Digital Twin of a Psychotherapist Dataset (PsyDTCorpus)

Dataset Overview

• Dataset Name: PsyDTCorpus
• Source: Based on real multi‑turn counseling cases of a specific therapist, synthesized via a digital‑twin data generation framework.
• Scale:
  • Training set: 4,760 dialogues, total 86,054 turns, average 18 turns per dialogue.
  • Test set: 240 dialogues, total 4,311 turns, average 18 turns per dialogue.
• Format: OpenAI format.
• Topic Distribution: The dataset covers various topics; the distribution is shown in the topic distribution chart.

Data Generation Method

• Framework: Using a small number of real counseling cases, combined with the Big Five personality analysis and LLM summarization capabilities, to generate multi‑turn dialogues that reflect the therapist's language style and counseling techniques.
• Generation Scale:
  • Single‑turn counseling database size: 5,000.
  • Specific therapist case count: 12 (typically not more than 20).

Dataset Download

• Download Methods:
  1. Using git-lfs:

     cd <project_path>/data
     git lfs install
     git clone https://www.modelscope.cn/datasets/YIRONGCHEN/PsyDTCorpus.git
     

  2. Using modelscope download:

     cd <project_path>/data
     mkdir PsyDTCorpus
     modelscope download --dataset YIRONGCHEN/PsyDTCorpus --include *
     

Sample Entry

{
    "id": 0,
    "normalizedTag": "婚恋",
    "messages": [
        {
            "role": "system",
            "content": "You are a psychotherapist proficient in Rational Emotive Behavior Therapy (REBT), capable of providing professional guidance and support to alleviate clients' negative emotions and behavioral responses, helping them achieve personal growth and mental health. REBT includes several stages, listed below with brief descriptions of each stage..."
        },
        ...
    ]
}