Answering Opinion Questions on Products by Exploiting Hierarchical Organization of Consumer Reviews

Proceedings of the 2012 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning, pages 391–401, Jeju Island, Korea, 12–14 July 2012. c⃝2012 Association for Computational Linguistics Answering Opinion Questions on Products by Exploiting Hierarchical Organization of Consumer Reviews Jianxing Yu, Zheng-Jun Zha, Tat-Seng Chua School of Computing National University of Singapore {jianxing, zhazj, chuats}@comp.nus.edu.sg Abstract This paper proposes to generate appropriate answers for opinion questions about prod- ucts by exploiting the hierarchical organiza- tion of consumer reviews. The hierarchy orga- nizes product aspects as nodes following their parent-child relations. For each aspect, the re- views and corresponding opinions on this as- pect are stored. We develop a new framework for opinion Questions Answering, which en- ables accurate question analysis and effective answer generation by making use the hierar- chy. In particular, we first identify the (ex- plicit/implicit) product aspects asked in the questions and their sub-aspects by referring to the hierarchy. We then retrieve the corre- sponding review fragments relevant to the as- pects from the hierarchy. In order to gener- ate appropriate answers from the review frag- ments, we develop a multi-criteria optimiza- tion approach for answer generation by simul- taneously taking into account review salience, coherence, diversity, and parent-child rela- tions among the aspects. We conduct eval- uations on 11 popular products in four do- mains. The evaluated corpus contains 70,359 consumer reviews and 220 questions on these products. Experimental results demonstrate the effectiveness of our approach. 1 Introduction With the rapid development of E-commerce, most retail websites encourage consumers to post reviews to express their opinions on the products. For exam- ple, the review “The battery of Nokia N95 is amaz- ing.” reveals positive opinion on the aspect “bat- Figure 1: Overview of product opinion-QA framework tery” of product Nokia N95. An aspect here refers to a component or an attribute of a certain prod- uct. Numerous consumer reviews are now available online, and these reviews contain rich opinionated information on various aspects of products. They are naturally a valuable resource for answering opin- ion questions about products, such as “How do peo- ple think about the battery of Nokia N95?” Opin- ion Question Answering (opinion-QA) on products seeks to uncover consumers’ thinking and feeling about the products or aspects of products. It is dif- ferent from traditional factual QA, where the ques- tions ask for the fact, such as “Where is the capital of United States?” and the answer is “Washington, D.C.” For a product opinionated question, the answer should not be just a best answer. It should reflect the opinions of various segments of users, and incorpo- 391 rate both positive and negative viewpoints. Hence the answer should be a summarization of public opinions and comments on the product or specific aspect asked in the question (Jiang et al., 2010). In addition, it should also include public opinions and comments on the sub-aspects. Such answers would help users to understand the inherent reasons of the opinions on the asked aspect. …

Content truncated to 500 words