Aspect-level sentiment analysis based on BERT fusion multi-attention

Abstract

In view of the fact that the existing aspect-level sentiment analysis (ABSA) model cannot effectively distinguish the importance of aspect words and words in the text, and lacks the utilization of the overall interaction between aspect words and text, an aspect-level approach based on BERT combined with multi-attention is proposed. The sentiment analysis model captures the interaction and correlation between aspect words and the overall text sentence through the text and aspect word interactive attention mechanism, thereby improving the accuracy of ABSA. Comparative experiments are carried out on the restaurant and laptop datasets of the Semeval2014 evaluation task. The experimental results show that the model proposed in this paper achieves good classification results in the aspect-level sentiment analysis task for short text reviews. This method provides a new idea for ABSA for review texts.