Psychological tutoring system based on LLM model
By combining large language models and psychological theories, a psychological counseling system is constructed, which solves the defects of the existing system's inability to identify and handle user psychological problems, and the provision of personalized intervention plans is achieved, which improves the effectiveness and user experience of psychological counseling.
Patent Information
- Application Number
- CN202410885052.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-03
- Publication Date
- 2024-11-22
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing artificial intelligence psychological counseling system cannot meet the actual application needs, lacks crisis intervention mechanisms and psychological technical support, cannot effectively identify and handle users' psychological problems, and cannot provide personalized and scalable psychological counseling services.
By combining large language model (LLM) with psychological theory, a psychological counseling system is built, including psychological portrait construction module, main complaint trouble collection module, relationship model establishment module and coaching goal confirmation and intervention plan recommendation module, using natural language processing and diagram On analytical technology, build a network model for users' psychological portraits and problem occurrence, and provide personalized intervention solutions.
It improves the effect and user experience of psychological counseling, can more accurately identify and handle users' psychological problems, provide personalized intervention plans, meet the overall scenario needs of psychological counseling, and enhances the scalability of the system and user trust.
Smart Images

Figure CN119007940A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of psychological counseling systems, and in particular to a psychological counseling system based on an LLM model. Background Art
[0002] Psychology, as the theoretical basis of psychological counseling, provides a scientific understanding of human behavior and psychological processes. Psychological treatment methods such as cognitive behavioral therapy (CBT) and psychoanalysis have been widely used in clinical practice and have been proven to be effective for many psychological disorders. However, these methods usually require a lot of manpower and time investment, which limits their feasibility in large-scale application.
[0003] In recent years, the rapid development of artificial intelligence technology, especially the emergence of large language models (LLMs), has brought new opportunities to the field of psychological counseling. LLMs can understand and generate natural language through deep learning technology, making communication between machines and humans more natural and smooth. This provides a technical foundation for the development of intelligent psychological counseling systems, making it possible to provide personalized and scalable psychological counseling services.
[0004] Although some AI + psychological technology products and systems have been developed, there are still some problems in practical applications:
[0005] Wide range of scenarios: Currently, more technologies are combined to create a chatbot with psychological knowledge, but in actual applications, this type of artificial intelligence cannot meet scenario requirements.
[0006] Lack of crisis intervention mechanism: Current AI + psychology products may not recognize or be overly sensitive to potential psychological problems displayed by users, thus immediately interrupting the chat or failing to have "empathy"
[0007] Lack of psychological technical support: The current LLM model is more integrated with psychological technology, but the process of psychological counseling has not been deeply studied, so it only has partial functions and cannot meet the overall scenario needs of psychological counseling. Summary of the invention
[0008] In response to the above technical problems, the present invention combines the natural language processing capabilities of the LLM large model with psychological theories to construct a psychological counseling system that can follow the workflow and understand and process students' psychological states and behavior patterns.
[0009] In order to achieve the above-mentioned purpose, the technical solution of the present invention provides a psychological counseling system based on the LLM model, which includes the following modules: a psychological portrait construction module, which is used to construct a user psychological portrait based on the collected user-related data through a pre-constructed psychological portrait model; a main complaint collection module, which is used to collect the user's main complaint information through the interaction with the user through the human-computer interaction module; a relationship model establishment module, which is used to construct a user problem occurrence network model based on the user's psychological portrait and the main complaint information, understand the causes of the user's troubles and predict the subsequent possible behaviors, so as to provide a basis for subsequent counseling and intervention; a counseling goal confirmation and intervention plan recommendation module, which is used to guide the user to determine the corresponding counseling goals and recommend appropriate intervention plans based on the problem relationship pattern determined by the relationship model establishment module.
