Repository files navigation

KokoroChat is the largest human-collected Japanese psychological counseling dialogue dataset to date (as of June 2025). It was created through role-playing between trained counselors and includes rich, long-form dialogues and detailed client feedback on counseling quality. The dataset supports research on empathetic response generation, dialogue evaluation, and mental health-oriented language modeling.

This work has beenaccepted to the main conference of ACL 2025.📄View Paper (arXiv)

• 6,589 dialogues, collected between 2020 and 2024
• Avg. 91.2 utterances per dialogue
• 480 trained counselors simulating online text-based counseling sessions
• 20-dimension Likert-scale client feedback for every session
• Broad topic coverage: mental health, school, family, workplace, romantic issues, etc.

Each sample contains:

• A full counseling dialogue with role labels (counselor / client) and message timestamps
• Structured client feedback on 20 dimensions (0–5 Likert scale)
• Flags for ethical concern checks (optional)
• Predicted topic label (automatically annotated by GPT-4o-mini)

👉 See thekokorochat_dialogues folder for the complete dataset.

You can also access our full dataset and fine-tuned models via Hugging Face:

• 📁Dataset:KokoroChat-dataset

We fine-tuned three counseling dialogue models based onLlama-3.1-Swallow-8B-Instruct-v0.3, using different subsets of the KokoroChat dataset filtered by client feedback score:

• 🔵Llama-3.1-KokoroChat-Low: Fine-tuned on 3,870 dialogues with feedback scores< 70
• 🟢Llama-3.1-KokoroChat-High: Fine-tuned on 2,601 dialogues with feedback scoresbetween 70 and 98
• ⚫Llama-3.1-KokoroChat-Full: Fine-tuned on 6,471 dialogues with feedback scores≤ 98

If you use this dataset, please cite the following paper:

@inproceedings{qi2025kokorochat,title     ={KokoroChat: A Japanese Psychological Counseling Dialogue Dataset Collected via Role-Playing by Trained Counselors},author    ={Zhiyang Qi and Takumasa Kaneko and Keiko Takamizo and Mariko Ukiyo and Michimasa Inaba},booktitle ={Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics},year      ={2025},url       ={https://github.com/UEC-InabaLab/KokoroChat}}

KokoroChat is released under theCreative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0) license.