Combination of Convolutional Neural Network and Gated Recurrent Unit for Aspect-Based Sentiment Analysis
Narisa Zhao, Huan Gao, Xin Wen, Hui Li
Abstract
Aspect-based sentiment analysis (ABSA) aims to identify views and sentiment polarities towards a given aspect in reviews. Compared with general sentiment analysis, ABSA can provide more detailed and complete information. Recently, ABSA has become an important task for natural language understanding and has attracted considerable attention from both academic and industry fields. The sentiment polarity of a sentence is not only decided by its content but also has a relatively significant correlation with the targeted aspect. For this reason, we propose a model for aspect-based sentiment analysis which is a combination of Convolutional Neural Network (CNN) and Gated Recurrent Unit (GRU), utilizing the local features generated by CNN and the long-term dependency learned by GRU. Extensive experiments have been conducted on datasets of hotels and cars, and results show that the proposed model achieves excellent performance in terms of aspect extraction and sentiment classification. Experiments also demonstrate the great domain expansion capability of the model.