Joint Learning for Aspect Category Detection and Sentiment Analysis in Chinese Reviews

Abstract

Aspect based sentiment analysis (ABSA) is a valuable task, aiming to predict the sentiment polarities of the given aspects (terms or categories) in review texts. Aspect-category sentiment analysis is a sub-task of ABSA, which mainly focus on aspect category detection and aspect category polarity identifying. Most of the previous methods employ a pipeline strategy, regarding aspect category detection and category sentiment analysis as two separate tasks, which could not meet the needs of practical application. To address this limitation, we propose an end-to-end neural network model based on joint learning, which can detect aspect category and identify aspect category polarity simultaneously. We conduct several comparable experiments on a Chinese review dataset and the experimental results show that our proposed model is simpler and more effective than the baseline models.