The FinChina SA dataset and code from FinLLM@IJCAI'23 paper:

Chinese Fine-Grained Financial Sentiment Analysis with Large Language Models

https://arxiv.org/abs/2306.14096

First, the data needs to be processed into a conversational form like this:

{
    "conversations":[
        {
            "from":"human",
            "value":"question"
        },
        {
            "from":"gpt",
            "value":"answer"
        }
    ]
}

The code is in the src folder and you need to modify the configuration in the shell script.

To fine-tune the model:

sh sft.sh

To run inference:

sh run_infer.sh

Our experiments is based on the following repo:

• Longformer-chinese
• Chinese-LLaMA-Alpaca
• LLMZoo
• ChatGLM
• MedicalGPT

Thanks for their efforts to make the code and model weights public.

If you find this project useful, please cite it using the following format

@article{lan2023chinese,
  title={Chinese Fine-Grained Financial Sentiment Analysis with Large Language Models},
  author={Lan, Yinyu and Wu, Yanru and Xu, Wang and Feng, Weiqiang and Zhang, Youhao},
  journal={arXiv preprint arXiv:2306.14096},
  year={2023}
}