MeChat is a Chinese mental health support chatbot developed using the SMILE methodology to expand single-turn dialogues into multi-turn conversations. It aims to address accessibility issues in mental healthcare by providing an AI-driven virtual counselor, benefiting individuals seeking support and researchers in the field.

How It Works

The project utilizes a technique called SMILE (Single-turn to Multi-turn Inclusive Language Expansion), which leverages ChatGPT to transform existing single-turn dialogues into more natural, multi-turn conversations. This approach enhances the realism and effectiveness of the chatbot by simulating human-like interactions between a client and a counselor, thereby improving the model's performance in mental health support scenarios.

Quick Start & Requirements

• Install dependencies: pip install -r requirements.txt
• Run interactive file: python MeChat.py
• Requires a single GPU with >= 20GB VRAM.
• Model available at: https://huggingface.co/qiuhuachuan/MeChat

Highlighted Details

• Accepted to EMNLP 2024 Findings.
• Trained on a 55k multi-turn dialogue dataset (SMILECHAT) generated using the SMILE method.
• Fine-tuned from ChatGLM2-6B using LoRA.
• Provides a comprehensive data conversion script and training entry point.

Maintenance & Community

• The project is associated with other related works: interactive-agents and PsyChat.
• Citation details are provided for academic referencing.

Licensing & Compatibility

• The project's license is not explicitly stated in the README.
• The model and data are publicly available for research purposes.

Limitations & Caveats

The project acknowledges that generated dialogues may have shorter turns compared to real counseling data and emphasizes the need for continued focus on model safety, including suicide intervention, handling sensitive information, and avoiding misinformation. The disclaimer explicitly states the chatbot is not a substitute for professional medical advice.