Fin-TuneNet: A Hierarchical Fine-Tuning Framework for Domain-Specific NLP Optimization
Fu Lei, Qianqian Xu, Yihong Jin, Yuan Tian, Kowei Shih, Kuan Lu · 2025
The challenges of financial natural language processing (NLP) are significant. Specialized terminology, complex dependencies, and strict compliance rules make deploying large language models (LLMs) in finance difficult. Existing models like GPT-3, FinBERT, and T5-Financial show potential but fail to generalize well across financial tasks. This paper presents Fin-TuneNet, a hierarchical fine-tuning framework to improve LLMs for financial applications. It integrates task-specific adapters, multi-scale attention mechanisms, and domain-guided prompt engineering. Fin-TuneNet aligns better with tasks and achieves robustness while keeping computational costs low. This frame-work is an efficient and domain-specific solution for applying LLMs in financial NLP.