Intelligent psychological counseling skill training and feedback platform based on AI large model

Abstract

The invention provides an intelligent psychological counseling skill training and feedback platform based on an AI large model. The platform comprises a case library module which stores a plurality of psychological counseling training cases; the user interaction module is used for a user to select training cases from the case library module; the AI visitor simulation module is constructed on the basis of a large language model subjected to psychological corpus adjustment, and is used for generating a dialogue response according to a virtual visitor in a case after the user selects the psychological consultation training case; the skill evaluation module is used for analyzing whole-course records of multiple rounds of conversations between the user and the virtual visitor through the user interaction module, performing multi-dimensional quantitative evaluation on consultation skills displayed by the user, and generating an evaluation result; and the feedback report generation module is used for generating a structured skill improvement suggestion report based on the evaluation result, and presenting the structured skill improvement suggestion report to the user through the user interaction module, so that the difficulty that scale, individuation and high-quality actual combat are difficult to consider in psychological counselor training is solved.

Description

Technical Field

[0001] This invention relates to the field of interdisciplinary technology of artificial intelligence and mental health, and in particular to an intelligent psychological counseling skills training and feedback platform based on a large AI model. Background Technology

[0002] In traditional psychotherapist training, students often face challenges such as insufficient practical experience, untimely and subjective feedback. The scarcity of real clients limits the number of cases each student can handle, making it difficult to cover diverse psychological issues and backgrounds. Simultaneously, supervision resources are strained, and the one-to-many supervision model makes it difficult to provide precise and frequent feedback, resulting in significant bottlenecks in students' counseling skill development. Furthermore, traditional training methods lack quantitative assessment of counseling skills, making it difficult to generate systematic growth data and leading to unclear development paths for students.

[0003] Existing digital training tools, such as virtual reality (VR) and standardized patients (SPs), have alleviated the problem of insufficient cases to some extent, but these methods are costly and difficult to promote on a large scale. At the same time, these tools mostly remain at the level of process rehearsal, lacking in-depth assessment of counseling skills and personalized feedback, and thus cannot meet the needs of high-quality, personalized training. Summary of the Invention

[0004] This invention provides an intelligent psychological counseling skills training and feedback platform based on an AI large model, which solves the technical problems of scarcity of real client resources, coarse granularity of supervision feedback, and strong subjectivity of assessment in existing technologies.

[0005] On the one hand, this invention provides an intelligent psychological counseling skills training and feedback platform based on an AI large-scale model, including: The case library module stores multiple psychological counseling training cases; each psychological counseling training case includes the background story, counseling issues, and behavioral patterns of a virtual client. The user interaction module is used to allow users to select training cases from the case library module and serves as an interactive interface for users to conduct multi-round dialogues with virtual visitors. The AI ​​simulated visitor module is built on a large language model adjusted with psychological corpus. It is used to dynamically generate dialogue responses that match the background, issues and behavioral patterns of the virtual visitor in the case after the user selects a psychological counseling training case. The skills assessment module is used to analyze the entire record of the user's multi-round dialogue with the virtual visitor through the user interaction module, and to conduct a multi-dimensional quantitative assessment of the consulting skills demonstrated by the user, generating assessment results. The feedback report generation module is used to generate a structured skills improvement suggestion report based on the evaluation results, and present it to the user through the user interaction module.

[0006] This invention provides an intelligent psychological counseling skills training and feedback platform based on an AI large-scale model. The case library module stores multiple psychological counseling training cases. After the user selects a psychological counseling training case, the AI ​​simulated client module dynamically generates dialogue responses based on the virtual client in that case. The skills assessment module performs multi-dimensional quantitative evaluation of the counseling skills demonstrated by the user and generates assessment results. The feedback report generation module generates a structured skills improvement suggestion report based on the assessment results. This effectively solves the dilemma in traditional psychological counselor training of balancing large-scale, personalized, and high-quality practical training. It provides psychology students and aspiring counselors with a repeatable, zero-risk, and highly realistic immersive training environment, significantly improving the efficiency and professionalism of counseling skills training, and at a relatively low cost. Attached Figure Description

[0007] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0008] Figure 1 This is a schematic diagram of the structure of the intelligent psychological counseling skills training and feedback platform based on an AI large model provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the architecture of the intelligent psychological counseling skills training and feedback platform based on an AI large model provided in an embodiment of the present invention; Figure 3 This is a flowchart illustrating the intelligent psychological counseling skills training and feedback method based on an AI large model provided in an embodiment of the present invention. Figure 4 This is a schematic diagram of the structure of the electronic device provided in an embodiment of the present invention. Detailed Implementation

[0009] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this 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 this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0010] Figure 1This is a schematic diagram of the structure of the intelligent psychological counseling skills training and feedback platform based on an AI large model provided in an embodiment of the present invention.

[0011] See Figure 1 The AI-based intelligent psychological counseling skills training and feedback platform may include a case library module, a user interaction module, an AI simulated client module, a skills assessment module, and a feedback report generation module.

[0012] The case library module stores multiple psychological counseling training cases; each psychological counseling training case includes the background story, counseling issues, and behavioral patterns of the virtual client.

[0013] Specifically, each psychological counseling training case can be a structured data entity, including a background story, counseling issues, and behavioral patterns. The background story includes the virtual client's personal resume, upbringing, social relationships, etc., providing rationale for their behavior and emotions. The counseling issues are clearly defined core psychological problems, such as academic anxiety, workplace stress, social phobia, etc., defining the main direction of counseling. The behavioral patterns include the client's typical emotional response patterns, cognitive distortions (such as catastrophic thinking), defense mechanisms, etc., used to guide the AI ​​in generating responses consistent with their personality traits. The case content is standardized based on real clinical scenarios, covering mainstream approaches such as cognitive behavioral therapy (CBT), narrative therapy, and acceptance and commitment therapy, ensuring the professionalism and diversity of the training. Each case can include six consecutive counseling sessions, forming a complete longitudinal training process to support phased skills training and assessment.

[0014] The user interaction module allows users to select training cases from the case library module and serves as an interactive interface for users to conduct multi-turn dialogues with virtual visitors.

[0015] Specifically, the user interaction module can be designed as a graphical interface or command-line tool to support user interaction with the system. For example, it can implement case selection functionality through drop-down menus or search boxes on a web page, or provide a multi-turn dialogue interface in the form of an instant messaging tool.

[0016] The AI ​​simulated visitor module is built on a large language model adjusted based on psychological corpus. It is used to dynamically generate dialogue responses that match the background, issues and behavioral patterns of the virtual visitor in the case after the user selects a psychological counseling training case.

