一种基于图卷积网络的文本情感分类方法

A text sentiment classification method based on graph convolution network

为了更好地分析单词和长距离依赖的作用,解释相关的句法约束,本文提出了一种基于多头注意力机制和图卷积网络模型MHGCN,在句子的依存关系树上建立一个图卷积网络,以利用句法信息和单词依存关系。利用多头注意力机制学习多个表示子空间的相关信息,并使用图卷积网络获得句法信息和长距离依赖。实验表明,MHGCN模型能有效完成情感分类任务,可为人机交互、医疗保健和社交媒体舆情监测等提供参考依据。

In order to better analyze the role of words and long-distance dependencies,and explain relevant syntactic constraints,this paper proposes a graph convolutional network model MHGCN based on multi head attention mechanism and graph convolutional network.A graph convolutional network is established on the dependency tree of a sentence to utilize syntactic information and single word dependencies.Utilize multi head attention mechanism to learn relevant information of multiple representation subspaces,and use graph convolutional networks to obtain syntactic information and long-distance dependencies.Experiments have shown that the MHGCN model can effectively complete sentiment classification tasks and provide reference basis for human-computer interaction,healthcare,and social media public opinion monitoring.