Exploring Cognitive Difference in Poetry Collection via Large Language Models and Metaphors: A Case Study of the Book of Songs

Hui Bao, Kai He, Yige Wang, Zeyu Gao

Abstract: Both modern and classical poetry collections consist of poems with various contents and styles. Each poem is imbued with rich metaphors, which reflect the emotions, thoughts, and cultural backgrounds of the poets. Previously, researchers have analyzed various poetry collections, but do it manually in most cases and focused on representative works within the collections, lacking a framework for large-scale systematic analysis. Metaphor analysis is an essential tool for reflecting language concept mapping and facilitating the comprehension of complex ideas. Recently, it has been widely applied in studies across politics, society, economics, and literature. With the advancement of large language models, metaphor analysis’ application scope and capabilities have been enhanced. In this study, we propose a framework “Poetic Insight” and conduct a comprehensive analysis of all 305 poems in the earliest Chinese poetry anthology, the “Book of Songs,” utilizing multiple commercial and open-source large language models. Through detailed analysis of semantic similarities and metaphorical expressions, we unveil the differences and connections among the poems, providing supporting evidence for the historical origins of several pieces, and offering insights for humanities scholars to leverage digital humanities methods in the future.