MSGAT-Based Sentiment Analysis for E-Commerce

Sentence-level sentiment analysis, as a research direction in natural language processing, has been widely used various fields. In order to address the problem that syntactic features were neglected previous studies on sentence-level multiscale graph attention network (MSGAT) analysis model based dependent syntax is proposed. The adopts RoBERTa_WWM text encoding layer, generates graphs basis of dependency trees, and obtains sentence at different scales for classification through multilevel network. Compared with existing mainstream models, proposed achieves better performance both hotel review dataset takeaway dataset, 94.8% 93.7% accuracy 96.2% 90.4% F1 score, respectively. results demonstrate superiority effectiveness Chinese analysis.