PsycoLLM - A Chinese Psychological Language Model Developed by Hefei University of Technology

What is PsycoLLM?

PsycoLLM, developed by the School of Computer Science and Information Engineering at Hefei University of Technology, is a large-scale Chinese psychological language model trained on high-quality psychological datasets. It enhances the understanding and assessment of mental health issues. The model's dataset covers single-turn question answering, multi-turn dialogue, and knowledge-based question answering, employing innovative data generation and optimization processes to ensure data authenticity and applicability. PsycoLLM performs exceptionally well in multi-dimensional psychological benchmark tests, including professional ethics, theoretical knowledge, and case analysis. Compared to other models, it demonstrates stronger performance and more accurate judgment capabilities, providing strong technical support for research and applications in the field of mental health.

Main functions of PsycoLLM

• Understanding and answering psychological problemsTo accurately understand the psychological problems raised by users and provide professional and accurate answers to help users obtain psychological support and guidance.
• Multi-turn dialogue interactionIt supports multi-round dialogues with users, and based on continuous question-and-answer interactions, it can gain a deeper understanding of users' psychological state and needs, and provide more targeted suggestions and assistance.
• Popularization and education of psychological knowledgeUsing a rich database of psychological knowledge, we popularize mental health knowledge among users, and improve their awareness of psychological problems and their ability to self-regulate.
• Emotion Recognition and SupportIt identifies users' emotional states, such as anxiety and depression, and provides corresponding emotional support and comfort to help users alleviate emotional distress.
• Mental health assessment and recommendationsThe system conducts a preliminary assessment of the user's mental health and provides corresponding suggestions based on the assessment results, such as seeking professional psychological counseling and engaging in self-regulation.

PsycoLLM Technical Principles

• Training on high-quality datasetsPsycoLLM is trained on high-quality psychological datasets, which include various types of data such as single-turn question answering, multi-turn dialogue, and knowledge-based question answering. These datasets cover a wealth of psychological knowledge and real-world psychological counseling scenarios, enabling the model to learn professional psychological knowledge and conversational skills.
• Multi-step data generation and optimization processThe process of generating multi-turn dialogue data involves a multi-step workflow, including multi-turn question-and-answer generation, evidence assessment, and dialogue optimization. First, preliminary multi-turn dialogues are generated. Then, it is determined whether each answer in the dialogue is supported by evidence. Finally, the dialogues are optimized to improve their coherence, authenticity, and applicability.
• Supervisory fine-tuningBuilding upon the pre-trained model, supervised fine-tuning further enhances the model's performance in the psychology field. During fine-tuning, the model is trained using high-quality psychology datasets to better understand and generate psychology-related texts.
• Transformer architectureBased on the Transformer architecture as the core model structure, it uses a self-attention mechanism to capture long-distance dependencies in text, achieving efficient understanding and generation of text.

PsycoLLM project address

• GitHub repository:https://github.com/MACLAB-HFUT/PsycoLLM
• HuggingFace model library:https://huggingface.co/MACLAB-HFUT/PsycoLLM
• arXiv technical paper:https://arxiv.org/pdf/2407.05721

Application scenarios of PsycoLLM

• Personal mental health supportWhen users are feeling down, they can talk to PsycoLLM to get emotional support and adjustment suggestions to help relieve stress and restore emotional balance.
• Psychological counselingUsers describe their psychological problems before psychological counseling, generating a pre-assessment report to provide counselors with reference information and improve counseling efficiency.
• Student mental health educationPsycoLLM is used in mental health courses to supplement teaching, explain psychological knowledge, help students understand and master skills such as emotion management, and improve mental health literacy.
• Community mental health servicesCommunity residents receive psychological support and counseling services to resolve psychological problems in their lives and promote community harmony.