[0017] Specifically, the AI-simulated client module is built upon large-scale language models (such as Qianwen and LLaMA). Its key lies not in using the original general model, but in adjusting it with psychological corpora. This process is typically achieved through fine-tuning or cue word engineering, allowing the model to learn the dialogue logic of psychological counseling, intervention techniques from specific schools of thought, and the expression styles of different individuals with psychological problems. When a user selects a case, the module loads the case's background story, counseling issues, and behavioral patterns as context and role settings. For each round of user input, the module does not randomly generate responses, but rather dynamically infers the client's current possible psychological state and emotional reactions based on these preset parameters, thereby generating personalized dialogue content that matches the client's role setting, simulating interactions in real counseling.

[0018] The skills assessment module is used to analyze the entire record of users' multi-round dialogues with virtual visitors through the user interaction module, and to conduct multi-dimensional quantitative assessment of the consulting skills demonstrated by users, generating assessment results.

[0019] Specifically, the assessment can be multi-dimensional, rather than a single score. Typical dimensions include, but are not limited to: empathy (the ability to accurately understand and respond to the client's emotions), questioning skills (whether the questions are open-ended and the guidance is appropriate), application of techniques (whether specific techniques from schools of thought such as CBT are correctly used), and structuring ability (whether the pace of the consultation is effectively controlled). The final result is a structured assessment, typically including quantitative scores for each dimension and a brief qualitative analysis.

[0020] The feedback report generation module generates a structured skills improvement suggestion report based on the assessment results and presents it to the user through the user interaction module.

[0021] Specifically, the feedback report generation module utilizes natural language generation technology to produce structured skill improvement suggestion reports. These reports not only summarize the user's strengths and weaknesses but also provide specific and actionable improvement suggestions. The reports can be presented in a combination of text and graphics, and support export and sharing, facilitating user review and teacher follow-up. Furthermore, growth curves can be generated based on the user's historical training data, enabling personalized learning path planning.

[0022] In this embodiment, the case library module stores multiple psychological counseling training cases. After the user selects a psychological counseling training case, the AI ​​simulated client module dynamically generates dialogue responses based on the virtual client in that case. The skills assessment module performs multi-dimensional quantitative assessment of the counseling skills demonstrated by the user and generates assessment results. The feedback report generation module generates a structured skills improvement suggestion report based on the assessment results. This effectively solves the dilemma in traditional psychological counselor training where it is difficult to balance large-scale, personalized, and high-quality practical training. It provides psychology students and aspiring counselors with a repeatable, zero-risk, and highly realistic immersive training environment, significantly improving the efficiency and professionalism of counseling skills training, and at a relatively low cost.

[0023] In one embodiment of this specification, the skills assessment module is further configured to: The dialogue records between the user and the virtual visitor are combined with the preset evaluation prompt word template to form prompt words; Input the prompt words into a large language model for evaluation; Obtain structured assessment results, including quantitative scores and qualitative analysis, from the output of a large language model used for evaluation.

[0024] In this embodiment, the preset assessment prompt template can be a structured task instruction, which typically includes: role setting: for example, you are a senior psychological counseling supervisor; assessment criteria: clearly listing the dimensions to be assessed (such as empathy, emotional understanding, question quality, and clarity of expression), and defining the scoring criteria for each dimension (such as a 1-5 point scale and its meaning); output format requirements: strictly specifying the structure of the model output, for example, requiring the content to be organized with specific tags such as scores, strengths, and improvement suggestions, ensuring that the generated results are structured and easy for subsequent program parsing and presentation. The dialogue record is embedded into the reserved space of the assessment prompt template to form a complete and context-rich prompt.

[0025] The large language model used for assessment can be an instance of the same model as the AI-simulated visitor module (distinguishing functions through different prompts), or it can be a smaller, dedicated model specifically tuned for the assessment task to optimize cost and speed. A smaller, dedicated model specifically tuned for the assessment task refers to a streamlined, efficient, and goal-oriented derived model obtained by using supervised fine-tuning or knowledge distillation techniques with assessment-specific corpora on the foundation of a core large model. The construction process of this dedicated model is as follows: First, a base model with a small parameter size is selected as the foundation; then, a large-scale "consultation dialogue-assessment report" pairing data annotated by psychology experts is collected or generated as training corpus; finally, the base model is fine-tuned end-to-end using this corpus, making its internal parameters specifically adapted to the task requirements and output format of consultation skills assessment. Compared to a general-purpose large language model, this dedicated model sacrifices broad general knowledge capabilities in exchange for higher accuracy, output stability, and lower computational resource consumption and response latency on specific assessment tasks, thereby significantly optimizing the system's operating costs and concurrent processing speed while ensuring the professionalism of the assessment. The model reasons and makes judgments based on its internal knowledge and understanding of the dialogue context, according to the standards in the template. Because the prompt word template constrains the output format, the model's output is no longer free-flowing text, but a structured evaluation result containing quantitative scores and qualitative analysis. Quantitative scores: Providing specific scores for each evaluation dimension (e.g., empathy: 4 points), achieving objectivity and measurability of the evaluation results, facilitating user tracking of growth. Qualitative analysis: Listing specific strengths and improvement suggestions in natural language, providing users with a clear and easy-to-understand action guide, explaining why a score was obtained.

[0026] In this embodiment, the assessment process is fully automated, eliminating complete reliance on human supervision and enabling the platform to support a massive number of users training simultaneously and receiving immediate feedback. By using unified assessment standards and models, the system avoids the subjective biases and inconsistent standards that can occur in manual assessments, ensuring the consistency of assessment results. By modifying the assessment prompt templates, assessment dimensions can be easily added, deleted, or modified to adapt to different counseling styles (such as switching from CBT to narrative therapy) or training focuses, demonstrating the system's strong adaptability.

[0027] In one embodiment of this specification, the dialogue record between the user and the virtual visitor is combined with a preset evaluation prompt template to form prompts, including: Identify the current counseling stage in the psychological counseling training case; Specifically, the goals and core techniques differ across different counseling stages (e.g., relationship building, problem assessment, cognitive restructuring, behavioral activation, and termination consolidation), and the evaluation criteria should also have different emphases. By analyzing the dialogue record, the current counseling stage can be identified. This can be achieved through keyword matching, conversation-structure-based classifiers, or small models.

[0028] Retrieve sub-templates that match the consultation stage from the pre-set assessment prompt template library; Extract key intervention statements from the user's dialogue records and label them with corresponding intervention categories based on the semantic roles of the key intervention statements; the intervention categories include exploratory, instructive, empathetic, and structured categories; Specifically, the user's (consultant's) statements are filtered from the dialogue record. These statements are then semantically labeled according to the intervention category (e.g., exploratory, directive, empathetic, structured). This can be achieved by comparing with a pre-built dictionary of technical keywords or by using a lightweight text classification model.

