计算机工程与应用 ›› 2021, Vol. 57 ›› Issue (9): 134-139.DOI: 10.3778/j.issn.1002-8331.2001-0341

• 模式识别与人工智能 • 上一篇    下一篇

基于注意力交互机制的层次网络情感分类

杨春霞,李欣栩,吴佳君,刘天宇   

  1. 1.南京信息工程大学 自动化学院,南京 210044
    2.江苏省大数据分析技术重点实验室,南京 210044
    3.江苏省大气环境与装备技术协同创新中心,南京 210044
    4.佳木斯市汤原县气象局,黑龙江 佳木斯 154000
  • 出版日期:2021-05-01 发布日期:2021-04-29

Hierarchical Network Sentiment Classification Based on Attention Interaction Mechanism

YANG Chunxia, LI Xinxu, WU Jiajun, LIU Tianyu   

  1. 1.School of Automation, Nanjing University of Information Science & Technology, Nanjing 210044, China
    2.Jiangsu Key Laboratory of Big Data Analysis Technology(B-DAT), Nanjing 210044, China
    3.Jiangsu Collaborative Innovation Center of Atmospheric Environment and Equipment Technology(CICAEET), Nanjing 210044, China
    4.Tangyuan County Meteorological Administration of Jiamusi City, Jiamusi, Heilongjiang 154000, China
  • Online:2021-05-01 Published:2021-04-29

摘要:

目前关于商品评论的深度网络模型难以有效利用评论中的用户信息和产品信息。提出一种基于注意力交互机制的层次网络(HNAIM)模型。该模型利用层次网络对不同粒度语义信息进行提取,并通过注意力交互机制在层次网络中通过捕捉用户、产品中的重要特征来帮助提取文本特征。最终将用户视角下的损失值和产品视角下的损失值作为辅助分类信息,并利用层次网络输出的针对用户或产品的关键文本特征进行训练和分类。三个公开数据集上对比结果表明,该模型较相关模型而言效果均有提升。

关键词: 情感分类, 粒度, 双向长短期记忆网络, 注意力交互机制

Abstract:

It is difficult for current deep network models for reviews of products to effectively use user information and product information in reviews. This paper proposes a hierarchical network(HNAIM) model based on the attention interaction mechanism. This model uses hierarchical networks to extract semantic information of different granularities, and uses the attention interaction mechanism in the hierarchical network to capture the important features of users and products to help extract text features. The loss value from the perspective of the user and the loss value from the perspective of the product are used as auxiliary classification information. And the hierarchical network is used to output for the key text features of users or products for training and classification. The comparison results on the three public data sets show that the model is more effective than related models.

Key words: sentiment classification, granularity, Bidirectional Long Short-Term Memory(BiLSTM), attention interaction mechanism