BIDIRECTIONAL SEMANTIC AND HIERARCHICAL SYNTACTIC SENTIMENT CLASSIFICATION BASED ON GCN

In order to solve the problems of insufficient coordination between global semantics and local syntactic features and noise interference of dependency parsing in aspect sentiment classification tasks, this paper proposes a bidirectional semantic enhancement and hierarchical syntactic analysis model based on graph convolutional network (GCN). The model effectively integrates semantic enhancement GCN and syntactic enhancement GCN for feature interaction to accurately model the complex hierarchical relationship between aspect words and sentiment words. In semantic modeling, self-attention and perceptual aspect attention (ASA) are integrated to extract deep semantic information through the attention fusion mechanism (GAFM). In terms of syntactic feature extraction, the syntactic distance mask matrix is introduced to measure semantic association, and the syntactic modeling is optimized in combination with the dependency relationship. In terms of structural optimization, the hierarchical phrase structure is adopted to fuse the syntactic dependency matrix with the phrase matrix to significantly reduce the noise of dependency tree parsing. Experimental results show that the model performs well on multiple datasets, with consistently improved accuracy and stability. Ablation experiments and visualization analysis further verify the effectiveness of each module, proving that the combination of bidirectional semantic enhancement and hierarchical syntactic analysis helps substantially improve the performance of aspect sentiment classification.