Pseudo-Relevance Feedback in Web Information Retrieval Using Segments’ Subjective Importance Values

Abstract

To make Web search more effective, we address the problem of articulating a user's information needs more effectively. This is done in an iterative way, by allowing the user to provide relevance feedback regarding individual segments of retrieved Web-pages. Previously applied methods are limited to discovering 'general importance values of segments' (based on the authors' 'objective views' i.e., main topics) rather than 'subjective importance values of segments' (based on a user's 'subjective view' i.e., personal information needs). In this paper, a user's interests are incrementally identified by allowing the user to iteratively select relevant keywords or phrases from a set of system-recommended candidate-keywords and candidate-phrases (i.e., pseudo-relevance feedback). It makes it possible to discover 'subjective importance values of segments' that can be dynamically changed by the user by indicating their interests regarding retrieved Web-pages. The important segments, selected by the user, provide higher precision of pseudo-relevance feedback for further Web information retrieval purposes.