Abstract: This study combines large language models (LLMs) with the difference-in-differences (DID) causal identification method, taking the An Lushan Rebellion as a quasi-natural experiment to conduct a quantitative emotional analysis of Du Fu's 1,496 extant poems. It systematically reveals the causal effect of historical shocks on the emotional expression in literary creation and its transmission mechanism. A stratified random sampling approach was adopted to select 200 poems for manual annotation verification, thereby establishing a reliable measurement quality evaluation system. The findings are as follows: (1) The An Lushan Rebellion exerted a significantly negative impact on the emotional scores of Du Fu's war-themed poems, and this result remained consistent across manual annotation verification, replacement of large language models, and multiple robustness tests; (2) Poetry creation exhibited a significant emotional regulation function — poems themed around personal experiences and social concerns could effectively mitigate the negative impacts of war shocks; (3) Heterogeneity analysis indicated that the negative effects of war shocks were mainly concentrated in guti shi (ancient-style poetry) and pailü (regulated verses with extended lines). Additionally, poems rhyming with rusheng (entering tones, a category of ancient Chinese tones) were most significantly affected, which reveals the differentiated characteristics of poetic genres and phonological forms as carriers of emotional expression.