[0010] Furthermore, constructing a psychological portrait model includes the following steps: S1: defining a psychological portrait label system; S2: based on the labels in the psychological portrait label system, constructing the following multiple types of data collection means and limiting the value range of their labels: literature collection means: setting questions and option tables corresponding to different labels, and assigning corresponding label values to each option; data analysis means: constructing a labeling agent based on expert experience to correspond user corpus to label values; statistical modeling means: for predictive class labels, by screening multiple labels related to the predictive class labels, constructing a statistical analysis model of the predictive class labels, so as to determine the predictive class label values through the statistical analysis model; S3: constructing label collection means; S4: constructing a user information table to fix users and their corresponding labels on a table.
[0011] Furthermore, the statistical modeling approach constructs a statistical analysis model through the following steps:
[0012] (1) Screen out n labels related to the construction of prediction class labels through literature collection and data analysis;
[0013] (2) Perform data cleaning;
[0014] (3) Construct a binary logistics regression model, where X1, X2, X3…Xn are the selected relevant label values, Y is the predicted class label value, β0, β1, β2…β n are model parameters;
[0015]
[0016] (4) Model parameter estimation: use maximum likelihood estimation to estimate model parameters:
[0017]
[0018] Its logarithmic function is:
[0019]
[0020] The maximum value of the logarithmic function is solved by the gradient descent method to obtain the estimated value of the parameter;
[0021] (5) Model testing: using the chi-square test to determine the goodness of model fit and using the Wald statistic to test the significance of the regression coefficient;
[0022] (6) Model evaluation: evaluate the model performance through test data.
[0023] Furthermore, in the data analysis method, the correspondence between the user's usage time and the frequency of users clicking on internal resources and the relevant psychological label values is also constructed.
[0024] Furthermore, in step S3, when constructing the label collection means, in addition to using labels for internal product functions, an active label assignment means is also provided.
[0025] Furthermore, when constructing a user psychological portrait based on the collected user-related data, when obtaining the text content generated by the user using the product function, the label value of the user is constructed based on the labeling agent through correlation analysis, wherein the correlation analysis adopts the Pearson rank correlation calculation formula, assuming that A is the obtained text content, that is, the user corpus, B is the label value, and r is the correlation coefficient:
[0026]
[0027] When r> sets the threshold, construct the corresponding label value.
[0028] Furthermore, the relationship model building module constructs a user problem occurrence network model according to the following steps: T1 node definition: define each specific problem of the user as a node in the network, and define the user's main complaint as a new node, where the node vi∈V represents the user's specific problem; T2 network construction: identify the association and interaction between problems, and construct a problem network graph, where G=(V,E) is set as a network graph, V is a set of nodes, E is a set of edges, and the edge eij∈E represents the user's specific problem; T3 relationship mapping: determine the relationship between nodes, and represent these relationships in the network model in the form of lines; T4 pattern recognition: use the centrality analysis algorithm in graph theory to identify patterns and key influencing factors in the problem network.
[0029] Furthermore, in step T3 relationship mapping, the LLM model is used to identify the relationship between nodes and determine the problem set V = {v1, v2, v3...vn}; the relationship between nodes is mainly determined by the partial correlation coefficient or is manually defined; the thickness of the edge represents the strength of the relationship, and the color of the edge represents whether the relationship is positively correlated or negatively correlated; the connection between nodes is connected through edges, and each edge is defined as eij.
[0030] Furthermore, in step T4 pattern recognition, the importance of the nodes is determined through optimized degree centrality analysis, and the constructed network is visualized through a network visualization tool.
[0031] Furthermore, the system also includes the following modules: an intervention effect evaluation module: used to evaluate the effect of the intervention plan based on the user feedback results after psychological counseling. BRIEF DESCRIPTION OF THE DRAWINGS
[0032] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.
[0033] Figure 1 It is a work flow chart of the psychological counseling system based on the LLM model of the present invention. DETAILED DESCRIPTION
[0034] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.
[0035] The purpose of the present invention is to propose a psychological counseling system and method based on the LLM language model in response to the deficiencies of the prior art. The system and method adopt the technology of the psychological portrait system, and pre-train the LLM on a large amount of psychological counseling data and diversified basic instructions, so that the large language model has professional knowledge and instruction-following skills in the psychological field, and different instructions are called through different process nodes to complete the entire workflow.
