基于双通道交互的图卷积网络方面级情感分析

Aspect-Level Sentiment Analysis Based on Bi-Channel Interactive Graph Convolutional Network

基于方面的情感分析旨在识别文本中特定方面的情感极性,目前,大多数研究将句法依赖树和图卷积神经网络应用于方面级情感分析,并取得了不错的结果。然而,当一个评论文本包含多个方面时,大多数方法对每个方面单独建模,从而忽略了方面词之间的情感联系。为了解决这一问题,本文提出了一种用于方面级情感分析的双通道交互式图卷积网络(BC-GCN)模型。该模型同时考虑了句子的句法结构信息以及多个方面的情感依赖性,并使用图卷积网络来学习其节点信息表示。特别是,为了更好地捕获方面词和观点词的表示,我们利用交互注意力机制来学习图卷积网络产生的句法信息特征和多方面情感依赖特征。在多组公开数据集上的实验结果表明,我们提出的改进方法能显著提高模型的性能。

Aspect-based sentiment analysis aims to identify the emotional polarity of specific aspects of text, and currently, most studies apply syntactically dependent trees and graph convolutional neural networks to aspect-level sentiment analysis with good results. However, when a review text con-tains multiple aspects, most methods model each aspect individually, ignoring the emotional connection between aspect words. To solve this problem, a Bi-channel interactive graph convolutional network (BC-GCN) model for aspect-based sentiment analysis is proposed. The model considers both the syntactic structure information of sentences and the emotional dependence of multiple aspects, and uses the graph convolutional network to learn its node information representation. In particu-lar, in order to better capture the representation of aspect words and opinion words, we use the interactive attention mechanism to learn the syntactic information features and multifaceted emotional dependency features generated by graph convolutional networks. Experimental results on multiple sets of public datasets show that the proposed improved method can significantly improve the performance of the model.