[0029] Based on the consultation stage and intervention category, a counterfactual reference fragment is dynamically inserted into the invoked sub-template; the counterfactual reference fragment describes a typical response that meets the best practice standards in the same situation; Specifically, a counterfactual fragment library is pre-built, storing typical responses defined by experts and conforming to best practices in various typical scenarios. Based on the consultation stage and intervention category identified in the previous steps, the system retrieves the most matching counterfactual reference fragment from the fragment library. For example, for a guiding user question in the cognitive restructuring stage, the system will retrieve a sample fragment of how experts would ask similar questions at this stage.

[0030] Dynamically insert this best practice snippet into the template of evaluation prompts.

[0031] The dialogue transcripts containing key intervention statements, intervention categories, and sub-templates with inserted counterfactual reference fragments are combined to form structured prompt words.

[0032] In this embodiment, by providing context-relevant best practices as direct references, the comprehension bias of the evaluation model is greatly reduced, making scores and comments more accurate and to the point. The evaluation report is no longer an isolated assessment, but rather a comparative learning process. Users not only know where they are lacking, but also intuitively see how to improve, greatly enhancing the specificity and actionability of the feedback. The evaluation criteria are no longer static, but dynamically adjusted according to the actual progress of the dialogue and the specific technical actions taken by the user, ensuring a close alignment between the evaluation and the training process.

[0033] In one embodiment of this specification, identifying the counseling stage of the current psychological counseling training case includes: The dialogue record is segmented into sentences, and sentences containing consulting technical terms are extracted to form a set of technical sentences; The set of technical sentences is matched with a pre-defined list of stage keywords, and the number of times each stage's corresponding keywords are matched is counted. The stage with the highest number of hits and exceeding the preset hit threshold is selected as the consultation stage; If the highest number of hits is lower than the preset hit threshold, the next round of dialogue will be collected and rematched until the preset hit threshold requirement is met.

[0034] In this embodiment, a predefined glossary of consulting techniques (such as automatic thinking, core beliefs, exposure, empathy, and reconstruction) is used as a filter to select sentences containing these technical terms, forming a set of technical sentences. The system also pre-defines a stage keyword list, which defines high-frequency words or phrases specific to each consulting stage (such as problem assessment or cognitive restructuring). The set of technical sentences is matched against the stage keyword list, and the number of keyword hits for each stage is counted.

[0035] In one embodiment of this specification, extracting key intervention statements from a user's dialogue log includes: Filter and retain statements from the dialogue record where the speaker is the consultant, as the remaining statements; The remaining sentences are scored according to their similarity to the pre-set intervention keyword dictionary to obtain the corresponding similarity score; Select the statements with the highest similarity as key intervention statements.

[0036] In one embodiment of this specification, based on the consultation stage and intervention category, a counterfactual reference fragment is dynamically inserted into the invoked sub-template, including: Search the counterfactual fragment library for candidate fragments that match the current stage label and intervention category; the counterfactual fragment library is derived from the Clinical Supervision Manual, CBT intervention model texts, and standardized responses reviewed by psychologists; Calculate the semantic similarity between key intervention statements and candidate segments, and select the segment with the highest similarity that is higher than the preset similarity threshold as the counterfactual reference segment; If the similarity is lower than the preset similarity threshold, then counterfactual segments that meet the best practice standards will be generated in real time based on the current stage and intervention category, and written to the segment library simultaneously. Insert the selected counterfactual fragment into the reserved space in the sub-template.

[0037] In this embodiment, a text embedding model (such as Sentence-BERT) is used to calculate the semantic similarity between the user's key intervention statement and each candidate fragment. If the similarity of all candidate fragments is lower than a preset similarity threshold, it indicates that the fragment library lacks a perfectly matching reference. At this point, the system activates a fallback mechanism: based on the current stage and intervention category, a large language model is immediately invoked to generate a counterfactual fragment that conforms to best practice standards. This newly generated fragment is synchronously written into the fragment library. This is a self-evolving design that allows the system's knowledge base to be continuously enriched and improved, adapting to more diverse dialogue scenarios.

[0038] In one embodiment of this specification, the skills assessment module is further configured to: The original dialogue record is input into the perturbation generation model to generate several semantically equivalent perturbed dialogue records; The original dialogue record and each perturbed dialogue record are combined with a preset evaluation prompt word template to form a set of parallel prompt words, which are then input into a large language model for evaluation to obtain a set of parallel evaluation results. Calculate the consistency score of the parallel evaluation results for this group. If the consistency score is lower than the preset consistency threshold, the evaluation dimensions with consistency scores lower than the preset consistency threshold are fed back to the feedback report generation module and marked in the generated skill improvement suggestions.

[0039] In this embodiment, the perturbation generation model refers to an algorithm or system capable of semantically preserving transformations of input text. This can be achieved through methods such as synonym replacement, sentence structure adjustment, or word order reordering. The aim is to simulate potential expression differences in real-world consultation scenarios, thereby testing and evaluating the model's robustness under semantic invariance. Parallel prompt words refer to multiple sets of prompt words generated based on the original dialogue record and its perturbation version. This can be achieved by dynamically inserting perturbation records and evaluation prompt word templates, aiming to capture the variability of the evaluation model's output and provide a comprehensive data foundation for subsequent consistency analysis. The consistency score is an indicator that quantifies the stability of the evaluation results. It can be achieved through statistical methods such as analysis of variance or similarity calculation, aiming to accurately identify unreliable evaluation steps and ensure that skill improvement suggestions are based on reliable evidence rather than random fluctuations.

[0040] In this embodiment, a perturbation verification mechanism is introduced to effectively address the issue of fluctuations in evaluation results due to subtle changes in input, thereby improving the stability of skill assessment and the reliability of feedback. The semantically equivalent records generated by the perturbation generation model are limited to changes in surface form while maintaining core semantics, avoiding irrelevant interference and ensuring that the perturbation results truly reflect input sensitivity. By combining prompt words with the original records and each perturbation record separately and inputting them into the evaluation model, multi-angle evaluation of the same dialogue content is achieved, and the system captures the variability of the evaluation model's output. Based on this, the consistency score of parallel evaluation results is calculated, and the problem dimension is fed back when it falls below a threshold, quantifying the stability of the evaluation results. This allows the feedback report to accurately identify unreliable evaluation steps and guide users to focus on skill points that need improvement.

