Acta Scientiarum Naturalium Universitatis Pekinensis ›› 2022, Vol. 58 ›› Issue (1): 21-28.DOI: 10.13209/j.0479-8023.2021.108

Previous Articles     Next Articles

An Emotion-Cause Pair Extraction Model Based on Multichannel Compact Bilinear Pooling

HUANG Jin1, XU Shi1, CAI Ercong1, WU Zhijie1, GUO Meimei1, ZHU Jia2,†   

  1. 1. School of Computer Science, South China Normal University, Guangzhou 510631 2. The Key Laboratory of Intelligent Education Technology and Application of Zhejiang Normal University, Jinhua 321004
  • Received:2021-06-08 Revised:2021-08-13 Online:2022-01-20 Published:2022-01-20
  • Contact: ZHU Jia, E-mail: jiazhu(at)


黄晋1, 许实1, 蔡而聪1, 吴志杰1, 郭美美1, 朱佳2,†   

  1. 1. 华南师范大学计算机学院, 广州 510631 2. 浙江师范大学智能教育技术与应用重点实验室, 金华 321004;
  • 通讯作者: 朱佳, E-mail: jiazhu(at)
  • 基金资助:


The authors propose a model based on multichannel compact bilinear pooling to rank pair candidates in a document. The proposed model firstly extracts the emotion-specific and cause-specific representation containing position information via graph attention network, then further learns the local relation representation between emotion clause and cause clause through the local relation-aware module. Finally, these representations are fused via multichannel compact bilinear pooling to learn the emotion-cause pairs representation for effective ranking. Experimental results show that the proposed approach achieves the best performance among all compared approaches on the task.

Key words: sentiment analysis, emotion-cause pair extraction, graph attention network, local relation-aware module, multichannel compact bilinear pooling


提出一个基于多通道压缩双线性池化的模型, 对文档中的候选情感?原因句子对进行排序。该模型利用图注意力网络提取包含位置信息的情感特定化表示和原因特定化表示, 通过局部关系学习模块, 进一步学习情感与原因句子之间的局部关系表示, 再使用多通道压缩双线性池化来融合学习情感?原因候选句子对表示。最后, 对候选句子对进行排序。实验结果表明, 与最新模型相比, 所提模型在多方面表现更优。

关键词: 情感分析, 情感?原因句子对提取, 图注意力网络, 局部关系提取, 多通道压缩双线性池化