计算机工程与应用 ›› 2023, Vol. 59 ›› Issue (3): 135-142.DOI: 10.3778/j.issn.1002-8331.2108-0215

• 模式识别与人工智能 • 上一篇    下一篇

基于AWI和GCN的方面级情感分类模型

王泽,孔韦韦,薛佳伟,平稳,李龙   

  1. 1.西安邮电大学,西安 710121
    2.陕西省网络数据分析与智能处理重点实验室,西安 710121
    3.桂林电子科技大学,广西 桂林 541004
    4.广西可信软件重点实验室,广西 桂林 541004
  • 出版日期:2023-02-01 发布日期:2023-02-01

Aspect Level Sentiment Classification Model Based on AWI and GCN

WANG Ze, KONG Weiwei, XUE Jiawei, PING Wen, LI Long   

  1. 1.Xi’an University of Posts and Telecommunications, Xi’an 710121, China
    2.Shaanxi Provincial Key Laboratory of Network Data Analysis and Intelligent Processing, Xi’an 710121, China
    3.Guilin University of Electronic Technology, Guilin, Guangxi 541004, China
    4.Guangxi Key Laboratory of Trusted Software, Guilin, Guangxi 541004, China
  • Online:2023-02-01 Published:2023-02-01

摘要: 目前大多数方面级情感分类研究都忽略了方面词的建模,以及方面词与上下文之间的交互信息,并且难以体现语法上与方面词有直接联系上下文单词的重要程度。针对上述问题,提出基于方面词交互(aspect word interaction,AWI)和图卷积网络(graph convolutional network,GCN)的方面级情感分类模型(AWI-GCN)。使用双向长短期记忆网络(bi-directional long short-term memory,Bi-LSTM)分别提取方面词和上下文的特征;采用GCN根据句法依存树进一步提取与方面词有直接语法联系的上下文情感特征;利用注意力机制学习方面词与上下文的交互信息,同时提取上下文中为方面词情感分类做出重要贡献的情感特征。针对3个公开数据集上的仿真实验结果表明,AWI-GCN模型相比当前代表模型取得了更好的情感分类效果。

关键词: 方面级情感分类, 方面词交互, 图卷积网络, 注意力机制, 句法依存树

Abstract: At present, in most studies of aspect-level sentiment classification, the modeling of aspect words and their interactions with context are ignored, and it is difficult to reflect the importance of context words which are directly related to aspect words grammatically. To address the problem above, based on aspect word interaction(AWI) and graph convolutional network(GCN), this paper puts forward a model for the classification—AWI-GCN. Bi-directional long short-term memory(Bi-LSTM) is used to extract aspect words and context features respectively. GCN is adapted to further extract emotional context features directly related to aspect words on the basis of  the dependency syntax tree. Attention-Mechanism is applied to learn the interactive information between aspect words and context, and at the same time to extract the emotional features of context that make important contributions to the emotion classification of aspect words. The simulation results on three open data sets show that the AWI-GCN model achieves better sentiment classification performance than the current representative one.

Key words: aspect sentiment classification, aspect word interaction, graph convolutional network, attention mechanism, syntactic dependency tree