[0041] In one embodiment of this specification, the AI-simulated visitor module is further used for: A psychological state vector is constructed and dynamically updated based on the background story, counseling issues, and behavioral patterns of virtual visitors. Each round of user dialogue input is used as an intervention on the mental state vector. The change in the mental state vector is calculated through an intervention-response prediction model trained on psychological dialogue data. Based on the updated psychological state vector, a dialogue response that matches the virtual visitor's personality traits and current psychological state is generated.

[0042] In this embodiment, the psychological state vector refers to a multidimensional numerical representation that can quantify the psychological characteristics of a virtual visitor. It can be implemented using embedding vector techniques in machine learning or by mapping psychological scale scores. The intervention-response prediction model can be understood as a deep learning model specifically designed for psychological counseling scenarios. Its purpose is to capture the complex correlation between user intervention and the psychological changes of virtual visitors, thereby improving the realism of psychological state evolution.

[0043] In this embodiment, the lack of realistic evolution in virtual client responses is effectively addressed by introducing a dynamic modeling mechanism based on psychological state vectors. First, an initial psychological state vector is constructed based on the virtual client's background story, counseling issues, and behavioral patterns. This process ensures consistency between the virtual character's personalized settings and subsequent dynamic evolution. Then, each round of user dialogue input is considered an intervention in the psychological state vector, and the state change is calculated using an intervention-response prediction model trained on psychological dialogue data. This design allows the evolution of psychological states to accurately reflect the progressive logic in real counseling. Dialogue responses are generated based on the updated psychological state vectors. This dynamic generation method not only aligns with the virtual client's personality traits but also accurately reflects their current psychological state, significantly enhancing the realism and relevance of the dialogue.

[0044] Personality traits refer to the inherent, relatively stable psychological characteristics and behavioral tendencies of a virtual client, pre-generated during the case initialization phase. They include inherent attributes such as emotional response baselines (e.g., anxiety tendency), cognitive styles (e.g., catastrophic thinking, core beliefs), behavioral patterns (e.g., conflict avoidance), and interpersonal styles (e.g., dependent). Personality traits constitute the client's basic psychological profile, determining their typical interpretation and reaction patterns to various events, ensuring consistent and credible role behavior in multiple rounds of dialogue. Current psychological state refers to the virtual client's real-time changing psychological condition during the consultation dialogue, quantified through a dynamically updated psychological state vector. It includes instantaneous emotions (e.g., intensity of anger), level of consultation motivation, degree of trust in the counselor, activation intensity of specific issues, and immediate changes in cognitive content. The current psychological state dynamically evolves under the influence of each round of user intervention, directly determining the specific content and emotional tone of the AI's response in the current dialogue round, reflecting the progress of the consultation process and the fluctuations in the client's psychological state.

[0045] In one embodiment of this specification, the AI-simulated visitor module is further used for: Cluster analysis was performed on the repeated consultation patterns of users in each training session to identify users' habitual consultation styles, which were used as the results of the cluster analysis. Based on the cluster analysis results, in subsequent training, response patterns that conflict with the user's habitual consultation style are configured for virtual visitors, in order to specifically challenge the user's consultation mindset.

[0046] In this embodiment, cluster analysis refers to the process of grouping and classifying behavioral characteristics in users' historical dialogue data using algorithms. This can be achieved using methods such as K-means clustering, hierarchical clustering, or density clustering, with the aim of extracting representative user consultation patterns from large amounts of unstructured data. Habitual consultation style refers to the stable and repetitive behavioral tendencies exhibited by users in multiple psychological counseling training sessions. It can be defined by statistically analyzing the types of intervention statements and dialogue strategies frequently used by users, with the aim of providing clear goals for subsequent personalized training. Reaction pattern refers to the behavioral logic dynamically adjusted by the virtual client based on the user's habitual style. It can be implemented based on a rule engine or reinforcement learning model, aiming to break the user's fixed thinking patterns by introducing conflict scenarios.

[0047] In this embodiment, the system uses clustering analysis algorithms to extract repeatedly used counseling skills and behavioral tendencies based on long-term accumulated user dialogue data, ensuring the objectivity and accuracy of the analysis results. The analysis results are then transformed into actionable style tags, which are used to dynamically adjust the response logic of the virtual client. For example, when the system detects that a user tends to frequently use instructive interventions, it configures the virtual client to exhibit resistance or skepticism, forcing the user to rethink and try new counseling strategies. This mechanism fully utilizes the real-time adaptability of the AI ​​module, not only accurately identifying the user's skill gaps but also promoting the flexible application and systematic improvement of the user's counseling skills through continuous feedback loops. Through the above technical solution, the platform can implement a highly personalized challenge mechanism in psychological counseling skills training, helping users break through the limitations of habitual thinking and thus significantly improving training effectiveness.

[0048] In one embodiment of this specification, the platform further includes a dialogue quality monitoring module, used for: Real-time monitoring of the dialogue process between users and virtual visitors to identify quality indicators for establishing consultation relationships; among them, the quality indicators for establishing consultation relationships include at least one of the following: alliance strength, working alliance depth, and trust level; When the quality indicators of establishing a consultation relationship are detected to be lower than the preset quality threshold, the relationship repair mechanism is triggered. The relationship repair mechanism includes: adjusting the dialogue response through the AI-simulated visitor module to proactively release signals for relationship repair; and / or pushing targeted relationship building tips to users through the user interaction module.

[0049] In this embodiment, the dialogue quality monitoring module can be implemented using natural language processing technology combined with psychological scale scoring rules. The quality indicators for establishing a counseling relationship can be understood as a set of core parameters reflecting the quality of interaction between the two parties during psychological counseling. These parameters can be generated through a computational model designed based on psychological theory, such as quantifying them using sentiment analysis algorithms combined with alliance relationship scale scores. In practical applications, the relationship repair mechanism refers to a set of dynamic response strategies, which can be implemented through a pre-set repair dialogue library combined with real-time generated personalized suggestions. The purpose is to intervene promptly in the deterioration of the relationship and ensure training effectiveness.

[0050] In this embodiment, the dialogue quality monitoring module dynamically captures changes in core indicators such as alliance strength, working alliance depth, and trust level through continuous analysis of the dialogue text between the user and the virtual visitor. These indicators, selected based on psychological counseling theory, accurately reflect the quality status of the counseling relationship. When an indicator falls below a preset threshold, the system automatically triggers a repair mechanism. On one hand, the AI-simulated visitor module adjusts its response strategy according to the current dialogue context, such as improving the interactive atmosphere by expressing understanding or apology. On the other hand, the user interaction module pushes specific relationship-building techniques to the user based on the dialogue situation, such as open-ended questions or methods of expressing empathy. This dual-path design not only repairs the dialogue relationship in real time but also helps users master relationship maintenance skills, thus forming a complete monitoring-identification-triggering-repair process, effectively solving the problem of missing relationship quality monitoring in psychological counseling training.

