A knowledge-enhanced interactive graph convolutional network for aspect-based sentiment analysis

Deep neural networks, especially graph neural networks, have made great progress in aspect-based sentiment analysis. Knowledge graphs can provide rich auxiliary information for aspect-based sentiment analysis. However, existing models cannot effectively learn aspect-specific sentiment features from the review text and external knowledge. They cannot accurately select knowledge entities that are highly relevant to the aspect. They also ignore the semantic interaction between the review text and external knowledge. To address these issues, we propose a knowledge-enhanced interactive graph convolutional network (KE-IGCN). First, we introduce a subgraph construction strategy to construct a syntax-guided knowledge subgraph, which can guide KE-IGCN in selecting highly relevant knowledge entities. Second, we propose a knowledge interaction mechanism to exploit the semantic interaction between external knowledge and the review text. We then use multilayer graph convolutional networks to learn aspect-specific sentiment features from the review text and external knowledge jointly and interactively. We also use a multilevel feature fusion mechanism to aggregate aspect-specific sentiment features from semantic and syntactic information of the review and external knowledge. Experimental results on four public datasets demonstrate that KE-IGCN outperforms other state-of-the-art baseline models.