A virtual psychological supervisor construction method based on multi-agent cooperation

Abstract

This invention discloses a method for constructing a virtual psychological supervisor based on multi-agent collaboration, belonging to the fields of digital mental health training and artificial intelligence technology. By constructing a multi-agent collaborative framework of "error-prone counselor agent - error-sensitive patient agent - error-correcting supervisor agent," and combining iterative optimization mechanisms guided by validators, high-quality "consultation dialogue - supervision feedback" aligned data pairs are generated. Based on data optimization of a large language model, a virtual psychological supervisor with error recognition, professional feedback generation, and skills assessment capabilities is ultimately constructed. This invention solves the problems of insufficient training data, inadequate feedback professionalism, and subjective assessment standards in traditional psychological counseling supervision, achieving safe, efficient, standardized, and scalable psychological counselor skills supervision, significantly improving supervision accuracy and training efficiency.

Description

Technical Field

[0001] This invention relates to the fields of digital mental health training and artificial intelligence technology, and in particular to a method for constructing a virtual mental health supervisor based on multi-agent collaboration. Background Technology

[0002] Traditional training for psychotherapists relies primarily on theoretical learning, limited clinical practice, and personal guidance from senior supervisors. This model faces the following bottlenecks in practical application: (1) Lack of standardization of case resources: Due to the strict privacy protection of actual patient cases, it is difficult to obtain and use high-quality training data in compliance with regulations, making it difficult for novice counselors to provide them with a consistent and representative training experience; (2) Lack of real-time supervision mechanism: In actual dialogue practice, novice counselors often cannot obtain timely professional feedback, making it difficult for them to identify and correct complex behavioral errors such as gender stereotypes, transference / countertransference, and violations of professional ethics in a timely manner. (3) The evaluation criteria are highly subjective: the quality of traditional training depends heavily on the personal experience of the instructors and lacks objective and quantitative evaluation indicators, making it difficult to scientifically evaluate the specific improvements of consultants in core skills such as empathy, listening and goal setting. (4) High training costs and limited efficiency: Senior supervisors are extremely limited and costly, making it difficult to promote high-quality training on a large scale. (5) Existing AI tools have limited functionality: Current auxiliary tools focus more on understanding patients’ emotions, but lack closed-loop support for counselors’ behavioral skills training, error identification and improvement strategies in complex dialogues.

[0003] Furthermore, due to the diversity of consultation methods, establishing unified evaluation criteria is extremely difficult. While large language models have shown auxiliary potential, their supervisory accuracy in professional fields is limited by the lack of highly aligned training data for "consultation dialogue-supervisory feedback". Summary of the Invention

[0004] The purpose of this invention is to propose a method for constructing a virtual psychological supervisor based on multi-agent collaboration to solve the problems mentioned in the background art: to solve the problem of scarce training data through a multi-agent collaboration framework, and to construct a safe and controllable virtual environment to systematically improve the professional skills of psychological counselors.

[0005] To achieve the above objectives, this invention provides a method for constructing a virtual psychological supervisor based on multi-agent collaboration, comprising the following steps: Step S1: Construct a virtual psychological supervisor platform, including a virtual patient simulation module, a dialogue interaction module, a supervision and feedback generation module, a data generation module, and a training and evaluation module; Step S2: The virtual patient simulation module constructs a virtual patient with a specific cognitive structure; the supervision and feedback generation module presets common behavioral guidelines for psychological counselors. Step S3: Use the data generation module to train the supervision and feedback generation module; Step S4: Connect the trained supervision and feedback generation module, virtual patient simulation module, and dialogue interaction module. The virtual patient and the psychologist will have a consultation dialogue. The dialogue interaction module provides a consultation dialogue interface between the psychologist and the virtual patient, supports natural language interaction between the psychologist and the virtual patient, and supports text and voice interaction. The dialogue interaction module can receive the psychologist's input and the virtual patient's response in real time and transmit them synchronously to the supervision and feedback generation module. Step S5: Integration of Training and Evaluation Modules: Identify and record the error behaviors of the consultation dialogue in the related dialogue interaction module and the supervision and feedback generation module, and generate an evaluation report.

