@inproceedings{wang2018multi,
title={Multi-Passage Machine Reading Comprehension with Cross-Passage Answer Verification},
author={Yizhong Wang and Kai Liu and Jing Liu and Wei He and Yajuan Lyu and Hua Wu and Sujian Li and Haifeng Wang},
booktitle={Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)},
volume={1},
year={2018}
}
title
Multi-Passage Machine Reading Comprehension with Cross-Passage Answer Verification
notes
刷MS MARCO。相比于SQUAD，MS数据集是阅读多篇文档，然后根据给出的query得到span类型的答案。本文提出首先从每篇文章里面选择出一个答案，然后使用这个答案去别的文章里面验证，基于的假设是正确的答案是经得起大部分文章验证的。首先从每篇文章里面抽取答案，然后建模抽取出的答案的语义。将这个答案和其他答案融合以鉴别这个答案的正确性。首先做QP matching。然后通过两步的decoder，pointer network。预测span起始的概率向量和span结束的概率向量。用这两个概率向量对原文加权求和，得到答案的语义表示。再用这个答案和其他答案去做attention，然后sum，然后通过全连接得到这个答案的正确概率。
bibtex
@inproceedings{wang2018multi, title={Multi-Passage Machine Reading Comprehension with Cross-Passage Answer Verification}, author={Yizhong Wang and Kai Liu and Jing Liu and Wei He and Yajuan Lyu and Hua Wu and Sujian Li and Haifeng Wang}, booktitle={Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)}, volume={1}, year={2018} }
link
https://arxiv.org/pdf/1805.02220
publication
ACL 2018 long accepted
open source
No
affiliated
Peking University Baidu Inc.