Hierarchical Graph Network for Multi-hop Question Answering

Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing, pages 8823–8838, November 16–20, 2020. c⃝2020 Association for Computational Linguistics 8823 Hierarchical Graph Network for Multi-hop Question Answering Yuwei Fang, Siqi Sun, Zhe Gan, Rohit Pillai, Shuohang Wang, Jingjing Liu Microsoft Dynamics 365 AI Research {yuwfan,siqi.sun,zhe.gan,rohit.pillai,shuohang.wang,jingjl}@microsoft.com Abstract In this paper, we present Hierarchical Graph Network (HGN) for multi-hop question an- swering. To aggregate clues from scattered texts across multiple paragraphs, a hierarchi- cal graph is created by constructing nodes on different levels of granularity (questions, paragraphs, sentences, entities), the representa- tions of which are initialized with pre-trained contextual encoders. Given this hierarchical graph, the initial node representations are up- dated through graph propagation, and multi- hop reasoning is performed via traversing through the graph edges for each subsequent sub-task (e.g., paragraph selection, supporting facts extraction, answer prediction). By weav- ing heterogeneous nodes into an integral uni- fied graph, this hierarchical differentiation of node granularity enables HGN to support dif- ferent question answering sub-tasks simultane- ously. Experiments on the HotpotQA bench- mark demonstrate that the proposed model achieves new state of the art, outperforming ex- isting multi-hop QA approaches.1 1 Introduction In contrast to one-hop question answering (Ra- jpurkar et al., 2016; Trischler et al., 2016; Lai et al., 2017) where answers can be derived from a single paragraph (Wang and Jiang, 2017; Seo et al., 2017; Liu et al., 2018; Devlin et al., 2019), many recent studies on question answering focus on multi-hop reasoning across multiple documents or paragraphs. Popular tasks include WikiHop (Welbl et al., 2018), ComplexWebQuestions (Talmor and Berant, 2018), and HotpotQA (Yang et al., 2018). An example from HotpotQA is illustrated in Fig- ure 1. In order to correctly answer the question (“The director of the romantic comedy ‘Big Stone Gap’ is based in what New York city”), the model is 1Code will be released at https://github.com/yuwfan/HGN. Figure 1: An example of multi-hop question answer- ing from HotpotQA. The model needs to identify rele- vant paragraphs, determine supporting facts, and then predict the answer correctly. required to first identify P1 as a relevant paragraph, whose title contains the keywords that appear in the question (“Big Stone Gap”). S1, the first sentence of P1, is then chosen by the model as a supporting fact that leads to the next-hop paragraph P2. Lastly, from P2, the span “Greenwich Village, New York City” is selected as the predicted answer. Most existing studies use a retriever to find para- graphs that contain the right answer to the question (P1 and P2 in this case). A Machine Reading Com- prehension (MRC) model is then applied to the selected paragraphs for answer prediction (Nishida et al., 2019; Min et al., 2019b). However, even after successfully identifying a reasoning chain through multiple paragraphs, it still remains a critical chal- lenge how to aggregate evidence from scattered 8824 sources on different granularity levels (e.g., para- graphs, sentences, entities) for joint answer and supporting facts prediction. To better leverage fine-grained evidences, some studies apply …

Content truncated to 500 words