[0006] Preferably, in step S1, the virtual psychological supervisor platform specifically includes: Virtual patient simulation module: Constructs virtual patients with specific cognitive structures based on a large language model; Dialogue interaction module: Used for psychological counselors to conduct multi-round real-time counseling dialogues with virtual patients, and supports contextual understanding; Supervision and Feedback Generation Module: Used to identify erroneous behaviors of psychotherapists from consultation dialogues and generate targeted supervisory feedback based on erroneous behavior guidelines; Data generation module: Employs a multi-agent collaboration framework to generate aligned data pairs of "consultation dialogue - supervision feedback" to improve the performance of the supervisor model; The training assessment module is used to quantitatively evaluate the performance of psychotherapists in core skills and generate assessment reports.

[0007] Preferably, in step S2, the specific cognitive structure is modeled through five dimensions: core beliefs, intermediate beliefs, automatic thoughts, emotional states, and coping strategies, integrating 106 clinically validated patient cognitive models, including those for depression, anxiety, and post-traumatic stress disorder.

[0008] Preferably, in step S2, the code of conduct for erroneous behaviors includes 15 types of erroneous behaviors, including: gender stereotypes, cultural insensitivity, ignoring patient emotions, violating ethical standards, and empathy issues.

[0009] Preferably, in step S3, the data generation module includes three agents: an error-prone counselor agent, an error-sensitive patient agent, and an error-correction supervision agent. The collaboration logic of the three agents is as follows: First, the error-prone consultant agent simulates preset error behaviors and interacts with the error-sensitive patient agent in multiple rounds to generate original consultation dialogues containing error behaviors. Then, the error-correcting supervisory agent generates critical supervisory feedback. Finally, a validator-guided optimization mechanism is introduced to evaluate and iteratively optimize data quality through the validator model, generating "consultation dialogue-supervisory feedback" aligned data pairs that meet the requirements of diversity, balance, and professionalism. The generated aligned data is used as input to the supervision and feedback generation module. Through fine-tuning or cue word learning, the accuracy of the supervision and feedback generation module in identifying erroneous behaviors and the professionalism of structured supervision feedback are optimized.

[0010] Preferably, the validator bootstrapping optimization mechanism includes the following steps: Evaluation phase: The quality of the initial aligned data pairs is scored using a validator model; Optimization phase: If the score is lower than the preset threshold, guide the multi-agent collaborative interaction to regenerate aligned data pairs until the quality requirements are met.

[0011] Preferably, in step S5, the training assessment module quantitatively assesses the core skills of the psychological counselor based on the CASES-R scale and calculates the improvement of individual skills and the comprehensive assessment score. The core skills specifically include listening, emotional response, interpretation skills, immediate response, goal setting, focus and participation, restatement skills, and direct guidance.

[0012] Preferably, in step S5, the improvement rate of a single skill Performance was quantified through multiple rounds of consultations and dialogues, and a comprehensive evaluation score was obtained. It can be obtained through the following formula: ; in These are the weighting coefficients for each skill. The assessment report mainly includes: a skills improvement trend chart, statistics on the improvement of incorrect behaviors, and a personalized follow-up training plan.

[0013] Therefore, the present invention employs the above-described method for constructing a virtual psychological supervisor based on multi-agent collaboration, which has the following advantages: (1) Safety and compliance: It provides a safe and controllable training environment, which effectively avoids the ethical and medical risks that may arise when novice counselors directly face real patients; (2) Data closed loop and continuous performance enhancement: The problem of scarcity of high-quality supervisory data is solved by using a multi-agent framework, and the discrimination ability and feedback accuracy of the large model are continuously optimized by using automatically generated aligned data; (3) Standardization and consistency: Based on evidence-based medicine guidelines and standard scales, the scientific nature and uniformity of training content and assessment standards are ensured, eliminating the subjective interference of traditional manual supervision; (4) Efficient and real-time skills enhancement: The instant feedback mechanism allows users to correct mistakes immediately in practice, which significantly improves the speed of skills internalization and training efficiency; (5) High scalability and low cost: The system supports cloud or local deployment and can provide standardized professional training for a large number of practitioners at extremely low marginal cost, effectively alleviating the problem of talent shortage.

