BERT-IAN Model for Aspect-based Sentiment Analysis

Abstract

Aspect-based sentiment analysis is different from document-level and sentence-level sentiment analysis, which aims to predict the sentiment polarity of a certain aspect in a sentence. The accuracy of the existing aspect-based sentiment analysis model still needs to be improved. A BERT-IAN sentiment analysis model that improves the Interactive Attention Networks (IAN) model is proposed to further improve the accuracy of the aspect-based sentiment analysis. First use the BERT pre-training model to encode aspects and context respectively. Then use a transformer encoder with interactive attention to interactively learn the attention of the aspect and context, and generate a final representation. Finally, through the sentiment classification layer, the aspect corresponding sentiment are analyzed. The experimental results on Restaurant and Laptop datasets show the effectiveness and superiority of the BERT-IAN model.