Web search system, for which we present a first evaluation.