Syntactic Structure-Enhanced Dual Graph Convolutional Network for Aspect-Level Sentiment Classification

Aspect-level sentiment classification (ALSC) is a fine-grained analysis task that aims to predict the of given aspect in sentence. Recent studies mainly focus on using Graph Convolutional Networks (GCN) deal with both semantics and syntax However, improvement limited since dependency trees are not aspect-oriented exploitation structure information inadequate. In this paper, we propose Syntactic Structure-Enhanced Dual Network (SSEDGCN) model for an ALSC task. Firstly, enhance relation between its opinion words, aspect-wise tree by reconstructing basic tree. Then, syntax-aware GCN encode new For learning, semantic-aware established. order exploit syntactic information, design syntax-guided contrastive learning objective makes aware improves quality feature representation aspect. The experimental results three benchmark datasets show our significantly outperforms baseline models verifies effectiveness model.