Modern Chinese novels contain rich emotional expressions. In this study, a BERT-BiGRU sentiment analysis model incorporating a sentiment lexicon is constructed for parsing the emotional color in modern Chinese novels. The novel text is preprocessed by jieba segmentation technique, the sentiment lexicon is constructed by combining Hownet and SentiWordNet, and the BERT-BiGRU network architecture is integrated to form a sentiment analysis model with bidirectional semantic comprehension capability. The experimental evaluation shows that the BERT-BiGRU model achieves 93.1%, 92.2%, and 92.6% in precision, recall, and F1-score metrics, respectively, which are 7.4%, 20.1%, and 14.3% higher than the GRU model. When applied to novel text analysis, the model successfully draws sentiment change curves, revealing the laws of sentiment flow in different novels. By calculating the Hurst parameter of 1514 modern Chinese novels, it is found that 93.2% of the excellent novels have a Hurst value greater than 0.5, 87% of which are concentrated in the range of 0.52-0.74, which indicates that the novels’ emotional dynamics generally have long-range correlation. The study confirms that the BERT-BiGRU model enhanced by the emotional lexicon can effectively capture the emotional veins in novel texts, providing a computational perspective for literary analysis, and revealing the common law of excellent novels in emotional construction, providing a quantitative reference for novel creation and evaluation.