[0014] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0015] Figure 1 This is an overall architecture diagram of a virtual psychological supervisor construction method based on multi-agent collaboration mentioned in an embodiment of the present invention; Figure 2 This is a schematic diagram of the multi-agent data generation framework in a method for constructing a virtual psychological supervisor based on multi-agent collaboration mentioned in an embodiment of the present invention. Figure 3 This is a flowchart of the supervision and feedback generation module in a method for constructing a virtual psychological supervisor based on multi-agent collaboration mentioned in an embodiment of the present invention. Figure 4 This is a diagram illustrating the full-process implementation stages of a virtual psychological supervisor construction method based on multi-agent collaboration mentioned in this embodiment of the invention. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0017] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0018] Example like Figure 1-4 As shown, this embodiment provides a method for constructing a virtual psychological supervisor based on multi-agent collaboration, including the following steps: Step S1: Construct a virtual psychological supervisor platform, including a virtual patient simulation module, a dialogue interaction module, a supervision and feedback generation module, a data generation module, and a training and evaluation module; Virtual Patient Simulation Module: Used to construct virtual patients with specific cognitive structures based on large language models; Dialogue interaction module: Used for psychological counselors to conduct multi-round real-time counseling dialogues with virtual patients, and supports contextual understanding; The supervision and feedback generation module is used to identify the counselor's erroneous behaviors from the dialogue history and generate targeted supervisory feedback based on the erroneous behavior guidelines. Data generation module: Employs a multi-agent collaboration framework to generate aligned data pairs of "consultation dialogue - supervision feedback" to improve the performance of the supervision model; The training assessment module is used to quantitatively assess the performance of psychotherapists in core skills using the CASES-R scale. Step S2: The virtual patient simulation module, based on the Big Language Model and Cognitive Behavioral Therapy (CBT) theory, constructs a virtual patient with a specific cognitive structure. The Big Language Model uses the existing open-source Qwen3-8B-Instruct model. CBT theory is a mature existing psychological counseling theory, specifically including: a three-layer cognitive model based on CBT theory—core beliefs, intermediate beliefs, and automatic thoughts. The Qwen3-8B-Instruct model is used to learn from corpora labeled with cognitive characteristics of specific psychological problems (such as anxiety and depression), enabling the virtual patient to possess dialogue logic and emotional feedback in a psychological counseling scenario. The specific cognitive structure is modeled through five dimensions: core beliefs, intermediate beliefs, automatic thoughts, emotional state, and coping strategies, integrating 106 clinically validated patient cognitive models covering depression, anxiety, and post-traumatic stress disorder. The monitoring and feedback generation module presets common behavioral guidelines for psychological counselors, including 15 types of erroneous behaviors: gender stereotypes, cultural insensitivity, ignoring patient emotions, violating ethical standards, and transference issues. Step S3: Data Generation and Model Enhancement Based on Multi-Agent Collaboration: Through multi-agent collaboration in the data generation module, the multi-agents include an error-prone counselor agent, an error-sensitive patient agent, and an error-correction supervisor agent. First, the error-prone counselor agent simulates preset error behaviors and interacts with the error-sensitive patient agent in multiple rounds to generate original consultation dialogues containing error behaviors. Then, the error-correction supervisor agent generates critical supervisory feedback. Finally, a validator-guided optimization mechanism is introduced. The validator model evaluates and iteratively optimizes the data quality to generate "consultation dialogue-supervisory feedback" aligned data pairs that meet the requirements of diversity, balance, and professionalism. The multi-agent collaboration logic of the data generation module is as follows: Error-prone counselor agent: Based on the error behavior guidelines, it simulates the tendency of a psychological counselor to exhibit error behaviors in counseling dialogues and generates counseling dialogues containing error behaviors; Error-sensitive patient agent: Produces logical feedback responses to the erroneous behaviors of the error-prone counselor agent; Error-correcting and supervisory intelligent agent: used to accurately locate erroneous behaviors in consultation dialogues and generate structured supervisory feedback; The validator bootstrapping optimization mechanism includes the following steps: Evaluation phase: The initial aligned data pairs are scored in terms of quality from three dimensions: diversity, balance, and professionalism, using a validator model; Optimization phase: If the score is lower than the preset threshold, guide the multi-agent collaborative interaction to regenerate aligned data pairs until the quality requirements are met; The generated aligned data is used as input to the supervision and feedback generation module. Through fine-tuning or cue word learning, the accuracy of the supervision and feedback generation module in identifying erroneous behaviors and the professionalism of structured supervised feedback are optimized. The fine-tuning or cue word learning methods specifically include: using supervised training (SFT) to optimize the model, specifically using LoRA (Low-Rank Adaptation) technology to efficiently fine-tune the parameters of the module's built-in basic model, updating only the low-rank adaptation matrix parameters to reduce training costs; at the same time, cue word learning is used to guide the model to focus on the tasks of erroneous behavior identification and structured supervised feedback generation. Step S4: Dialogue Interaction Module Integration: Data exchange is established between the trained supervision and feedback generation module, virtual patient simulation module, and dialogue interaction module. The virtual patient engages in consultation dialogue with the psychologist. The dialogue interaction module provides an interface for consultation dialogue between the psychologist and the virtual patient, supporting natural language interaction, text and voice interaction, and maintaining the dialogue context. The dialogue interaction module can receive the psychologist's input and the virtual patient's response in real time and transmit them synchronously to the supervision and feedback generation module. The supervision and feedback generation module analyzes the consultation dialogue content in real time, uses a chain-reasoning architecture to identify erroneous behaviors in the consultation dialogue, and generates structured supervisory feedback. The structured supervisory feedback uses a four-tuple structure, including error category, error location, error analysis, and improvement suggestions. The implementation process of the chain-reasoning architecture is as follows: first, determine whether there are erroneous behaviors in the consultation dialogue; then, accurately identify the utterance fragments containing erroneous behaviors; next, classify the erroneous behaviors into error behavior criteria; and finally, generate structured supervisory feedback. Step S5: Integration of Training and Assessment Modules: The historical data of consultation dialogues from the linked dialogue interaction module and the identification records of erroneous behaviors from the supervision and feedback generation module are used by the training and assessment module to quantitatively assess the core skills of the psychological counselor based on the CASES-R scale. This calculates the improvement rate of individual skills and the overall assessment score, generates an assessment report, and completes the training loop for virtual psychological supervisors. The CASES-R scale quantitatively assesses the core skills of the psychological counselor, specifically including listening, emotional response, interpretation skills, immediate response, goal setting, focus and participation, restatement skills, and direct guidance. Individual skill improvement Performance was quantified through multiple rounds of consultations and dialogues, and a comprehensive evaluation score was obtained. It can be obtained through the following formula: ; in These are the weighting coefficients for each skill. The assessment report mainly includes: a skills improvement trend chart, statistics on the improvement of incorrect behaviors, and a personalized follow-up training plan.

