CoI-Psychotherapy

This is the official implementation of the IEEE ICHI 2024 paper:

Chain-of-Interaction: Enhancing Large Language Models for Psychiatric Behavior Understanding by Dyadic Contexts

Dependence Installation

git clone git@github.com:trust-nlp/CoI-Psychotherapy.git
conda create -n [YOUR_ENV] python=3.10
conda activate [YOUR_ENV]
cd CoI-Psychotherapy
pip install -r requirements.txt

Data Preparation

Please refer to data/README.md for building your own dataset.

And modify your dataset path in src/llmtuner/eval/MMCBevaluator.py.

Evaluate frozen LLMs

CUDA_VISIBLE_DEVICES=0 python src/evaluate.py \
    --model_name_or_path path_to_LLM \
    --template vanilla \
    --task mmlu \
    --split test \
    --lang [Prompting method] \
    --n_shot 0 \
    --batch_size 4

Citation

If this work is helpful, please kindly cite as:

@article{han2024chainofinteraction,
      title={Chain-of-Interaction: Enhancing Large Language Models for Psychiatric Behavior Understanding by Dyadic Contexts}, 
      author={Guangzeng Han and Weisi Liu and Xiaolei Huang and Brian Borsari},
      journal={arXiv preprint arXiv:2403.13786},
      year={2024},
      eprint={2403.13786},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2403.13786}
}

Acknowledgements

This project is a fork of hiyouga/LLaMA-Factory. We are grateful for their work and contributions to the LLM community. This fork aims to evaluate the performance of multiple prompting methods for automated coding of motivational interviews. Please visit the original repository to learn more about the project and support the original creators.