Theme Classification of Chinese Classical Poetry Based on the GB-MSA-BGTCN Hybrid Model

The theme of Chinese classical poetry reveal its connotations and artistic conceptions, making them essential for understanding the poetry. Therefore, studying the theme classification of Chinese classical poetry is crucial. However, existing methods have struggled to effectively leverage key information and extract deep textual features. Traditional word embedding models have also failed to address polysemy, limiting their ability to convey the complex semantics of classical poetry. To overcome these challenges, we propose the GB-MSA-BGTCN model for theme classification. This model integrates GuwenBERT, a Multi-head Self-Attention (MSA) mechanism, Bidirectional Gated Recurrent Unit (BiGRU), and Temporal Convolutional Network (TCN). GuwenBERT, pre-trained on classical text corpora, is used for word embedding to effectively represent the semantic information of classical poetry. The MSA mechanism enhances classification accuracy by focusing on key textual information at multiple scales. The combined use of BiGRU and TCN captures long-term dependencies and contextual information, enabling deep feature extraction and improving performance. Due to the lack of public datasets on Chinese classical poetry themes, we constructed a new dataset for evaluation. Experimental results show that our model achieved precision, recall, and F1 scores of 84.10%, 81.65%, and 82.64%, respectively, surpassing other comparative models.