Aspect-Based Sentiment Analysis Using Graph Convolutional Networks and Co-attention Mechanism

Abstract

Aspect-based sentiment analysis (ABSA) refers to classifying the sentiment polarity of a specific aspect in a sentence. Recently, attention-based deep learning approaches are proposed to capture the semantic information and achieve satisfying results. However, due to the significance of syntactic structure, syntactic information is also analyzed for ABSA. As such, this work proposes a model that integrates the graph convolution network (GCN) and the co-attention mechanism to deal with the aspect-based information and remove the noise from unrelated context words. Both the semantic information and the syntactic information are conveyed by the representation for sentiment analysis. Experimental results show our model achieves a better working performance, which establishes a strong evidence of the capability.