One vs. Many QA Matching with both Word-level and Sentence-level Attention Network

Proceedings of the 27th International Conference on Computational Linguistics, pages 2540–2550 Santa Fe, New Mexico, USA, August 20-26, 2018. 2540 1 One vs. Many QA Matching with both Word-level and Sentence-level Attention Network Lu Wang1, 2, Shoushan Li1, 2, *, Changlong Sun3, Xiaozhong Liu3, Luo Si3, Min Zhang1,2, Guodong Zhou1 1NLP Lab, School of Computer Science and Technology, Soochow University, China 2Collaborative Innovation Center of Novel Software Technology and Industrialization 3Alibaba Group, China 1wanglu_1994@126.com, {lishoushan, minzhang, gdzhou}@suda.edu.cn 3{changlong.scl, xiaozhong.lxz, luo.si}@alibaba-inc.com Abstract Question-Answer (QA) matching is a fundamental task in the Natural Language Processing community. In this paper, we first build a novel QA matching corpus with informal text which is collected from a product reviewing website. Then, we propose a novel QA matching approach, namely One vs. Many Matching, which aims to address the novel scenario where one question sentence often has an answer with multiple sentences. Furthermore, we improve our matching approach by employing both word-level and sentence-level attentions for solving the noisy problem in the informal text. Empirical studies demonstrate the effectiveness of the proposed approach to question-answer matching. 1 Introduction Question answer (QA) matching is a task to determine whether an answer is answering a given ques- tion. For instance, in Figure 1, the question “Where is dear john filmed at?” in E1 has two candidate answers “The movie was filmed in 2009 in Charleston.” and “The file was released on May 25, 2010 on DVD.” The first answer is determined with the “Matching” category since it answers the question while the second answer is “Non-matching” since it could not answer the question. The past five years have witnessed a huge exploding interest in the research on QA matching, due to its widely applica- tions, such as question answering (Yang et al., 2015; Tan et al., 2016; Wang et al., 2017) and reading comprehension (Trischler et al., 2016; Dhingra et al., 2017). However, all existing QA matching studies only focus on formal text. In real applications, there exists many scenarios where the QA text is informal. For instance, E2 is a question-answer pair ext- *Corresponding author. This work is licensed under a Creative Commons Attribution 4.0 International License. License details: http://creativecom- mons.org/licenses/by/4.0/ Q2:Where is dear john filmed at? A2: The film was released on May 25, 2010 on DVD. Label: Non-matching E2: Three QA pairs in informal text Q: Will the response time slow after updating os? What about the battery? What about the screen? A: The response time is not a spark of slow, it has 4G RAM. But you must use it carefully, my phone's secreen has broke down. Q2: What about the battery? Q3: What about the screen? A3: The response ...... broke down. A2: The response ...... broke down. Label: Non-matching Label: Matching E1: Two QA pairs in formal text Q1: Where is dear john filmed at? A1: The movie was filmed in 2009 in Charleston. Label: Matching Q1: Will the response time slow after updating os? A1: The response ...... broke down. …

Content truncated to 500 words