\modelname for Multi-evidence Question Answering
Victor Zhong Paul G. Allen School of Computer Science、Engineering, University of Washington, Seattle, WA Caiming Xiong Salesforce Research, Palo Alto, CA Nitish Shirish Keskar Salesforce Research, Palo Alto, CA Richard Socher Salesforce Research, Palo Alto, CA
Abstract
End-to-end neural models have made significant progress in question answering, however recent studies show that these models implicitly assume that the answer and evidence appear close together in a single document. In this work, we propose the \modelname (CFC), a new question answering model that combines information from evidence across multiple documents. The CFC consists of a coarse-grain module that interprets documents with respect to the query then finds a relevant answer, and a fine-grain module which scores each candidate answer by comparing its occurrences across all of the documents with the query. We design these modules using hierarchies of coattention and self-attention, which learn to emphasize different parts of the input. On the Qangaroo WikiHop multi-evidence question answering task, the CFC obtains a new state-of-the-art result of 70.6% on the blind test set, outperforming the previous best by 3% accuracy despite not using pretrained contextual encoders.
中文速览
为了让机器能够跨多篇文档整合证据来回答问题，研究者提出了"粗细粒度融合"模型（Coarse-to-Fine Comprehension，CFC）：粗粒度模块先把所有支持文档与问题做联合注意力编码、生成整体摘要，再与候选答案比对；细粒度模块则专门定位每个候选答案在文档中的所有出现位置，逐一与问题匹配打分；两个模块均采用协同注意力（coattention）与自注意力（self-attention）的分层结构，最终将两路得分相加选出答案。在专为多文档推理设计的WikiHop测试集上，CFC以70.6%的准确率刷新了最优记录，在未使用预训练上下文编码器的前提下仍比此前最好结果高出3个百分点；在TriviaQA任务中，用CFC对候选答案重排序也带来了3%以上的提升。这项工作表明，将"宏观文档摘要"与"候选答案精确定位"分而治之，是让模型在海量长文档中高效聚合跨篇章证据的有效途径。
原文 arXiv:1901.00603;中英对照 + 大白话阅读 https://aha.fim.ai/paper/1901.00603v2