电子学报 ›› 2020, Vol. 48 ›› Issue (3): 442-448.DOI: 10.3969/j.issn.0372-2112.2020.03.004

• 学术论文 • 上一篇    下一篇

一种基于混合神经网络的命名实体识别与共指消解联合模型

郜成胜1, 张君福1, 李伟平1, 赵文2, 张世琨2   

  1. 1. 北京大学软件与微电子学院, 北京 100871;
    2. 北京大学软件工程国家工程研究中心, 北京 100871
  • 收稿日期:2018-09-24 修回日期:2018-11-22 出版日期:2020-03-25
    • 作者简介:
    • 郜成胜 男,1975年出生,山西长治人,现为北京大学软件与微电子学院工程博士研究生,主要研究领域为知识图谱、软件工程.E-mail:gaocs@pku.edu.cn;赵文 男,1967年出生,辽宁大连人.现为北京大学软件工程国家工程研究中心研究员、博士生导师,主要研究领域为知识图谱、软件工程、软件安全.E-mail:zhaowen@pku.edu.cn

A Joint Model of Named Entity Recognition and Coreference Resolution Based on Hybrid Neural Network

GAO Cheng-sheng1, ZHANG Jun-fu1, LI Wei-ping1, ZHAO Wen2, ZHANG Shi-kun2   

  1. 1. School of Software and Microelectronics, Peking University, Beijing 100871, China;
    2. National Engineering Research Center for Software Engineering, Peking University, Beijing 100871, China
  • Received:2018-09-24 Revised:2018-11-22 Online:2020-03-25 Published:2020-03-25

摘要: 命名实体识别与共指消解均依赖于对实体相邻文本信息的学习,本文提出一种基于混合神经网络的命名实体识别与共指消解联合模型,共用双向长短时记忆模型LSTM编码层对输入序列中每个词前后方向上下文信息进行编码,并通过训练学习得到上下文信息传递到前馈神经网络FFNN模型以提高共指消解精度,通过将领域文档及篇章语义向量加入FFNN,改进共指消解算法并优化共指消解模型.基于领域文本数据集进行联合模型训练,实验结果表明该联合模型可以有效地提高共指消解精度.

关键词: 神经网络, 命名实体识别, 共指消解, 联合神经网络模型

Abstract: Considering that both named entity recognition and coreference resolution depend on the same context of the entity word, this paper proposes a hybrid neural network model to settle these problems which contains a named entity recognition (NER) module and a coreference resolution (CR) module. NER and CR share the same bidirectional LSTM encoding layer, which is used to encode each input word by taking into account the context on both sides of the word. The contextual information of entities obtained in BiLSTM encoding layer further pass through to FFNN module to improve the coreference resolution. Furthermore, by adding domain documents and chapter semantic vectors to FFNN, the coreference resolution algorithm is improved and the coreference resolution model is optimized. Finally, we conduct experiments on the domain dataset to verify the effectiveness of our method. The joint model can effectively improve the accuracy of coreference resolution task.

Key words: neural network, named entity recognition, coreference resolution, hybrid neural network model

中图分类号: