Why Question Answering using Sentiment Analysis and Word Classes

Proceedings of the 2012 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning, pages 368–378, Jeju Island, Korea, 12–14 July 2012. c⃝2012 Association for Computational Linguistics Why Question Answering using Sentiment Analysis and Word Classes Jong-Hoon Oh∗ Kentaro Torisawa† Chikara Hashimoto ‡ Takuya Kawada§ Stijn De Saeger¶ Jun’ichi Kazama∥ Yiou Wang∗∗ Information Analysis Laboratory Universal Communication Research Institute National Institute of Information and Communications Technology (NICT) {∗ rovellia,† torisawa,‡ ch,§ tkawada,¶stijn,∥ kazama,∗∗wangyiou}@nict.go.jp Abstract In this paper we explore the utility of sen- timent analysis and semantic word classes for improving why-question answering on a large-scale web corpus. Our work is moti- vated by the observation that a why-question and its answer often follow the pattern that if something undesirable happens, the reason is also often something undesirable, and if some- thing desirable happens, the reason is also of- ten something desirable. To the best of our knowledge, this is the first work that intro- duces sentiment analysis to non-factoid ques- tion answering. We combine this simple idea with semantic word classes for ranking an- swers to why-questions and show that on a set of 850 why-questions our method gains 15.2% improvement in precision at the top-1 answer over a baseline state-of-the-art QA system that achieved the best performance in a shared task of Japanese non-factoid QA in NTCIR-6. 1 Introduction Question Answering (QA) research for factoid ques- tions has recently achieved great success as demon- strated by IBM’s Watson at Jeopardy: its accuracy has been reported to be around 85% on factoid ques- tions (Ferrucci et al., 2010). Although recent shared QA tasks (Voorhees, 2004; Peñas et al., 2011; Fuku- moto et al., 2007) have stimulated the research com- munity to move beyond factoid QA, comparatively little attention has been paid to QA for non-factoid questions such as why questions and how to ques- tions, and the performance of the state-of-art non- factoid QA systems reported in the literature (Mu- rata et al., 2007; Surdeanu et al., 2011; Verberne et al., 2010) remains considerably lower than that of factoid QA (i.e., 34% in MRR at top-150 results on why-questions (Verberne et al., 2010)). In this paper we explore the utility of sentiment analysis (Pang et al., 2002; Turney, 2002; Nakagawa et al., 2010) and semantic word classes for improv- ing why-question answering (why-QA) on a large- scale web corpus. The inspiration behind this work is the observation that why-questions and their an- swers often have the following tendency: • if something undesirable happens, the reason is often also something undesirable, and • if something desirable happens, its reason is of- ten also desirable. Consider the following question Q1, and its an- swer candidates A1-1 and A1-2. • Q1: Why does cancer occur? • A1-1: Carcinogens such as nitrosamine and benzopyrene may increase the risk of cancer by altering DNA in cells. • A1-2: Maintaining a healthy weight may lower the risk of various types of cancer. Here A1-1 describes an undesirable event related to cancer, …

Content truncated to 500 words