[0051] In one embodiment of this specification, the case library module contains several longitudinal training cases designed in accordance with the theoretical framework of cognitive behavioral therapy (CBT); The longitudinal training case consists of multiple consecutive interviews, in which the initial psychological state vector of the virtual visitor in the later interview is inherited from the state at the end of the previous interview, and reflects the progress of cognitive reconstruction, changes in emotion regulation and behavioral experiment feedback. Users gradually advance the counseling process by interacting with the same virtual client in multiple consecutive sessions, thereby training their systematic skills and ability to apply phased strategies in structured psychological intervention.

[0052] In this embodiment, the longitudinal training case refers to a multi-stage psychological counseling simulation tool designed based on the theoretical framework of cognitive behavioral therapy. It can be implemented through phased task setting, dynamic difficulty adjustment, and phased goal decomposition, aiming to provide users with a continuous training environment that meets clinical practice standards. In practical applications, the initial psychological state vector of the virtual client refers to a data structure used to quantitatively describe the psychological characteristics of the virtual client. It can be generated through machine learning models or constructed based on psychological scale scores, aiming to realistically reproduce the psychological evolution trajectory of the client in continuous counseling. Cognitive reconstruction progress refers to the dynamic process of positive changes in the maladaptive cognition of the virtual client under the intervention of the counselor. This progress is specifically manifested in the identification, evaluation, and correction of the client's initial automatic negative thoughts (such as "I will definitely fail") and deep-seated core beliefs (such as "I am incompetent"). The system represents these changes in the initial psychological state vector of subsequent sessions by quantifying dimensions such as the reduction in the frequency of cognitive distortion, the adoption rate of alternative rational cognitions, and the improvement in cognitive flexibility, thereby reflecting the phased effects of counseling at the cognitive level. Emotional regulation change refers to the evolution of a virtual client's ability to manage and cope with negative emotions during the counseling process. This change is specifically manifested in: a decrease in the intensity of the client's perception of dominant negative emotions (such as anxiety and depression), an increase in emotional stability, and an improvement in the effectiveness of the counselor-instructed regulation strategies (such as mindful breathing and emotion naming). The system dynamically updates the client's psychological state vector by tracking indicators such as the amplitude of emotional fluctuations, the speed of emotional recovery, and the richness of emotional expression, visually demonstrating the trajectory of the client's improved emotional state. Behavioral experiment feedback refers to the results and experiences obtained by the virtual client after performing real-world behavioral tasks between sessions, as suggested by the counselor. This feedback is specifically manifested in: the client's execution of new behavioral patterns (such as proactive socializing and task decomposition) attempted to address their original avoidance or fear behaviors, the objective evidence collected during the process, and the verification results of their original negative expectations. The system integrates this feedback information and updates the client's psychological state vector and behavioral pattern data accordingly to simulate changes in behavioral activation levels and self-efficacy in subsequent counseling sessions.

[0053] In this embodiment, the longitudinal training cases in the case library module strictly match the phased logic of structured intervention, ensuring the scientific and standardized nature of the training foundation. The design of multiple consecutive sessions forces users to experience the complete consultation cycle, avoiding the fragmented application of skills in single case training and strengthening the overall grasp of the consultation process. The dynamic continuation mechanism of the initial psychological state vector of the virtual client not only realistically restores the client's psychological evolution trajectory but also requires users to adjust subsequent strategies based on historical intervention effects, overcoming the limitation of ignoring long-term effects in isolated training. At the same time, by concretizing the progress of cognitive reconstruction, changes in emotion regulation, and behavioral experiment feedback into quantitative indicators, users can clearly identify the phased intervention effectiveness and flexibly switch technology applications. Ultimately, the process of users interacting with the same virtual client in multiple consecutive sessions forms a complete skill chain from initial assessment to goal achievement, effectively training their ability to integrate phased strategies in structured psychological intervention.

[0054] In one embodiment of this specification, the skills assessment module is further configured to: After completing the standard consultation dialogue between the user and the virtual visitor, the mirrored reverse consultation mode is automatically initiated: All intervention statements made by the user in the standard dialogue are returned to the user, and the AI ​​simulates the visitor to play the role of counselor, while the original user receives follow-up questions based on their own intervention logic as the visitor. Collect user experience texts submitted in the reverse role and compare them with their intervention statements in the standard dialogue in terms of emotional consistency and cognitive bias. Based on the comparison results, an intervention self-reflection score is generated and this score is included in the self-awareness sub-dimensional of the skills assessment results; When the score is lower than the preset score threshold, the platform locks access to the next stage of the case until the user completes the mirror reverse consultation and the score reaches the threshold, thereby forcing the formation of an intervention-experience-reflection closed loop.

[0055] In this embodiment, the mirrored reverse counseling mode refers to a mechanism for achieving deep introspection through role reversal. It can be dynamically triggered, activated immediately after the standard counseling dialogue concludes. This design aims to ensure that the reflection process is closely linked to the user's actual intervention behavior, avoiding the hollow reflection problem caused by detachment from specific context in traditional assessments. Specifically, the AI-simulated client-councilor function generates targeted follow-up questions based on the user's original dialogue content. It can accurately reproduce user intervention statements through semantic and logical analysis, aiming to amplify the user's cognitive blind spots without introducing external interference. Furthermore, the intervention self-reflection score is a quantitative indicator used to map the user's introspection depth. It can be generated through an algorithm comparing emotional consistency and cognitive bias, aiming to compensate for the lack of a self-awareness dimension in existing assessment systems.

[0056] In this embodiment, the above-mentioned technical solution systematically addresses the lack of self-awareness in psychological counseling skills training by constructing a mandatory role reversal mechanism. After the standard counseling dialogue is completed, the platform automatically triggers a mirrored reverse counseling mode, using the user's own intervention statements as reflection material, shifting the assessment from one-way skill analysis to two-way introspection-driven evaluation. During this process, AI simulates a client to generate targeted follow-up questions based on the user's original dialogue content, ensuring that the reflection phase is closely coupled with the user's actual intervention behavior. Subsequently, the platform collects the user's experience text in the reversed role and compares it with the intervention statements in the standard dialogue for emotional consistency and cognitive bias, revealing implicit cognitive biases that are difficult to capture by traditional assessments. Finally, by generating an intervention self-reflection score and incorporating it into the self-awareness sub-dimensional, combined with an access control mechanism, reflection is ensured to become a necessary condition for skill advancement, thereby driving users from passively receiving feedback to actively introspecting and achieving sustainable improvement in counseling skills.

