Query Graph Generation for Answering Multi-hop Complex Questions from Knowledge Bases

Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 969–974 July 5 - 10, 2020. c⃝2020 Association for Computational Linguistics 969 Query Graph Generation for Answering Multi-hop Complex Questions from Knowledge Bases Yunshi Lan Singapore Management University yslan.2015@phdcs.smu.edu.sg Jing Jiang Singapore Management University jingjiang@smu.edu.sg Abstract Previous work on answering complex ques- tions from knowledge bases usually separately addresses two types of complexity: questions with constraints and questions with multiple hops of relations. In this paper, we handle both types of complexity at the same time. Motivated by the observation that early incor- poration of constraints into query graphs can more effectively prune the search space, we propose a modified staged query graph gener- ation method with more flexible ways to gen- erate query graphs. Our experiments clearly show that our method achieves the state of the art on three benchmark KBQA datasets. 1 Introduction Knowledge base question answering (KBQA) aims at answering factoid questions from a knowledge base (KB). It has attracted much attention in recent years (Bordes et al., 2014; Xu et al., 2016; Yu et al., 2017; Liang et al., 2017; Hu et al., 2018; Petrochuk and Zettlemoyer, 2018). Early work on KBQA focused on simple questions containing a single relation (Yih et al., 2014; Bordes et al., 2015; Dong et al., 2015; Hao et al., 2017). However, real ques- tions are often more complex and recently some studies looked into complex KBQA. Two different types of complexity have been studied: (1) Single-relation questions with con- straints. For example, in the question “Who was the first president of the U.S.?” there is a single relation “president of” between the answer entity and the entity “U.S.,” but we also have the con- straint “first” that needs to be satisfied. For this type of complex questions, a staged query graph generation method has been proposed, which first identifies a single-hop relation path and then adds constraints to it to form a query graph (Yih et al., 2015; Bao et al., 2016; Luo et al., 2018). (2) Ques- tions with multiple hops of relations. For example, for the question “Who is the wife of the founder of Facebook?” the answer is related to “Facebook” through two hops of relations, namely, “wife of” and “founder of.” To answer this type of multi- hop questions, we need to consider longer relation paths in order to reach the correct answers. The main challenge here is how to restrict the search space, i.e., to reduce the number of multi-hop re- lation paths to be considered, because the search space grows exponentially with the length of rela- tion paths. One idea is to use beam search. For example, Chen et al. (2019) and Lan et al. (2019b) proposed to consider only the best matching re- lation instead of all relations when extending a relation path. However, little work has been done to deal with both types of complexity together. In this paper, we handle both constraints and multi-hop relations together for complex …

Content truncated to 500 words