BiGRU-Multi-Head Self-Attention Network for Chinese sentiment classification

Abstract

The rapid development of Internet and the extensive user base have led to a significant increase in the propagation speed of information and public opinion. Quick and accurate sentiment analysis of public opinion can help analyze and manage public opinion. Since BiGRU can simplify network parameters, and the Multi-Head Self-Attention(MHSA) mechanism can learn long-distance features more accurately and quickly. In this study, we propose a BiGRU-MHSA network to perform Chinese sentiment classification. The innovation of the model lies in the combination of BiGRU and MHSA in order to focus on several semantic centers and comprehend the whole text more precisely and efficiently. At the same time, we use the Chinese comment datasets on Weibo to verify the effectiveness and accuracy of the model. From the experimental results, we find that our proposed model improves the accuracy by about 1% on different comments datasets without consuming too much computing resources.