Chinese Short Text Classification Based on Semantic Fusion and Syntactic Analysis

Most existing text classification models are designed for English and often perform poorly on Chinese texts due to the latter's complex syntax and lack of explicit word boundaries. To address these challenges, we propose BRGAT-SF (BERT-wwm Graph Attention Network with Semantic Fusion), a novel model that integrates semantic features with syntactic dependency information. By introducing a semantic fusion module, BRGAT-SF effectively integrates character-level and word-level information, enhancing the quality of semantic representations for Chinese text. A dedicated syntactic encoding module jointly models semantic and syntactic information, allowing the network to capture deeper structural patterns in Chinese sentences. Experiments on four public Chinese short text classification datasets show that BRGAT-SF consistently outperforms strong baselines in terms of accuracy, recall, and F1-score. These results confirm the effectiveness of incorporating syntactic structures into semantic modeling and demonstrate the robustness of BRGAT-SF for Chinese text classification.