Target-Aspect-Sentiment Joint Detection for Aspect-Based Sentiment Analysis

• DOI: 10.1609/aaai.v34i05.6447
• Publication Date: 2020-06-29T19:05:48Z
• Authors: Hai Wan; Yufei Yang; Jianfeng Du; Yanan Liu; Kunxun Qi; Jeff Z. Pan
• Keywords: SemEval; Sentiment Analysis; Polarity (international relations)

Aspect-based sentiment analysis (ABSA) aims to detect the targets (which are composed by continuous words), aspects and polarities in text. Published datasets from SemEval-2015 SemEval-2016 reveal that a polarity depends on both target aspect. However, most of existing methods consider predicting either or but not both, thus they easily make wrong predictions polarities. In particular, where is implicit, i.e., it does appear given text, do work. To tackle these limitations ABSA, this paper proposes novel method for target-aspect-sentiment joint detection. It relies pre-trained language model can capture dependence prediction. Experimental results restaurant show proposed achieves high performance detecting triples even implicit cases; moreover, outperforms state-of-the-art those subtasks detection competent to.