Dynamically Fused Graph Network for Multi-hop Reasoning

Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 6140–6150 Florence, Italy, July 28 - August 2, 2019. c⃝2019 Association for Computational Linguistics 6140 Dynamically Fused Graph Network for Multi-hop Reasoning Lin Qiu1† Yunxuan Xiao1† Yanru Qu1† Hao Zhou2 Lei Li2 Weinan Zhang1 Yong Yu1 1 Shanghai Jiao Tong University 2 ByteDance AI Lab, China {lqiu, kevinqu, yyu}@apex.sjtu.edu.cn {xiaoyunxuan, wnzhang}@sjtu.edu.cn {zhouhao.nlp, lilei.02}@bytedance.com Abstract Text-based question answering (TBQA) has been studied extensively in recent years. Most existing approaches focus on finding the an- swer to a question within a single paragraph. However, many difficult questions require multiple supporting evidence from scattered text across two or more documents. In this pa- per, we propose the Dynamically Fused Graph Network (DFGN), a novel method to answer those questions requiring multiple scattered evidence and reasoning over them. Inspired by human’s step-by-step reasoning behavior, DFGN includes a dynamic fusion layer that starts from the entities mentioned in the given query, explores along the entity graph dynam- ically built from the text, and gradually finds relevant supporting entities from the given documents. We evaluate DFGN on HotpotQA, a public TBQA dataset requiring multi-hop reasoning. DFGN achieves competitive results on the public board. Furthermore, our analy- sis shows DFGN could produce interpretable reasoning chains. 1 Introduction Question answering (QA) has been a popular topic in natural language processing. QA pro- vides a quantifiable way to evaluate an NLP sys- tem’s capability on language understanding and reasoning (Hermann et al., 2015; Rajpurkar et al., 2016, 2018). Most previous work focus on find- ing evidence and answers from a single para- graph (Seo et al., 2016; Liu et al., 2017; Wang et al., 2017). It rarely tests deep reasoning ca- pabilities of the underlying model. In fact, Min et al. (2018) observe that most questions in exist- ing QA benchmarks can be answered by retrieving †These authors contributed equally. The order of author- ship is decided through dice rolling. Work done while Lin Qiu was a research intern in ByteDance AI Lab. The Sum of All Fears is a best-selling thriller novel by Tom Clancy ... It was the fourth of Clancy's Jack Ryan books to be turned into a film ... Dr. John Patrick Jack Ryan Sr., KCVO (Hon.), Ph.D. is a fictional character created by Tom Clancy who appears in many of his novels and their respective film adaptations ... Net Force Explorers is a series of young adult novels created by Tom Clancy and Steve Pieczenik as a spin-off of the military fiction series ... Question: What fiction character created by Tom Clancy was turned into a film in 2002? Answer: Jack Ryan Input�Paragraphs: Original�Entity�Graph Second�Mask�Applied First�Mask�Applied Figure 1: Example of multi-hop text-based QA. One question and three document paragraphs are given. Our proposed DFGN conducts multi-step reasoning over the facts by constructing an entity graph from multiple paragraphs, predicting a dynamic mask to select a sub- graph, propagating information along the graph, and finally transfer the information from the graph back …

Content truncated to 500 words