Research On Aspect Based Sentiment Analysis Method Based On Graph Attention Network

In recent years,with the development of mobile Internet,people are eager to post comments on an event or a product online.These comments often hide users’ sentiment tendencies,and conducting sentiment analysis on them can provide comprehensive and scientific decision support for consumers,enterprises and governments.Aspect based sentiment analysis is a fine-grained sentiment analysis that aims to determine the sentiment polarity of a given aspect in a text.Compared with sentiment analysis tasks at other levels,aspect based sentiment analysis has a finer granularity and requires higher performance of the model,so it has received more attention.Most of the current mainstream methods use dependency syntax tree and graph attention network to analyze and extract the sentiment polarity of aspect words,but these methods usually ignore the information redundancy problem of dependency syntax tree and cannot effectively deal with syntactically structure-insensitive comments.To address the problems in the current study,this paper further investigates aspect based sentiment analysis based on graph attention networks.The main work of this paper is as follows:(1)In order to make full use of the syntactic information of the text,a graph attention network model based on the dependency syntax tree is designed in this paper.The model uses the graph attention network to extract the syntactic relationship in the dependency syntax tree,and according to the dependency relationship between words,it can more accurately capture the opinion words related to the emotional polarity of the aspect words.Firstly,the original dependency syntax tree is pruned according to syntactic distance to retain only the information that is closer to the target aspect.Secondly,considering the influence of dependency relation type and lexicality,additional relation heads are extended in the graph attention network to fuse relation type and lexicality information into the node representation.Finally,the results of experiments on three datasets show that the model in this paper has some performance improvement compared with the baseline model.(2)In response to the problem of incorrect dependency parsing caused by syntactic insensitivity of some comments and the problem of important information loss caused by excessive pruning,this paper introduces a semantic information extraction module and further proposes an aspect based sentiment analysis model incorporating semantic information.The semantic information extraction module uses a self-attention mechanism to generate the adjacency matrix and encodes it using a graph attention network to extract semantic features and fully exploit the internal relationships between words.In addition,the model uses positional encoding to enhance the word representation to better mine the sequence information.Finally,the syntactic feature representation and semantic feature representation are fused by a gating mechanism to obtain a richer sentence representation.The model is experimented on three classical datasets in several dimensions,and the results of experiments show that the model in this paper has high superiority.