SDEGCN: Syntactic dependency enhanced and integrated graph convolutional network for aspect-based sentiment analysis

Abstract

Aspect-based sentiment analysis (ABSA) aims to identify the sentiment polarity of specific aspects within a sentence. Existing graph convolutional network (GCN) approaches often suffer from insufficient modeling of dependency relations on specific aspects and the shallow integration of syntactic information. To address these issues, this paper proposes a syntactic dependency enhanced and integrated graph convolutional network (SDEGCN), which aims to effectively mine syntactic dependency relations and deeply integrate them into the model. Firstly, in the syntactic dependency enhancement layer, the dependency location-aware algorithm highlights the core syntactic roles of aspect terms and their relevant context, while the syntactic consistency constraint guide the syntactic graph convolutional network to learn more effective representations. Then, the semantic encoding layer calculates attention scores through self-attention mechanisms to optimize the adjacency matrix input of the graph convolutional network, thereby capturing semantically relevant features of sentences. Finally, the feature fusion layer employs a biaffine attention transformation mechanism to fuse syntactic and semantic features, and after pooling and concatenation aggregation, the classification is completed. Extensive experimental results on five benchmark datasets demonstrate that the SDEGCN significantly outperforms existing graph convolutional baseline models, proving its effectiveness in ABSA tasks.