This study systematically explores the affective tendency of traditional Chinese cultural texts and their communication effects by constructing a corpus of traditional Chinese culture (COCC) and proposing a multimodal feature fusion and structural correlation analysis method, which covers five genres: spoken language, novels, magazines, newspapers, and academic essays, and is combined with cultural loaded word filtering to ensure the representativeness of the data. In order to solve the problem of cross-domain sentiment semantic ambiguity, a feature fusion method based on domain label description is proposed, which generates domain-specific labels through TF-IDF and word vector techniques, and designs a bi-directional attention-gated recurrent unit Bi-AUGRU model to optimize the text feature extraction. The clause association analysis model (DSAM-CC) is further proposed to integrate the discourse structure tree RST with sentiment association features to capture the coherence and logic of sentiment propagation at the document level. Experiments show that the DSAM-CC model achieves an accuracy of 82.65% and an F1 value of 77.76% in the sentiment analysis task, which is significantly improved over the benchmark models MemNet and LSTM. Word frequency statistics and visual analysis show that high-frequency adjectives such as “benevolence” and “loyalty” and nouns such as “Confucianism” and “solar terms” embody the ethics and pluralism of traditional culture, while the emotional distribution reveals the coexistence of positive reinforcement of traditional values and modern criticism. This study provides a data-driven analytical framework for the sentiment transmission mechanism of traditional culture, which helps the digital research of cultural heritage.
