LCQMC:A Large-scale Chinese Question Matching Corpus

Proceedings of the 27th International Conference on Computational Linguistics, pages 1952–1962 Santa Fe, New Mexico, USA, August 20-26, 2018. 1952 LCQMC: A Large-scale Chinese Question Matching Corpus Xin Liu†, Qingcai Chen†∗, Chong Deng‡, Huajun Zeng#, Jing Chen†, Dongfang Li†, Buzhou Tang† †Shenzhen Calligraphy Digital Simulation Technology Lab, Harbin Institute of Technology, China ‡Machine Intelligence Technology, Alibaba Group †{hit.liuxin, qingcai.chen, mcdh.chenjing, crazyofapple, tangbuzhou}@gmail.com, ‡dengchong.d@alibaba-inc.com, #aaahchi@hotmail.com Abstract The lack of large-scale question matching corpora greatly limits the development of matching methods in question answering (QA) system, especially for non-English languages. To amelio- rate this situation, in this paper, we introduce a large-scale Chinese question matching corpus (named LCQMC), which is released to the public1. LCQMC is more general than paraphrase corpus as it focuses on intent matching rather than paraphrase. How to collect a large number of question pairs in variant linguistic forms, which may present the same intent, is the key point for such corpus construction. In this paper, we first use a search engine to collect large-scale question pairs related to high-frequency words from various domains, then filter irrelevant pairs by the Wasserstein distance, and finally recruit three annotators to manually check the left pairs. After this process, a question matching corpus that contains 260,068 question pairs is construct- ed. In order to verify the LCQMC corpus, we split it into three parts, i.e., a training set containing 238,766 question pairs, a development set with 8,802 question pairs, and a test set with 12,500 question pairs, and test several well-known sentence matching methods on it. The experimen- tal results not only demonstrate the good quality of LCQMC, but also provide solid baseline performance for further researches on this corpus. 1 Introduction Question matching is a fundamental task of QA, which is usually recognized as a semantic matching task, sometimes a paraphrase identification task. The goal of the task is to search questions that have similar intent as the input question from an existing database. QA system learns from a large and noisy question matching corpus, and can span a diverse set of topics (Fader et al., 2013). Semantic match- ing is often regarded as a binary classification problem. Given a pair of sentences, the system is asked to judge whether two sentences express the same meaning. Semantic matching algorithms are widely used in NLP applications, such as information retrieval, machine translation and knowledge-based ques- tion answering. Emerging research also shows that first story detection (Petrovic et al., 2012) and text normalization (Xu et al., 2013; Ling et al., 2013) can also benefit from semantic matching techniques. Chinese is one of the most widely used languages in the world, and the proportion of Chinese using Internet is increasing rapidly (Qiu et al., 2013). Just the same as the English sentence matching task, many Chinese applications can also benefit from Chinese sentence matching. Though some English paraphrase corpora were published for sentence-level matching tasks, the scales are still not enough to meet the data requirements of deep learning techniques. The problem is …

Content truncated to 500 words