基于图神经网络的方面级情感分析

Aspect-based sentiment analysis based on graph neural network

目前基于循环神经网络和注意力机制的方面级情感分析模型缺乏解释相关句法约束和远程单词依赖关系。针对该问题提出结合句子依存树和单词序列信息建立句子关系图模型。首先将句子表示为图,单词作为图的节点,依存句法树的边和单词序列作为图的边;然后提出邻接矩阵标记方案对句子关系图进行标记;最后利用图神经网络实现节点和边的分类任务。该模型在SemEval2014任务中的restaurant和laptop两个数据集上进行实验,在两个数据集上F 1值提升了5%左右。实验结果表明,将句子转换成图利用图神经网络对句子进行方面级情感分析是有益的。

Most aspect sentiment analysis is based on recurrent neural networks and attention mechanisms,and these models lack a mechanism to account for relevant syntactical constraints and long-range word dependencies.To tackle this problem,this paper proposed a sentence relationship graph model based on sentence dependency tree and word sequence information.Firstly,it transformed the sentence into a graph,constructed the words as nodes of the graph,and the edges in the graph were composed of word dependency and word sequence information.Then it proposed an adjacency matrix tagging scheme to mark the sentence relation graphs.Finally,it used graph neural network to realize the classification task of nodes and edges.This model was tested on two datasets of restaurant and laptop in SemEval2014,and the classification F 1-score of the two datasets was increased by about 5%.The experimental results show that it is beneficial to transform sentences into graphs and use the graph neural network to perform aspect-based sentiment analysis on sentences.