基于门控思想的改进交互关系提取的多跳阅读理解研究
Research on Multi-Hop Reading Comprehension Based on Improved Interaction Extraction with Gate Mechanism
DOI: 10.12677/ORF.2022.122016, PDF,   
作者: 何奏捷：贵州大学数学与统计学院，贵州 贵阳；杜逆索, 欧阳智：贵州大学数学与统计学院，贵州 贵阳；贵州大学贵州省大数据产业发展应用研究院，贵州 贵阳
关键词: 多跳阅读理解；门控机制；注意力机制；Multi-Hop Reading Comprehension； Gate Mechanism； Attention Mechanism
摘要: 多跳阅读理解需要搜集多个支持文档中的各个证据，然后利用搜集到的证据，进行多级跳跃的推理过程确认答案。目前图神经网络被大量运用在解决多跳阅读理解问题，针对目前图神经网络相关模型中问题与节点的交互信息提取不充分的问题，提出基于门控思想的改进交互关系提取的多跳阅读理解模型。首先，将支持文档中与候选答案或问题中实体完全一致的词作为实体图节点，将同一段落中的不同实体相连，不同段落中的同一实体相连成边构建实体图。然后，对提取的实体进行信息编码处理，并通过图卷积网络模拟推理过程。最后，利用改进的交互关系提取模型，将推理过后的数据与原始节点进行信息对比与聚合，保留更有效的交互信息进行结果预测。在WikiHop数据集中进行实验验证，结果表明改进交互关系提取方法取得了更好的效果。
Abstract: Multi-hop reading comprehension needs leaping reasoning over multiple supporting documents to obtain the correct answer. Graph neural network is widely used to solve the problem of multi hop reading comprehension. Aiming at the insufficient extraction of interactive information between problems and nodes in the current graph neural network related models, the multi hop reading comprehension model based on improved interactive relationship extraction based on gating mechanism is proposed. Firstly, take the entities in the supporting document that are equal to the entities in the candidate set or question as the entity graph nodes, connect different entities in the same paragraph and the same entity in different paragraphs as edges to construct the entity graph. Then, embedding the graph nodes, and using graph convolution network to reason. Finally, the improved interactive relationship extraction model is used to compare and aggregate the information with the original graph nodes, and retain more effective interactive information for result prediction. The WikiHop experimental results show that the improved interactive relationship extraction method has achieved better results.
文章引用：何奏捷, 杜逆索, 欧阳智. 基于门控思想的改进交互关系提取的多跳阅读理解研究[J]. 运筹与模糊学, 2022, 12(2): 169-176. https://doi.org/10.12677/ORF.2022.122016

参考文献

[1] Devlin, J., Chang, M.W., Lee, K., et al. (2019) Bert: Pre-Training of Deep Bidirectional Transformers for Language Understanding. Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (NAACL-HLT), Volume 1, 4171-4186.
[2] Seo, M., Kembhavi, A., Farhadi, A., et al. (2017) Bidirectional Attention Flow for Machine Comprehension. Proceedings of the 5th International Conference on Learning Representations, Toulon, 24-26 April 2017.
[3] 万静, 郭雅志. 基于多段落排序的机器阅读理解研究 [J]. 北京化工大学学报(自然科学版), 2019, 46(3): 93-98.
[4] Wang, S., Yu, M., Jiang, J., et al. (2018) A Co-Matching Model for Multichoice Reading Comprehension. In: Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (ACL), Association for Computational Linguistics, Melbourne, 746-751. [Google Scholar] [CrossRef] 
[5] Dhingra, B., Jin, Q., Yang, Z.L., et al. (2018) Neural Models for Reasoning over Multiple Mentions Using Coreference. Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies: NAACL-HLT, Volume 2, 42-48. [Google Scholar] [CrossRef] 
[6] Feldman, Y. and El-Yaniv, R. (2019) Multi-Hop Paragraph Retrieval for Open Domain Question Answering. In: Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics (ACL), Association for Computational Linguistics, Florence, 2296-2309. [Google Scholar] [CrossRef] 
[7] Jiang, Y., Joshi, N., Chen, Y., et al. (2019) Explore, Propose, and As-semble: An Interpretable Model for Multi-Hop Reading Comprehension. In: Proceedings of the 57th Conference of the Association for Computational Linguistics (ACL), Association for Computational Linguistics, Florence, 2714-2725. [Google Scholar] [CrossRef] 
[8] Gori, M., Monfardini, G. and Scarselli, F. (2005) A New Model for Learning in Graph Domains. Proceedings 2005 IEEE International Joint Conference on Neural Networks, Volume 2, 729-734.
[9] Santoro, A., Raposo, D., Barrett, D.G., et al. (2017) A Simple Neural Network Module for Relational Reasoning. Annual Conference on Neural Information Processing Systems 2017, Long Beach, 4-9 December 2017, 4967-4976.
[10] Ding, M., Zhou, C., Chen, Q.B., et al. (2019) Cognitive Graph for Multi-Hop Reading Comprehension at Scale. In: Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics (ACL), Associa-tion for Computational Linguistics, Florence, 2694-2703. [Google Scholar] [CrossRef] 
[11] Ye, D., Lin, Y., Liu, Z., et al. (2019) Multi-Paragraph Reasoning with Knowledge Enhanced Graph Neural Network.
[12] Cao, D.N., Aziz, W. and Titov, I. (2019) Question Answering by Reasoning across Documents with Graph Convolutional Networks. Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (NAACL-HLT), Volume 1, 2306-2317.
[13] Cao, Y., Fang, M. and Tao, D. (2019) Bag: Bi Directional Attention Entity Graph Convolutional Network for Multi-Hop Reasoning Question Answering. Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1, 357-362.
[14] Song, L., Wang, Z., Yu, M., et al. (2018) Exploring Graph-Structured Passage Representation for Multi-Hop Reading Comprehension with Graph Neural Net-works.
[15] Tu, M., Huang, K., Wang, G., et al. (2020) Select, Answer and Explain: Interpretable Multi-Hop Reading Comprehension over Multiple Documents. In: The Thirty-Fourth AAAI Conference on Artificial Intelligence (AAAI), AAAI Press, New York, 9073-9080. [Google Scholar] [CrossRef] 
[16] Tang, Z., Shen, Y., Ma, X., et al. (2020) Multi-Hop Reading Comprehension across Documents with Path-Based Graph Convolutional Network. Proceedings of the Twenty-Ninth International Joint Conference on Artificial Intelligence (IJCAI), Yokohama, 11-17 July 2020, 3905-3911. [Google Scholar] [CrossRef] 
[17] Zhong, V., Xiong, C., Keskar, N.S., et al. (2019) Coarse-Grain Fine-Grain Coat Tention Network for Multi-Evidence Question Answering. 7th International Conference on Learning Representations (ICLR), New Orleans, 6-9 May 2019.

为你推荐