HPL-2012-191An Efficient Passage Ranking Technique for Question Answering System
A, Pooja A; Krishnan, Vinodh; Manjunath, Geetha
Keyword(s): information retrieval; algorithms; ranking
Abstract: Question answering systems provide an intuitive way of requesting concise information from web or database (Information QA systems) or from community sites (Community QA systems). This concise answer is obtained by filtering and ranking the data available with respect to the question. Thus, one of the most important stages of any QA system is the passage ranking stage, which ranks the possible answers based on their relevance to the question. Much of the previous literature solved the passage ranking problem by considering the bag-of-words model that ignored the relationship among the words or structural model that ignored the semantic equivalence of words. In this paper, we propose a more comprehensive passage ranking technique that goes beyond these limitations and leverages lexical, semantic and syntactic features together to rank the possible answer snippets for a given factoid question. Apart from comparing the proposed technique against existing state-of-the-art techniques, we analyze the improvement achieved in the ranking with respect to different features, thus, giving insights on the role of each feature. The experimental results on TREC QA datasets show that the proposed technique (MRR 0.62) gives a significant 26.5% improvement in MRR over existing state-of-the- art (MRR 0.49), thus, highlighting the importance of the proposed technique.
External Posting Date: September 6, 2012 [Abstract Only]. Approved for External Publication - External Copyright Consideration
Internal Posting Date: September 6, 2012 [Fulltext]