[0057] Generally, the system's AI uses LangChain as the agent framework to build the agent structure, uses Huggingface to download the Qianwen large model, and uses the CBT-Case corpus for LoRA fine-tuning, collects data on psychological terminology and knowledge, and uses Huggingface to fine-tune the large model, giving it psychological common sense. During the dialogue, the intention recognition model (RoBERTa-psych) analyzes the user's intervention intention in real time, triggers the corresponding skill slot, and then, driven by the recall recommendation algorithm and the RoBERTa-psych intention recognition model, generates accurate and interpretable multi-turn dialogues with visitors in real time.

[0058] Figure 2 This is a schematic diagram of the architecture of an intelligent psychological counseling skills training and feedback platform based on an AI large-scale model provided in an embodiment of the present invention. The architecture may include a large-scale model service layer, a storage layer, an access layer, an application layer, and a user layer.

[0059] 1. Large Model Service Layer: This layer is the AI ​​intelligence hub of the system, comprising a finely tuned large language model (Qianwen 7B), an external psychology knowledge base, and reinforcement learning and human annotation optimization modules. Its core function is to drive intelligent services across the entire platform: fine-tuning enables the model to master professional dialogue skills in psychological counseling, simulating highly realistic virtual clients; utilizing external knowledge bases (such as DSM-5 and CBT intervention manuals) ensures the professionalism and accuracy of dialogue; and through continuous optimization and iteration, improves AI performance in areas such as emotional understanding, risk assessment, and skills assessment. Later optimizations can utilize NLP analysis of text emotions, speech recognition of anxious tone, and video micro-expression extraction to train students' core competency of "multi-dimensionally capturing client emotions" after employment.

[0060] 2. Storage Layer: This layer forms the system's data foundation. It employs a hybrid cloud model, with its core components including a master-slave MySQL database and Alibaba Cloud storage services. The uses of private and public clouds are clearly defined. Its function is to achieve secure, efficient, and elastic data storage and management: the private cloud ensures the confidentiality and protection of sensitive core data such as dialogue logs and training scores locally; the public cloud supports the synchronization and backup of user basic information and skill development trajectories; and the master-slave database design guarantees efficient and reliable data read / write operations. Local storage: Stores student training dialogue logs and scoring details, allowing students to review their job skills gaps; Cloud storage: Synchronizes with students' "job skills development trajectory" on the public cloud, serving as proof of practical skills to demonstrate to employers during job applications.

[0061] 3. Access Layer (Gateway Service Layer): This layer serves as the command center for internal and external communication within the system. It includes not only the FastAPI interface gateway, Nginx load balancer, and Redis caching, but also integrates network governance functions such as authentication, routing, and rate limiting, and collaborates with the underlying cloud and local server hardware resources. Its core function is to ensure high availability and security: FastAPI encapsulates AI capabilities into a standard API; Nginx balances traffic to handle high-concurrency training scenarios; Redis caches user vulnerabilities to enable personalized recommendations; and comprehensive authentication and security policies safeguard all data interaction entry points.

[0062] 4. Application Layer: This layer directly carries and presents the product's business functions, integrating psychological counseling skills training, AI psychological assessment, AI emotion recognition, AI psychological profiling, an AI dialogue early warning system, and a campus mental health module. Its role is to transform underlying AI capabilities into concrete and usable services, such as: guiding users through the entire training process from establishing a counseling relationship to intervention; analyzing user emotions in real time and generating psychological profiles; providing early warnings of risks in dialogues; and offering customized mental health services for campus scenarios. For example, it simulates the entire post-employment counseling process, establishing a counseling relationship, problem exploration, CBT intervention, termination, and referral, supporting switching between "clients of different professions / problem types"; intelligent assessment: quantifying students' performance in key employment dimensions such as "empathy ability, questioning skills, and CBT strategy application" in real time, generating improvement suggestions.

[0063] 5. User Layer: This layer is the interface between the system and end users, covering multiple terminals including Android APP, iOS APP, web page, and WeChat mini-program. Its purpose is to provide psychology students and teachers with a convenient, easy-to-use, and professionally relevant operational entry point, allowing users to select cases anytime, anywhere, engage in immersive dialogue training with AI clients, view detailed quantitative assessment reports, and review historical dialogues, thus gaining a one-stop skills enhancement experience. It supports selecting consultation scenarios, such as 'teenagers with academic anxiety' or 'workplace interpersonal conflict clients,' and allows users to view training score reports and review historical dialogues. The interface logic aligns with the operational workflow of post-employment psychological counseling platforms.

[0064] The following is a brief introduction to the selection of artificial intelligence technology in this invention.

[0065] Deep learning framework: PyTorch + HuggingFace; 1. PyTorch supports dynamic computation graphs, which facilitates fine-tuning of model parameters according to psychology employment scenarios and improves scenario adaptability; 2. HuggingFace provides a rich library of pre-trained models, which can quickly load models and fine-tune them with psychology corpora, reducing development costs and allowing students to focus on employment skills training.

[0066] Intelligent agent building tool: LangChain; 1. Use LangChain to build intelligent agents, supporting the arrangement of "virtual visitor interaction - dialogue recording - rating feedback", which can simulate the "consultation - recording - review" workflow after employment, helping students adapt to professional habits in advance; 2. Supports memory mechanism, which can record the context of students' multiple rounds of training, and improve employment shortcomings in a targeted manner.

[0067] Microservice backend framework: FastAPI; 1. FastAPI is used to encapsulate intelligent agents to form a server backend. Its asynchronous and non-blocking characteristics support high concurrency and simulate the scenario of multiple visitors consulting at the same time after employment; 2. Swagger API documentation is automatically generated to facilitate students' understanding and improve their ability to collaborate with technical teams after employment.

[0068] Pre-trained large model base: Qianwen Large Model 7B; 1. The technical solution clearly states "using Qianwen Large Model 7B as an open source base". Its 7B parameter scale balances performance and deployment cost, making it suitable for universities to carry out large-scale student training and avoid the impact of hardware limitations on employment skills training; 2. The open source feature supports local deployment, which can train students' "model localization maintenance" ability and adapt to the "data privacy protection" needs of small and medium-sized psychological institutions after employment (avoiding the upload of core consultation data to the cloud).

