Akademik Veri Yönetim Sistemi
5th International Artificial Intelligence and Data Science Congress (ICADA), İzmir, Türkiye, 24 - 25 Nisan 2025, sa.50, ss.757-767, (Tam Metin Bildiri)
This study investigates the application of fine-tuned large language models for gen- erating stylistically hybrid poetry inspired by William Shakespeare and William Blake. A dataset comprising over 400 sonnets and poems from both authors, retrieved via the Po etryDB API, was constructed for model training. Visual inputs were translated into descrip- tive prompts using the BLIP image captioning model. Two configurations were evaluated: a baseline GPT-3.5-Turbo model and a custom fine-tuned variant. Four distinct images served as prompts, generating eight poems in total. Outputs were assessed by a panel of English literature graduates, who rated each poem’s stylistic alignment (on a Shake- speare–Blake continuum) and semantic relevance to the associated image. The fine-tuned model demonstrated superior performance, producing more semantically coherent and stylistically integrated poems. The frequent emergence of lexical markers such as “thy,” “Je- rusalem,” “thee,” and “Lamb of God” supported evidence of effective stylistic fusion. These results highlight that fine-tuning large language models can effectively balance classical poetic forms, thematic elements, and visual prompt relevance, enabling the generation of hybrid literary styles.