Understanding and Exploiting User’s Navigational Intent in Community Question Answering

Abstract

Verbose or colloquial queries take up a small but non-negligible proportion in the modern searching paradigms, and are commonly used in other platforms such as Community Question Answering (CQA), where answerers often include URLs as part of answers to provide further information. To begin with, we define questions resolved (or largely explained) by the linked web pages (i.e., in the corresponding answers) as navigational question, which are simulated as verbose queries to evaluate the performance of search engines (i.e., by considering the associated linked web pages as relevant documents). Then we experiment with the process of identifying new navigational questions from CQA, from which we demonstrate that navigational intent detection can be effectively automated by using textual features and a set of metadata features. Lastly, to effectively identify relevant navigational questions, we present a hybrid approach which blends several language modelling techniques, namely, the classic (query-likelihood) language model, the state-of-the-art translation-based language model, and our proposed intent-based language model. Our experiments on two real-world datasets show that the proposed mixture language model leads to a significant performance boost compared to that of the state-of-the-art language modelling approach.