# Instruct-FinGPT: Financial Sentiment Analysis by Instruction Tuning of General-Purpose Large Language Models

**Authors:** Boyu Zhang, Hongyang Yang, Xiao-Yang Liu  
**arXiv:** [2306.12659](https://arxiv.org/abs/2306.12659)  
**Date:** 2023-06-22  
**Summary page:** https://www.summarizepaper.com/en/paper/2306.12659v1/

## Key points

- Sentiment analysis is important for uncovering insights from financial articles, news, and social media
- Large language models (LLMs) struggle with interpreting numerical values and understanding financial context
- The authors propose an instruction tuning approach to address these issues
- The method transforms supervised financial sentiment analysis data into instruction data and fine-tunes a general-purpose LLM
- Results show advancements in financial sentiment analysis compared to other models like ChatGPT and LLaMAs
- The proposed approach outperforms other models, especially in scenarios requiring numerical understanding and contextual comprehension
- This research improves the accuracy and effectiveness of sentiment analysis in the financial domain.

## Summary

In the paper titled "Instruct-FinGPT: Financial Sentiment Analysis by Instruction Tuning of General-Purpose Large Language Models," authors Boyu Zhang, Hongyang Yang, and Xiao-Yang Liu discuss the importance of sentiment analysis in uncovering insights from financial articles, news, and social media. They emphasize that while large language models (LLMs) have impressive capabilities in financial natural language processing (NLP), they struggle to accurately interpret numerical values and understand financial context which limits their effectiveness in predicting financial sentiment. To address these issues, the authors propose a simple yet effective instruction tuning approach. This method transforms a small portion of supervised financial sentiment analysis data into instruction data and fine-tunes a general-purpose LLM using this technique. The results show remarkable advancements in financial sentiment analysis compared to state-of-the-art supervised sentiment analysis models and widely used LLMs like ChatGPT and LLaMAs. The experiment demonstrates that the proposed approach outperforms other models, particularly in scenarios where numerical understanding and contextual comprehension are essential. This research contributes to improving the accuracy and effectiveness of sentiment analysis in the financial domain.