[0019] A specific implementation process is as follows: like Figure 1-4 As shown, this embodiment provides a specific implementation of a method for constructing a virtual psychological supervisor based on multi-agent collaboration. Data exchange is achieved through a standardized API interface, building a complete training ecosystem. The core logic and agent functions of the system are implemented using a Large Language Model (LLM, such as Qwen3-8B-Instruct) and rely on carefully designed prompts and a structured instruction set.

[0020] The specific implementation process is as follows: (a) System initialization and knowledge base loading: The system sets up a phased training framework and loads a preset clinical knowledge base from the data storage layer.

[0021] (1) Cognitive model library: loaded with 106 patient cognitive models validated by clinical experts. It covers disease types such as depression, anxiety, and post-traumatic stress disorder (PTSD).

[0022] (2) Set of Error Behaviors: Loads a set of 15 common mistakes made by novice consultants. Each element Defined as a quadruple: These represent the error category, behavior description, correction strategy, and typical example, respectively.

[0023] (ii) Virtual patient simulation and background construction: The virtual patient simulation module is responsible for constructing virtual individuals with pathological characteristics.

[0024] (1) Analysis of cognitive model elements: each cognitive model It includes five core dimensions: core beliefs, intermediate beliefs, automatic thoughts, emotions, and coping strategies.

