Sentiment analysis of Chinese ancient poetry based on multidimensional knowledge attention.

This article focuses on the development and evaluation of a Sentiment Analysis Model of Chinese Ancient Poetry based on Multidimensional Knowledge Attention (SAMCAP). Recognizing the unique linguistic features and implicit sentiment expressions in Chinese ancient poetry, the model integrates semantic textual features extracted via GuwenBERT and BiLSTM with explicit and implicit knowledge features—including poet's background, allusion sentiment, imagery sentiment, and sentiment terms—retrieved from a constructed knowledge base. Experiments on the Fine-grained Sentimental Poetry Corpus (FSPC) demonstrate that SAMCAP outperforms several baseline models, achieving the highest F1 score of 0.9372, with ablation studies highlighting the significant contribution of explicit sentiment terms over implicit knowledge. The study suggests that incorporating multidimensional domain knowledge enhances the accuracy of sentiment analysis in Chinese ancient poetry, while noting challenges remain in fully capturing implicit sentiments and extending the approach to other classical Chinese literary genres.