An End-to-End Model for Question Answering over Knowledge Base with Cross-Attention Combining Global Knowledge

Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics, pages 221–231 Vancouver, Canada, July 30 - August 4, 2017. c⃝2017 Association for Computational Linguistics https://doi.org/10.18653/v1/P17-1021 Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics, pages 221–231 Vancouver, Canada, July 30 - August 4, 2017. c⃝2017 Association for Computational Linguistics https://doi.org/10.18653/v1/P17-1021 An End-to-End Model for Question Answering over Knowledge Base with Cross-Attention Combining Global Knowledge Yanchao Hao1,2, Yuanzhe Zhang1,3, Kang Liu1, Shizhu He1, Zhanyi Liu3, Hua Wu3 and Jun Zhao1,2 1 National Laboratory of Pattern Recognition, Institute of Automation, Chinese Academy of Sciences, Beijing, 100190, China 2 University of Chinese Academy of Sciences, Beijing, 100049, China 3 Baidu Inc., Beijing, 100085, China {yanchao.hao, yzzhang, kliu, shizhu.he, jzhao}@nlpr.ia.ac.cn {liuzhanyi, wu hua}@baidu.com Abstract With the rapid growth of knowledge bases (KBs) on the web, how to take full advan- tage of them becomes increasingly impor- tant. Question answering over knowledge base (KB-QA) is one of the promising ap- proaches to access the substantial knowl- edge. Meanwhile, as the neural network- based (NN-based) methods develop, NN- based KB-QA has already achieved im- pressive results. However, previous work did not put more emphasis on question representation, and the question is con- verted into a fixed vector regardless of its candidate answers. This simple represen- tation strategy is not easy to express the proper information in the question. Hence, we present an end-to-end neural network model to represent the questions and their corresponding scores dynamically accord- ing to the various candidate answer aspect- s via cross-attention mechanism. In ad- dition, we leverage the global knowledge inside the underlying KB, aiming at inte- grating the rich KB information into the representation of the answers. As a result, it could alleviates the out-of-vocabulary (OOV) problem, which helps the cross- attention model to represent the question more precisely. The experimental result- s on WebQuestions demonstrate the effec- tiveness of the proposed approach. 1 Introduction As the amount of the knowledge bases (KBs) grows, people are paying more attention to seeking effective methods for accessing these precious in- tellectual resources. There are several tailor-made languages designed for querying KBs, such as SPARQL (Prudhommeaux and Seaborne, 2008). However, to handle such query languages, users are required to not only be familiar with the partic- ular language grammars, but also be aware of the architectures of the KBs. By contrast, knowledge base-based question answering (KB-QA) (Unger et al., 2014), which takes natural language as query language, is a more user-friendly solution, and has become a research focus in recent years. Given natural language questions, the goal of KB-QA is to automatically return answers from the KB. There are two mainstream research direc- tions for this task: semantic parsing-based (SP- based) (Zettlemoyer and Collins, 2009, 2012; K- wiatkowski et al., 2013; Cai and Yates, 2013; Be- rant et al., 2013; Yih et al., 2015, 2016; Red- dy et al., 2016) and information retrieval-based (IR-based) (Yao and Van Durme, 2014; Bordes et al., 2014a,b, 2015; Dong et al., …

Content truncated to 500 words