Rel-Div: Generating Diversified Query Interpretations from Semantic Relations

Ramakrishna Bairi, A. Ambha, Ganesh Ramakrishnan

Abstract: Accelerated growth of the World Wide Web has resulted in an increase in appetite for searching over Internet to fulfill the information needs. Understanding user intent plays a pivotal role in determining the quality of search results and improving user satisfaction. But short, ambiguous or underspecified queries make the process of determining the concealed user intention harder. Identifying diversified but relevant interpretations for a query with an impressive accuracy, is still an active area of research. These varied interpretations originate from entities associated with the user query and relations between the entities. We address the problem of generating diverse but relevant interpretations by utilizing an Internet encyclopedia (Wikipedia) as a primary source entities and their relations. By preprocessing the encyclopedia, we build a rich repository of Semantic Relations, which characterize these entities and their relationships. We present algorithms to enumerate pertinent interpretations for a query based on the repository. The proposed approach uses the repository to generate candidate interpretations over which we apply graph based iterative approaches to generate diversified and relevant results. We empirically evaluate the effectiveness of our approach, with the ‘Query Relevance and Understanding’ dataset of TREC 2011 workshop and AMBIENT (Ambiguous Entities) dataset.