TS-GCN : Aspect-level Sentiment Classification Model for Consumer Reviews

Shunxiang Zhang, Tong Zhao, Houyue Wu, Guangli Zhu, KuanChing Li

The goal of aspect-level sentiment classification (ASC) task is to obtain the sentiment polarity of aspect words in the text. Most existing methods ignore the implicit aspects, resulting in low classification accuracy. To improve the accuracy, this paper proposes a classification model for consumer reviews, abbreviated as TS-GCN (Truncated history attention and Selective transformation network-Graph Convolutional Networks). TS-GCN can classify sentiment from both explicit and implicit aspects. Firstly, we process the text by the BERT model and the BiLSTM model to obtain the text features. Secondly, the GCN model completes explicit sentiment classification by training text features. Due to the lack of implicit words, the GCN model cannot classify implicit sentiments. Finally, we predict implicit words based on the TS model, which makes up for the deficiency of the GCN model and completes the sentiment classification of implicit words. TS-GCN is proved on several datasets in the consumer reviews field. The results of experiments show that the TS-GCN can improve the accuracy and F1 of ASC.