UInterestManager : A Generic Framework to Model User Interests

Abstract

With the proliferation of documents on the Internet, it has become increasingly difficult for users to search and retrieve the information they want: information that is relevant and interesting. To address this problem, researchers have adopted personalized services and filtering systems that rely on user preferences. In this thesis, we present a novel method to model user interests automatically and generate a ranked list of preferences for each user in real time. The proposed framework here consists of three main steps to automatically model user preferences and generate a ranked list of the user's interests in real time. 1) Through the use of WordNet, a list of keywords which represent explicit interests is extracted from a text document and then implicit user preferences are determined by employing domain ontology. 2) A hierarchical data structure for each user stores interests and is updated for each new document read by that user. Additionally, there is a queue of recently accessed interests for each user that maintains these preferences based upon arrival time. 3) The final step includes generating a ranked list of top K interests for each user in real time. This ranking system employs the user model to create a combined list of both short-term and long-term interests. The results show significant improvement over some widely used methods.