Development and evaluation of a fine-tuned model to improve lifestyle management for prostate cancer patients, suggesting effective health literacy intervention.
Key Points
The study aims to develop and evaluate a supervised fine-tuned large language model for lifestyle management tailored to prostate cancer patients.
Constructed a structured lifestyle management knowledge base using literature from PubMed.
Developed bilingual QA pairs through retrieval-augmented generation and independent English and Chinese test sets.
Trained Baichuan2-7B-Chat with a 2-stage strategy and assessed outputs using double-blind evaluations against other models.
Developed a comprehensive dataset with over 42,330 single-turn QA pairs and 3008 multiturn dialogues.
PCaPLMM_SFT outperformed Baichuan2-7B-Chat and showed comparable or superior performance to GPT-3.5-Turbo.
Consistency analyses revealed robust agreement between referee models across evaluation rounds.