Aspect-Based Sentiment Analysis Based on Dual GraphSAGE Using External Knowledge
Guijun Luo, Qilie Liu, Qian Liu · 2023
Aspect-based sentiment analysis (ABSA) is the process of breaking down sentence structures to identify referenced entities and evaluate the sentiment orientation towards each aspect. While Graph Convolutional Network (GCN) models leveraging textual syntactic and semantic information have demonstrated strong performance, they typically assign equal weights to all edges between words. GraphSAGE, on the other hand, allows for the assignment of varying weights to different edges by aggregating edge weights between nodes. Moreover, existing models often solely concentrate on syntactic context information, overlooking external textual information. To address these limitations, we propose a dual GraphSAGE network that integrates both syntactic and semantic information while incorporating external sentiment knowledge. Ultimately, we harness the learning potential of this integrated information alongside n-gram grammar features. The empirical results from the evaluation conducted on three standard datasets provide corroborating evidence for the effectiveness of our proposed model.