[0025] (2) Language behavior modeling: The virtual patient in the first Response to the round of dialogue Implemented based on the following generator function: in, Indicates the preceding The history of dialogue between wheels For the counselor in the first The words of the wheel, For the currently selected cognitive model, Provide the patient's personal background information (age, gender, occupation, etc.). These are the generation parameters for the large language model.

[0026] (III) Dialogue Interaction and Real-time Monitoring: The dialogue interaction module provides an interface similar to instant messaging, supporting natural language interaction between counselors and virtual patients. The system adopts streaming generation technology to ensure that the average response time is controlled within 2 seconds.

[0027] (1) User input: The counselor inputs the dialogue. .

[0028] (2) Context maintenance: The system maintains the dialogue history list in real time. .

[0029] (iv) Supervision and Feedback Generation (Core Closed Loop): The supervision and feedback generation module analyzes the dialogue content in real time. The supervision task is formalized as a joint probability distribution calculation: in, Indicates the location of incorrect statements. Indicates error category classification, This indicates that feedback content has been generated.

[0030] The specific implementation process is as follows: (1) Chain-of-Track (CoT) Judgment: ① Error detection: based on Set judgment Does it contain erroneous behavior?

[0031] ② Error Location: If errors exist, accurately identify the speech segments. .

[0032] ③ Multi-classification task: Classify errors to .

[0033] (2) Structured feedback generation: Generate a feedback template based on the formula. Content: feedback It consists of three components: error location (quoting the original text), error analysis (theoretical basis), and improvement suggestions (alternative expressions).

[0034] (v) Data generation and model enhancement based on multi-agent collaboration: like Figure 2 As shown, the data generation module autonomously generates high-quality training data through a multi-agent framework to fine-tune and optimize the supervised large language model.

[0035] (1) Agent collaboration logic: Error-prone consultant agent: based on Simulate incorrect behavior to generate discourse: .

[0036] Error-sensitive patient agent: based on Product reaction: .

[0037] Error-correcting and supervisory agent: Generating annotations based on an "omniscient perspective": .

[0038] (2) Validator-guided optimization mechanism: Quality assessment of initial feedback using a validator model .like If the value falls below a set threshold, the optimization function will be activated: Generated through this mechanism The data pairs are used for continuous training of the model.

[0039] (vi) Training assessment and report generation: The training assessment module assesses eight core skills (listening, emotional response, etc.) based on the CASES-R scale.

[0040] (1) Calculation of skill improvement: Individual skill improvement Performance was quantified through multiple rounds of dialogue, and a comprehensive skills assessment score was obtained. It can be obtained through the following formula: in These are the weighting coefficients for each skill.

[0041] (2) Report output: The system automatically generates a report that includes a skills improvement trend chart, error behavior improvement statistics and personalized follow-up training plan.

[0042] (vii) Implementation of the complete training phase: such as Figure 4 As shown, the training process is divided into four stages: (1) Pre-training phase: Baseline skills assessment.

[0043] (2) Formal training phase: 15-30 minute structured dialogue to obtain immediate feedback .

[0044] (3) Reflection and summary stage: error pattern identification and knowledge consolidation.

[0045] (4) Evaluation and feedback stage: Generate the training plan for the next stage.

[0046] It is worth noting that the contents not described in detail in this invention (such as the specific fine-tuning algorithm of LLM, the underlying protocol of API interface, etc.) are all existing technologies and are well known to those skilled in the art.

[0047] The method and system provided in this embodiment solve the problem of scarce professional supervision data by generating a closed loop of multi-agent data, and significantly improve the professionalism and accuracy of AI supervision.

[0048] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for constructing a virtual psychological supervisor based on multi-agent collaboration, characterized in that, Includes the following steps: Step S1: Construct a virtual psychological supervisor platform, including a virtual patient simulation module, a dialogue interaction module, a supervision and feedback generation module, a data generation module, and a training and evaluation module; Step S2: The virtual patient simulation module constructs a virtual patient with a specific cognitive structure; the supervision and feedback generation module presets common behavioral guidelines for psychological counselors. Step S3: Use the data generation module to train the supervision and feedback generation module; Step S4: Connect the training supervision and feedback generation module, virtual patient simulation module and dialogue interaction module to exchange data. The virtual patient and the psychologist conduct consultation dialogue. The dialogue interaction module provides a consultation dialogue interface between the psychologist and the virtual patient, supports natural language interaction between the psychologist and the virtual patient, and supports text and voice interaction. The dialogue interaction module can receive input from the psychological counselor and responses from the virtual patient in real time, and transmit them synchronously to the supervision and feedback generation module. Step S5: Integration of Training and Evaluation Modules: Identify and record the error behaviors of the consultation dialogue in the related dialogue interaction module and the supervision and feedback generation module, and generate an evaluation report.

