A Chinese Implicit Sentiment Analysis Model Based on Relational Weighted Graph Convolutional Network

Abstract

Implicit sentiment analysis is used to deal with text corpora that are devoid of sentiment words, which is a difficult task for text sentiment analysis. Implicit sentiment statements in the Chinese corpus are dominated by euphemistic and implicit expressions, and implicit sentiment analysis research can aid in improving the performance of Chinese sentiment analysis tasks. We propose a Chinese Implicit Sentiment Analysis Based on Relational Weighted Graph Convolutional Network (CISA-RWGCN) in this paper, which builds a model with RWGCN, sets different weight mapping matrices for different types of edge relations on GCN, and calculates the weights of neighbor nodes using node representation, node labels, and edge relation types of target and neighbor nodes. On the SMP-ECISA 2019 dataset, the results of comparison and ablation experiments show that the model's full utilization of syntactic information can help the Chinese sentiment classification task achieve 87.4% on the F1 score.