Nonintrusive Personalization in Interactive Information Retrieval

Information retrieval systems are critical for overcoming information overload. A major deficiency of existing retrieval systems is that they generally lack user modeling and are not adaptive to individual users; information about the actual user and search context is largely ignored. Personalization is expected to break this deficiency and significantly improve retrieval accuracy. However, there are three limitations of existing personalization studies. First, they have a limited notion of personalization. Most existing studies are focused on ranking of information retrieval. Nevertheless, there are several stages in interactive information retrieval, query formulation, ranking, result presentation and relevance feedback. Second, many studies are doing personalization at the server side, where privacy is a serious concern. Third, most studies address the user modeling or context-sensitive ranking separately; they do not develop practical systems to demonstrate the effectiveness of personalized search. In this thesis, we break these limitations and develop a full-fledged nonintrusive personalized retrieval system. We take a broader view of personalization and apply the personalization in the ranking, result representation and relevance feedback stages of interactive information retrieval. We present a decision theoretic framework and propose specific retrieval methods based on statistical language model. Nonintrusiveness is characterized as protected privacy, progressive personalization and minimal user effort. To demonstrate the effectiveness of personalization in the real search, we design and develop a UCAIR (User-Centered Adaptive Information Retrieval) search agent, a personalized search system, which incorporates the personalization into retrieval, result representation and relevance feedback stages of interactive information retrieval in a nonintrusive way.