Personalization of Information Retrieval in Different Types of Tasks

[email protected]; [email protected]; [email protected]; [email protected] ABSTRACT In this paper, we propose to personalize search result content through modeling multiple user behavioral measures in different ways as evidence for implicit relevance feedback for different types of search tasks. The point of this personalization is to predict potentially useful documents based on the type of task, and on multiple behaviors indicative of document usefulness. In particular, task type was regarded as the key contextual factor in information retrieval. To represent task types, a faceted classification scheme was used. User interaction behaviors were logged from the clientside for analysis. Both Recursive Partitioning and Logistic Regression methods were employed to identify the most important behavioral signals and to generate the predictive models to predict document usefulness for relevance feedback. Our results demonstrate that combining multiple behaviors on content pages and search result pages can improve the prediction of useful documents. However, different combinations of behavioral measures should be used to predict document usefulness in different task types.