PsyBP: A Method for Constructing Process-Standardized LLMs for Psychological Counseling

The application of large language models (LLMs) in the field of psychological counseling has become a research hotspot. However, existing methods for constructing psychological counseling LLMs lack a clear and professional framework for counseling processes. Therefore, the constructed LLMs lack structured dialogue guidance when facing various psychological counseling tasks, leading to chaotic or inefficient counseling processes. To address this issue, a novel framework for constructing psychological counseling LLMs, called the Psychological Counseling Blueprint (PsyBP), was proposed in this study. This framework integrates multiple psychological counseling theories and methods to establish a well-structured and phased psychological counseling dialogue process. Subsequently, based on this process, targeted optimization and rewriting of publicly available multi-turn emotional support dialogue datasets were performed, resulting in the creation of the PsyBP dataset with clear boundaries for each counseling stage. Using this dataset, a general-purpose LLM was fine-tuned, successfully developing the PsyBPLLM specifically suited for psychological counseling tasks. Experimental evaluations demonstrated that both the PsyBP dataset and PsyBPLLM exhibited superior performance compared with existing baseline datasets and models, validating the effectiveness and applicability of the PsyBP framework in practical psychological counseling scenarios.The application of large language models (LLMs) in the field of psychological counseling has become a research hotspot. However, existing methods for constructing psychological counseling LLMs lack a clear and professional framework for counseling processes. Therefore, the constructed LLMs lack structured dialogue guidance when facing various psychological counseling tasks, leading to chaotic or inefficient counseling processes. To address this issue, a novel framework for constructing psychological counseling LLMs, called the Psychological Counseling Blueprint (PsyBP), was proposed in this study. This framework integrates multiple psychological counseling theories and methods to establish a well-structured and phased psychological counseling dialogue process. Subsequently, based on this process, targeted optimization and rewriting of publicly available multi-turn emotional support dialogue datasets were performed, resulting in the creation of the PsyBP dataset with clear boundaries for each counseling stage. Using this dataset, a general-purpose LLM was fine-tuned, successfully developing the PsyBPLLM specifically suited for psychological counseling tasks. Experimental evaluations demonstrated that both the PsyBP dataset and PsyBPLLM exhibited superior performance compared with existing baseline datasets and models, validating the effectiveness and applicability of the PsyBP framework in practical psychological counseling scenarios.