Emotional Drifting of Chinese Characters in Tang Poetry: A Bipartite Network Analysis of Cultural Variability

The same Chinese character often conveys opposing emotional valences when combined with different morphemes in Tang poetry, a contextual affective phenomenon we term emotional drifting. Existing quantitative studies treat whole poems or isolated words as analytical units, overlooking this character-level cultural variability. This study constructs a character-bigram bipartite network and proposes the Emotional Drifting Index (EDI) to quantify such semantic variation based on Complete Tang Poetry (42,452 valid verses). After BPE segmentation and NPMI-based correlation filtering, multiple large language models assist in filtering non-poetic fixed phrases, followed by expert review. Separate positive and negative co-occurrence networks are built, and EDI is calculated via Euclidean distance of two types of node weighted degrees to quantify the intensity with which each character participates in both positive and negative emotional contexts, that is, a corpus-level measure of emotional-polarity coexistence. Statistical tests including ANOVA and Tukey HSD verify significant disparities across representative poets and four Tang historical stages. Empirically, around 23.2% of Chinese characters present measurable emotional drifting; Li Bai’s average EDI is over twice higher than Bai Juyi’s (p<0.001), and the overall drifting degree forms an inverted-V curve peaking in High Tang, which statistically corroborates the classic literary viewpoint of High Tang ethos. This bipartite-network-based quantitative scheme provides a feasible microscopic research path for traditional Tang poetic affective research.