Acta Scientiarum Naturalium Universitatis Pekinensis ›› 2022, Vol. 58 ›› Issue (1): 13-20.DOI: 10.13209/j.0479-8023.2021.107

Previous Articles     Next Articles

Research on Dialogue Entity Relation Extraction with Enhancing Character Information

XU Yang1, JIANG Yuru1,2,†, ZHANG Yuyao1, HE Weikai1   

  1. 1. Institute of Intelligent Information Processing, Beijing Information Science and Technology University, Beijing 100101 2. Beijing Laboratory of National Economic Security Early-Warning Engineering, Beijing 100044
  • Received:2021-06-08 Revised:2021-08-15 Online:2022-01-20 Published:2022-01-20
  • Contact: JIANG Yuru, E-mail: jiangyuru(at)bistu.edu.cn

融合角色指代的多方对话关系抽取方法研究

徐洋1, 蒋玉茹1,2,†, 张禹尧1, 何威恺1   

  1. 1. 北京信息科技大学智能信息处理研究所, 北京 100101 2. 国家经济安全预警工程北京实验室, 北京 100044
  • 通讯作者: 蒋玉茹, E-mail: jiangyuru(at)bistu.edu.cn
  • 基金资助:
    国家自然科学基金(61602044, 61772081)和北京市自然科学基金(4204100)资助

PDF

Abstract:

A multi-party dialogue relationship extraction model that integrates character reference information is proposed based on a previous model of graph attention network. Specifically, this article adds character nodes in graph, connects them with the word nodes referred by the corresponding character and uses graph attention networks for encoding. The F1 score based on DialogRE Dataset improved by 2.9% on the valid set and 4.6% on the test set compared with the baseline model.

Key words: entity relation extraction, hierarchical encoding, graph attention network, dialogue structure

摘要：

在前期基于图网络的模型基础上, 引入角色指代信息, 提出融合角色指代的多方对话关系抽取模型。在构建图节点时加入角色节点, 将其与对应角色指代的词节点进行连接, 并使用图注意力网络进行编码。在DialogRE数据集上的实验效果与基线模型相比, F1值在验证集上提升2.9%, 在测试集上提升4.6%。

关键词: 实体关系抽取, 层次化编码, 图注意力网络, 对话结构

Copyright © Acta Scientiarum Naturalium Universitatis Pekinensis, All Rights Reserved.
Powered by Beijing Magtech Co., Ltd.