Deep learning for Aspect-based Sentiment Analysis

With the popularity of the Internet and the amount of information people can get from the Internet increases exponentially, sentiment analysis depending on text information has become a rather important tool in daily life, which helps people improve the quality of products and their lifestyle. According to the different granularity of sentiment text, sentiment analysis can be divided into three levels: word, sentence, and document level. While coarse-grained sentiment analysis only focuses on the polarity of emotions and neglects the finer-grained tendency of emotions. Driven by this demand, some researchers put forward the aspect-based sentiment analysis. ABSA can be categorized into two basic steps: aspect extraction and sentiment classification, on which researchers are mainly focusing to improve its performance. Deep learning has gradually become a research hotspot in recent years and has been widely used in CV and NLP fields. Recently, some efforts have begun to apply deep learning to ABSA tasks. In this paper, we further summarize and analyze the recent achievements of ABSA in the application of different deep learning models in terms of their completeness, accuracy, and detail. We also summarize ABSA's current challenges and difficulties and look forward to the future development and improvement of ABSA based on the existing experience.