[0069] To ensure the stable delivery of core AI capabilities and the closed-loop business process, the project adopted a mature and reliable supporting technology system. Python was chosen as the development language due to its status as a mainstream tool in the intersection of psychology and AI, offering comprehensive data analysis and modeling libraries such as Pandas and Scikit-learn, enabling efficient processing of training and scoring data. MySQL was used as the database, leveraging its transactional features to ensure the integrity and consistency of student training records and consultation dialogues, simulating rigorous record management standards for future employment. Its database sharding and partitioning capabilities also prepare for future massive data storage. The system was ultimately deployed on a Linux server environment to ensure the stability, efficiency, and security of production-grade services. This entire technology selection, from core algorithms to the supporting environment, was designed with the fundamental principle of "simulating real-world workflows and cultivating core employment skills."

[0070] Deployment Example: This project uses a Docker containerization solution deployed on a cloud server, achieving high availability and resource isolation through a multi-container architecture. The deployment process uses Ubuntu 22.04 as the base environment. First, the Docker engine and Compose tools are installed, followed by the creation of a custom image containing Python 3.10 and Miniconda. The core services are divided into two main containers: the AI ​​engine and the database. The AI ​​container is based on an NVIDIA CUDA image, integrating Hugging Face Transformers, LangChain, LangGraph, and FastAPI technology stacks. It accelerates large model inference through GPU passthrough and exposes port 8000 to provide RESTful API services. The MySQL container runs the database independently, achieving data persistence through volume mounting. Sensitive configurations such as database passwords and Hugging Face Tokens are injected via environment variables, and model files are pre-downloaded to the persistent storage volume during the build phase. Services are orchestrated and started with a single click using Docker Compose, supporting 200+ concurrent sessions and a response time within 1.5 seconds. A built-in Swagger UI interface documentation facilitates verification. This solution has been validated in the AWS EC2 production environment and features elastic resource scaling, data disaster recovery snapshots, and hot model updates to ensure stable and efficient service operation.

[0071] Application Scenario Example: AI can simulate clients, for example, addressing the common psychological problem of "academic anxiety among college students." This simulates a client exhibiting typical characteristics of academic anxiety, such as persistent anxiety and tension, difficulty falling asleep, frequent awakenings at night, dreams of failing courses, loss of appetite, and an all-or-nothing mindset: "I must get 90 points, otherwise I'm a failure, I know nothing, I'm not suited for studying," or catastrophic thinking: "If I don't do well this time → I'm not suited for studying → I can't get into graduate school → my life is ruined → I've let my parents down, I'm a failure." Users can respond based on the AI ​​client's reactions, deepening their understanding of the skills. AI-simulated clients overcome the time, space, and resource limitations of traditional counseling skills training, providing users with highly realistic counseling training scenarios. It can simulate diverse client backgrounds, psychological states, and problem types, such as college students with academic anxiety or painters whose childhood lack of parental love led to mental illness. This allows learners to repeatedly practice in different situations, enhancing their ability to cope with complex counseling scenarios. Simultaneously, it also assists psychology teachers and trainers in their teaching.

[0072] Personalized Assessment: After students complete their consultation training with the AI ​​client, the system can conduct personalized assessments based on dialogue data, considering multiple dimensions such as empathy, questioning skills, and CBT strategy application. For example, when guiding the client to identify automatic thinking, if a student accurately points out that "doing poorly on the exam is disastrous" is a catastrophic thought, the system will give a high score in the "cognitive recognition" dimension. If the user uses prompts multiple times during training, the system will deduct corresponding points in the application dimension. Finally, the system will generate a personalized assessment report based on the user's overall performance, providing targeted feedback. Real-time feedback and personalized assessment can generate quantifiable personal growth data reports. Through multi-dimensional analysis, the system objectively evaluates the user's consultation performance, accurately identifies the user's problems, and generates clear improvement suggestions. This effectively addresses the industry pain points of traditional training, such as subjective feedback and lack of dynamic assessment, helping learners to identify weaknesses and consolidate learned skills.

[0073] Based on the same general inventive concept, this invention also protects a method for training and providing feedback on intelligent psychological counseling skills based on a large AI model, such as... Figure 3 As shown, Figure 3This is a flowchart illustrating the AI-based intelligent psychological counseling skills training and feedback method provided in this embodiment of the invention. The following describes the AI-based intelligent psychological counseling skills training and feedback method provided by this invention. The AI-based intelligent psychological counseling skills training and feedback method described below can be referred to in correspondence with the AI-based intelligent psychological counseling skills training and feedback platform described above. The AI-based intelligent psychological counseling skills training and feedback method can be applied to any of the AI-based intelligent psychological counseling skills training and feedback platforms in the above embodiments.

[0074] AI-based large-scale model-based intelligent psychological counseling skills training and feedback methods include: Step 301: The case library module stores multiple psychological counseling training cases; each psychological counseling training case includes the background story, counseling issues, and behavioral patterns of a virtual client. Step 302: The user interaction module allows users to select training cases from the case library module, which serves as the interaction interface for users to conduct multi-round dialogues with virtual visitors. Step 303: Through the AI ​​simulated visitor module built on a large language model adjusted by psychological corpus, after the user selects a psychological counseling training case, the AI ​​simulated visitor module dynamically generates dialogue responses that match the background, issues and behavioral patterns of the virtual visitor in the case. Step 304: Analyze the entire record of the user's multi-round dialogue with the virtual visitor through the user interaction module using the skills assessment module, and conduct a multi-dimensional quantitative assessment of the consulting skills demonstrated by the user to generate assessment results. Step 305: Based on the assessment results, the feedback report generation module generates a structured skills improvement suggestion report and presents it to the user through the user interaction module.

[0075] Figure 4 This is a schematic diagram of the structure of the electronic device provided in an embodiment of the present invention.

[0076] like Figure 4 As shown, the electronic device may include a processor 410, a communications interface 420, a memory 430, and a communication bus 440. The processor 410, communications interface 420, and memory 430 communicate with each other via the communication bus 440. The processor 410 can call logical instructions from the memory 430 to execute an AI-based intelligent psychological counseling skills training and feedback method.

[0077] Furthermore, the logical instructions in the aforementioned memory 430 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0078] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer is able to execute the AI-based intelligent psychological counseling skills training and feedback method provided by the above methods.

[0079] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the AI-based intelligent psychological counseling skills training and feedback methods provided by the above methods.

