Dual Graph Convolutional Networks with Semantic and Syntax Reinforcement for Aspect-based Sentiment Analysis

Aspect-based sentiment analysis (ABSA) aims to identify the sentiment polarity of the given aspect. Dependency trees contain different types of syntactic dependencies, but existing approaches do not consider the impact of different syntactic relations. Besides, how to efficiently capture the connection between aspects and context is also a challenge. Hence, we propose Dual Graph Convolutional Networks with Semantic and Syntax Reinforcement (SSRGCNs) for ABSA task. Specifically, we introduce a semantic context module using aspect-focused attention to capture aspect-related semantic context and self-attention to understand global information of the sentence. A weight allocation module is also proposed to assign different weights to different syntactic labels to fully consider the different influence. Then, we stack Graph Convolutional Networks (GCNs) layers over them to extract latent context representations. Finally, we propose a feature fusion mechanism to better integrate all features. Experiments on benchmark datasets demonstrate the effectiveness of our model.