A Syntax-based BSGCN Model for Chinese Implicit Sentiment Analysis with Multi-classification

Abstract

There are two types of sentiment analysis about Chinese sentences: explicit sentiment analysis and implicit sentiment analysis. Implicit sentiment, unlike explicit sentiment, lacks a well-defined sentiment vocabulary. Currently, the majority of relevant research has focused on extracting implicit emotion via word analysis, ignoring the role of syntactic structures and relationships between words in the analysis of implicit emotions in Chinese. In this paper, we use a graph convolutional neural network (GCN) to analyze the syntactic structure of implicit sentiment texts, then combine it with a Bidirectional Encoder Representations from Transformers (BERT) to extract contextual information to create the BSGCN, a multi-classification Chinese implicit sentiment analysis model, that can classify implicit sentiment Chinese sentences into five types: happiness, sadness, disgust, surprise, and neutral. In the experiment based on the dataset SMP-ECISA, the accuracy of the Chinese implicit sentiment analysis model proposed in this paper was 82.1%, which is a significant improvement over existing models.