KokoroChat: A Japanese Psychological Counseling Dialogue Dataset Collected via Role-Playing by Trained Counselors

TL;DR

KokoroChat is a high-quality Japanese psychological counseling dialogue dataset created through role-playing by trained counselors, which enhances LLM response quality and authenticity while addressing privacy concerns.

Contribution

This paper introduces KokoroChat, a novel dataset built via role-playing that improves the authenticity and diversity of counseling dialogues for training language models.

Findings

Fine-tuning LLMs with KokoroChat enhances response quality.

The dataset improves automatic evaluation metrics.

Role-playing ensures high-quality, privacy-preserving data.

Abstract

Generating psychological counseling responses with language models relies heavily on high-quality datasets. Crowdsourced data collection methods require strict worker training, and data from real-world counseling environments may raise privacy and ethical concerns. While recent studies have explored using large language models (LLMs) to augment psychological counseling dialogue datasets, the resulting data often suffers from limited diversity and authenticity. To address these limitations, this study adopts a role-playing approach where trained counselors simulate counselor-client interactions, ensuring high-quality dialogues while mitigating privacy risks. Using this method, we construct KokoroChat, a Japanese psychological counseling dialogue dataset comprising 6,589 long-form dialogues, each accompanied by comprehensive client feedback. Experimental results demonstrate that fine-tuning open-source LLMs with KokoroChat improves both the quality of generated counseling responses and the automatic evaluation of counseling dialogues. The KokoroChat dataset is available at https://github.com/UEC-InabaLab/KokoroChat.