Identification of Conflict Opinion in Aspect-Based Sentiment Analysis using BERT-based Method

Abstract

Aspect-based sentiment analysis (ABSA) is an NLP task for predicting sentiment polarities of specific aspects in a given opinion sentence. Recent research shows that deep learning and language modeling like BERT has become state-of-the-art in NLP tasks, including ABSA. However, most methods still ignore conflict opinion or methods that reached high performance in 2-class (positive and negative), and 3-class (positive, negative, and neutral) classification will be degraded when applied in a 4-class classification where conflict opinion is included. In this paper, we propose a BERT-based method that can identify and handle aspects containing conflict opinions in three steps: (i) designing input representation for BERT-based sentence-pair classification task, (ii) processing two-label sentiment classification for each aspect, and lastly (iii) translating the second step result to 4-class sentiment classification. Experimental results on the SemEval-2014 restaurant domain dataset demonstrate that our proposed method has effectively identified conflict opinion and achieved better results on 3-class and 4-class classification tasks.