[0080] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0081] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0082] 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 the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An intelligent psychological counseling skills training and feedback platform based on an AI large-scale model, characterized in that, include: The case library module stores multiple psychological counseling training cases; each psychological counseling training case includes the background story, counseling issues, and behavioral patterns of a virtual client. The user interaction module is used to allow users to select training cases from the case library module and serves as an interactive interface for users to conduct multi-round dialogues with virtual visitors. The AI ​​simulated visitor module is built on a large language model adjusted with psychological corpus. It is used to dynamically generate dialogue responses that match the background, issues and behavioral patterns of the virtual visitor in the case after the user selects a psychological counseling training case. The skills assessment module is used to analyze the entire record of the user's multi-round dialogue with the virtual visitor through the user interaction module, and to conduct a multi-dimensional quantitative assessment of the consulting skills demonstrated by the user, generating assessment results. The feedback report generation module is used to generate a structured skills improvement suggestion report based on the evaluation results, and present it to the user through the user interaction module.

2. The AI-based intelligent psychological counseling skills training and feedback platform according to claim 1, characterized in that, The skills assessment module is also used for: The dialogue records between the user and the virtual visitor are combined with the preset evaluation prompt word template to form prompt words; The prompt words are input into a large language model for evaluation; Obtain the structured evaluation results, including quantitative scores and qualitative analysis, output by the large language model used for the evaluation.

3. The intelligent psychological counseling skills training and feedback platform based on an AI large model as described in claim 2, characterized in that, The dialogue records between the user and the virtual visitor are combined with preset assessment prompt templates to form prompts, including: Identify the current counseling stage in the psychological counseling training case; From the preset assessment prompt template library, call the sub-template that matches the consultation stage; Extract the user's key intervention statements from the dialogue records and label the corresponding intervention categories based on the semantic roles of the key intervention statements; Based on the consultation stage and intervention category, a counterfactual reference fragment is dynamically inserted into the invoked sub-template; wherein, the counterfactual reference fragment describes a response that conforms to the best practice standard in the same situation; The dialogue transcripts containing key intervention statements, intervention categories, and sub-templates with inserted counterfactual reference fragments are combined to form structured prompt words.

4. The intelligent psychological counseling skills training and feedback platform based on an AI large model as described in claim 3, characterized in that, Identify the current counseling stage of the psychological counseling training case, including: The dialogue record is segmented into sentences, and sentences containing consulting technical terms are extracted to form a set of technical sentences; The set of technical sentences is matched with a preset list of stage keywords, and the number of times the corresponding keywords for each stage are matched is counted. The stage with the highest number of hits and exceeding the preset hit threshold is selected as the consultation stage; If the highest number of hits is lower than the preset hit threshold, the next round of dialogue will be collected and rematched until the preset hit threshold requirement is met. Based on the consultation stage and intervention category, a counterfactual reference fragment is dynamically inserted into the invoked sub-template, including: Retrieve candidate fragments from the counterfactual fragment library that simultaneously match the current stage label and intervention category; Calculate the semantic similarity between key intervention statements and candidate segments, and select the segment with the highest similarity that is higher than the preset similarity threshold as the counterfactual reference segment; If the similarity is lower than the preset similarity threshold, then counterfactual segments that meet the best practice standards will be generated in real time based on the current stage and intervention category, and written to the segment library simultaneously. Insert the selected counterfactual fragment into the reserved space in the sub-template.

5. The AI-based intelligent psychological counseling skills training and feedback platform according to claim 2, characterized in that, The skills assessment module is also used for: Based on the original dialogue records, generate several semantically equivalent perturbed dialogue records; The original dialogue record and each perturbed dialogue record are combined with a preset evaluation prompt word template to form a set of parallel prompt words, which are then input into the large language model used for evaluation to obtain a set of parallel evaluation results. Calculate the consistency score of the parallel evaluation results. If the consistency score is lower than the preset consistency threshold, the evaluation dimensions with consistency scores lower than the preset consistency threshold are fed back to the feedback report generation module and marked in the generated skill improvement suggestions.

6. The intelligent psychological counseling skills training and feedback platform based on an AI large model as described in claim 1, characterized in that, The AI-simulated visitor module is also used for: Based on the background story, counseling issues, and behavioral patterns of the virtual visitor, a psychological state vector is constructed and dynamically updated. Each round of user dialogue input is used as an intervention on the mental state vector, and the change in the mental state vector is calculated. Based on the updated psychological state vector, a dialogue response that matches the virtual visitor's personality traits and current psychological state is generated.

7. The intelligent psychological counseling skills training and feedback platform based on an AI large model according to claim 1, characterized in that, The AI-simulated visitor module is also used for: Cluster analysis was performed on the repeated consultation patterns of users in each training session to identify users' habitual consultation styles, which were used as the results of the cluster analysis. Based on the clustering analysis results, in subsequent training, response patterns that conflict with the habitual consultation style are configured for virtual visitors in order to specifically challenge the user's consultation mindset.

8. The intelligent psychological counseling skills training and feedback platform based on an AI large model as described in claim 1, characterized in that, The platform also includes a dialogue quality monitoring module, used for: Real-time monitoring of the dialogue process between users and virtual visitors; identification of consultation relationships; and establishment of quality indicators. When the quality index of the consultation relationship establishment is detected to be lower than the preset quality threshold, the relationship repair mechanism is triggered; The relationship repair mechanism includes: adjusting the dialogue response method through the AI ​​simulated visitor module to proactively release signals for relationship repair; And / or push targeted relationship-building tips to users through the user interaction module.

9. The intelligent psychological counseling skills training and feedback platform based on an AI large model according to claim 1, characterized in that, The case library module contains several longitudinal training cases designed in accordance with the theoretical framework of cognitive behavioral therapy. The longitudinal training case consists of multiple consecutive interviews, in which the initial psychological state vector of the virtual visitor in the later interview is inherited from the state at the end of the previous interview, and reflects the progress of cognitive reconstruction, changes in emotion regulation and behavioral experiment feedback. Users advance the consultation process by interacting with the same virtual visitor in multiple consecutive sessions.

10. The intelligent psychological counseling skills training and feedback platform based on an AI large model according to claim 1, characterized in that, The skills assessment module is also used for: After completing the consultation dialogue between the user and the virtual visitor, all the intervention statements made by the user in the dialogue are returned to the user, and the AI ​​simulates the visitor to play the role of the counselor, while the original user receives follow-up questions based on their own intervention logic as the visitor. Collect the experience text submitted by users in the reverse role and compare it with their intervention statements in the standard dialogue in terms of emotional consistency and cognitive bias to obtain the comparison results; Based on the comparison results, an intervention self-reflection score is generated and this score is included in the self-awareness sub-dimensional of the skills assessment results; When the score is lower than the preset score threshold, access to the next stage of the case is locked until the user completes the mirror reverse consultation and the score reaches the preset score threshold.

Images