Reproduce “Pre-trained Large Language Models for Financial Sentiment Analysis” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Wei Luo, Dihong Gong
Researchers from Xi'an Jiaotong-Liverpool University and the University of Florida adapted a Llama2-7B model for financial sentiment analysis, classifying financial news titles into sentiment categories. The approach achieved an accuracy of 0.90 on the Financial PhraseBank benchmark, exceeding prior state-of-the-art methods by 0.04 percentage points.
Key insights
- Supervised fine-tuning is essential for adapting pre-trained LLMs to achieve high performance in domain-specific tasks, significantly outperforming few-shot prediction.
- Both generation-based (SFT) and direct classification (ClassHead) fine-tuning methods for Llama2-7B can yield equivalent high accuracy in financial sentiment analysis.
- The ClassHead method offers the practical advantage of providing classification confidence scores, which is valuable for applications requiring uncertainty quantification.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.