CoCQA: Co-Training over Questions and Answers with an Application to Predicting Question Subjectivity Orientation

Proceedings of the 2008 Conference on Empirical Methods in Natural Language Processing, pages 937–946, Honolulu, October 2008. c⃝2008 Association for Computational Linguistics CoCQA: Co-Training Over Questions and Answers with an Application to Predicting Question Subjectivity Orientation Baoli Li Emory University csblli@gmail.com Yandong Liu Emory University yliu49@emory.edu Eugene Agichtein Emory University eugene@mathcs.emory.edu Abstract An increasingly popular method for finding information online is via the Community Question Answering (CQA) portals such as Yahoo! An- swers, Naver, and Baidu Knows. Searching the CQA archives, and rank- ing, filtering, and evaluating the sub- mitted answers requires intelligent processing of the questions and an- swers posed by the users. One impor- tant task is automatically detecting the question’s subjectivity orientation: namely, whether a user is searching for subjective or objective information. Unfortunately, real user questions are often vague, ill-posed, poorly stated. Furthermore, there has been little la- beled training data available for real user questions. To address these prob- lems, we present CoCQA, a co-training system that exploits the association be- tween the questions and contributed answers for question analysis tasks. The co-training approach allows CoCQA to use the effectively unlim- ited amounts of unlabeled data readily available in CQA archives. In this pa- per we study the effectiveness of CoCQA for the question subjectivity classification task by experimenting over thousands of real users’ questions. 1 Introduction Automatic question answering (QA) has been one of the long-standing goals of natural lan- guage processing, information retrieval, and artificial intelligence research. For a natural language question we would like to respond with a specific, accurate, and complete an- swer that addresses the question. Although much progress has been made, answering complex, opinion, and even many factual questions automatically is still beyond the current state-of-the-art. At the same time, the rise of popularity in social media and collabo- rative content creation services provides a promising alternative to web search or com- pletely automated QA. The explicit support for social interactions between participants, such as posting comments, rating content, and responding to questions and comments makes this medium particularly amenable to Ques- tion Answering. Some very successful exam- ples of Community Question Answering (CQA) sites are Yahoo! Answers 1 and Naver2, and Baidu Knows3. Yahoo! Answers alone has already amassed hundreds of mil- lions of answers posted by millions of par- ticipants on thousands of topics. The questions posted to such CQA portals are typically complex, subjective, and rely on human interpretation to understand the corre- sponding information need. At the same time, the questions are also usually ill-phrased, vague, and often subjective in nature. Hence, analysis of the questions (and of the corre- sponding user intent) in this setting is a par- ticularly difficult task. At the same time, CQA content incorporates the relationships between questions and the corresponding an- swers. Because of the various incentives pro- vided by the CQA sites, answers posted by users tend to be, at least to some degree, re- sponsive to the question. This observation suggests investigating whether the relation- 1 http://answers.yahoo.com …

Content truncated to 500 words