Generating Natural Answers by Incorporating Copying and Retrieving Mechanisms in Sequence-to-Sequence Learning
notes
基于知识库的问答系统。在传统seq2seq with attention中加入KB信息。每次生成一个词的时候，有三个可能的来源，一个是从词典中生成，一个是从原问题中copy，还有一个是从KB中生成。首先通过topic匹配从KB库中检索出一些候选facts。然后再模型中计算问题和每个facts的匹配度，作为对于fact的attention。facts的编码是将三元组的word embedding拼起来作为这条fact的表示。当前时刻输出由问题的attention sum，KB的attention sum，decoder的隐状态和上一时刻输出共同决定。
bibtex
@inproceedings{he2017generating,
title={Generating natural answers by incorporating copying and retrieving mechanisms in sequence-to-sequence learning},
author={He, Shizhu and Liu, Cao and Liu, Kang and Zhao, Jun},
booktitle={Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)},
volume={1},
pages={199--208},
year={2017}
}
title
Generating Natural Answers by Incorporating Copying and Retrieving Mechanisms in Sequence-to-Sequence Learning
notes
基于知识库的问答系统。在传统seq2seq with attention中加入KB信息。每次生成一个词的时候，有三个可能的来源，一个是从词典中生成，一个是从原问题中copy，还有一个是从KB中生成。首先通过topic匹配从KB库中检索出一些候选facts。然后再模型中计算问题和每个facts的匹配度，作为对于fact的attention。facts的编码是将三元组的word embedding拼起来作为这条fact的表示。当前时刻输出由问题的attention sum，KB的attention sum，decoder的隐状态和上一时刻输出共同决定。
bibtex
@inproceedings{he2017generating, title={Generating natural answers by incorporating copying and retrieving mechanisms in sequence-to-sequence learning}, author={He, Shizhu and Liu, Cao and Liu, Kang and Zhao, Jun}, booktitle={Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)}, volume={1}, pages={199--208}, year={2017} }
link
http://www.aclweb.org/anthology/P17-1019
publication
ACL 2017 long accepted
open source
http://www.nlpr.ia.ac.cn/cip/shizhuhe/data/coreqa_data.zip
affiliated
National Laboratory of Pattern Recognition, Institute of Automation, Chinese Academy of Sciences Kang Liu