Aspect-level Sentiment Analysis with Dual-strategy Graph Convolutional Networks
Aiming at the problem of poor sentiment classification effect of aspect words caused by inaccurate dependency parsing results and complexity of online reviews,this paper proposes a method based on the Dual-strategy Graph Convolutional Network.Firstly,the probability ma-trix which reflects all dependencies in the dependency parser is used to construct a SynGCN module with rich syntactic knowledge.Secondly,the self-attention mechanism is utilized to ob-tain the attention score matrix of the adjacency matrix as the adjacency matrix of the WcoGCN module,and the regularizer is exploited to constrain the attention score in the module to accu-rately capture the correlations between words.Experimental results show that the accuracy and F1 value of the proposed model are better than other models,and it can improve the effect of sentiment classification on public data sets.
万方数据
维普