2. The method for constructing a virtual psychological supervisor based on multi-agent collaboration according to claim 1, characterized in that: In step S1, the virtual psychological supervisor platform specifically includes: Virtual patient simulation module: Constructs virtual patients with specific cognitive structures based on a large language model; Dialogue interaction module: used for multi-round real-time consultation dialogues between psychological counselors and virtual patients, and supports contextual understanding; Supervision and Feedback Generation Module: Used to identify erroneous behaviors of psychotherapists from consultation dialogues and generate targeted supervisory feedback based on erroneous behavior guidelines; Data generation module: Employs a multi-agent collaboration framework to generate aligned data pairs of "consultation dialogue - supervision feedback" to improve the performance of the supervisor model; The training assessment module is used to quantitatively evaluate the performance of psychotherapists in core skills and generate assessment reports.

3. The method for constructing a virtual psychological supervisor based on multi-agent collaboration according to claim 1, characterized in that: In step S2, specific cognitive structures are modeled through five dimensions: core beliefs, intermediate beliefs, automatic thoughts, emotional states, and coping strategies. This integrates 106 clinically validated patient cognitive models, including those for depression, anxiety, and post-traumatic stress disorder.

4. The method for constructing a virtual psychological supervisor based on multi-agent collaboration according to claim 1, characterized in that: In step S2, the guidelines for wrongdoing include 15 types of wrongdoing, including: gender stereotypes, cultural insensitivity, ignoring patient feelings, violating ethical standards, and empathy issues.

5. The method for constructing a virtual psychological supervisor based on multi-agent collaboration according to claim 1, characterized in that: In step S3, the data generation module includes three agents: an error-prone counselor agent, an error-sensitive patient agent, and an error-correction supervisor agent. The collaboration logic of the three agents is as follows: First, the error-prone consultant agent simulates preset error behaviors and interacts with the error-sensitive patient agent in multiple rounds to generate original consultation dialogues containing error behaviors. Then, the error-correcting supervisory agent generates critical supervisory feedback. Finally, a validator-guided optimization mechanism is introduced to evaluate and iteratively optimize data quality through the validator model, generating "consultation dialogue-supervisory feedback" aligned data pairs that meet the requirements of diversity, balance, and professionalism. The generated aligned data is used as input to the supervision and feedback generation module. Through fine-tuning or cue word learning, the accuracy of the supervision and feedback generation module in identifying erroneous behaviors and the professionalism of structured supervision feedback are optimized.

6. The method for constructing a virtual psychological supervisor based on multi-agent collaboration according to claim 5, characterized in that: The validator bootstrapping optimization mechanism includes the following steps: Evaluation phase: The quality of the initial aligned data pairs is scored using a validator model; Optimization phase: If the score is lower than the preset threshold, guide the multi-agent collaborative interaction to regenerate aligned data pairs until the quality requirements are met.

7. The method for constructing a virtual psychological supervisor based on multi-agent collaboration according to claim 1, characterized in that: In step S5, the training assessment module uses the CASES-R scale to quantitatively assess the core skills of the counselor and calculates the improvement of individual skills and the overall assessment score. The core skills specifically include listening, emotional response, interpretation skills, immediate response, goal setting, focus and participation, restatement skills, and direct guidance.

8. The method for constructing a virtual psychological supervisor based on multi-agent collaboration according to claim 7, characterized in that: In step S5, the improvement rate of a single skill. Performance was quantified through multiple rounds of consultations and dialogues, and a comprehensive evaluation score was obtained. It can be obtained through the following formula: ； in These are the weighting coefficients for each skill. The assessment report mainly includes: a skills improvement trend chart, statistics on the improvement of incorrect behaviors, and a personalized follow-up training plan.

Images