[0036] The key points of the present invention are:
[0037] (1) Methods for constructing a psychological portrait model: including steps such as data collection, model training, label generation, and how to use the model to analyze user data and construct a user psychological portrait.
[0038] (2) Methods for collecting complaints and building relationship models: including how to use LLM to communicate with users, collect complaints, and how to build a problem occurrence network model to analyze the correlation and interaction between problems.
[0039] (3) Personalized intervention plan recommendation algorithm: including how to recommend appropriate intervention plans based on the user’s psychological profile, main complaints and counseling goals.
[0040] The workflow that the present invention should follow is as follows:
[0041] Collect user data: Utilize the LLM model and multimodal data such as voice and text generated during user interaction to build a student psychological profile based on the self-developed psychological state model, including psychological state, behavioral patterns, family environment, psychological traits, beliefs and attitudes, emotional feelings, values, neural & cognitive abilities and other dimensions.
[0042] Collection of main complaints: Build a professional psychological counselor agent to collect information about students’ current psychological problems. It is necessary to collect three pieces of information from users: “what is the problem?”, “what are the feelings?” and “what are the thoughts?”
[0043] Establish a relationship model: Build a relationship network by collecting user psychology in the early stage and combining it with the user's main complaints: the rules followed by the relationship network are mainly the (stressor-trait-psychological problem-behavior) model, to understand the causes of students' troubles and predict subsequent behaviors.
[0044] Determine counseling goals: Build a psychological counselor agent to understand the current problems that students need to solve and the expected results.
[0045] Provide improvement suggestions: Combine the network model of student goals and current student problems to find the causes of students' current problems in the network, and provide structured and executable suggestions based on the causes and goals.
[0046] Student feedback: Through the above communication, students rate the psychological counseling, understand its effectiveness, and provide reliable opinions and suggestions.
[0047] Specifically, see Figure 1 , the training method and specific steps are as follows:
[0048] (1) Label generation based on psychological portrait model
[0049] Purpose: To create a personalized psychological portrait for the user to guide the subsequent counseling process.
[0050] Operation: Collect users’ basic information, psychological state, behavioral patterns, etc., and use LLM to generate labels such as “anxiety tendency”, “academic pressure”, etc.
[0051] The specific operations are as follows:
[0052] ① Invite relevant school psychology experts and internal psychology employees with graduate degrees to define students' psychological characteristics, and use discussion and voting to generate a total of 134+ student psychological portrait labels, covering 9 major label themes: psychological state, family environment, emotional feelings, somatization, neurological & cognitive abilities, psychological characteristics, values, new year attitudes, and behaviors / events.
[0053] ② Based on the labels in the "label system", through literature collection, expert experience, data analysis and other methods, build multiple types of data collection methods and limit the value range of their labels. The following table 1 is an example:
[0054] Table 1
[0055] Data collection method
[0056] part
[0057] Literature collection For example, the label of life satisfaction refers to the subjective well-being scale, one of which is "I am satisfied with my life". There are 7 options based on the Likert scale, and students are given different attribute label values based on different options.
[0058] 1. High life satisfaction
[0059] 2. Average life satisfaction
[0060] 3. Lower life satisfaction
[0061] Expert experience refers to the professional label judgment of multiple experts on the same student question and answer content, and the label value is assigned. If different experts have the same or similar labels for the same question and answer, the label value is taken.
[0062] Match the corpus with the label values and build a labeling agent.
[0063] Data analysis - Conduct data analysis based on user usage records or functions, such as analyzing the content in the "Mood Diary" through voice-to-text conversion, and then passing the text content to the labeling agent to generate specific user tags.
[0064] Conduct corresponding data analysis on user usage time, frequency of user clicks on internal resources and related psychological label values. When a user has multiple labels, we know from literature analysis that there are existing labeling models and the relationship between manual labeling. Use the logistic regression method to construct a function of the independent variable and the dependent variable to find the real independent variable.
[0065] For example, in the calculation of the possibility of suicide, we construct suicide through manual labeling. For example, if the user has said that he wants to commit suicide before, and the counselor confirms that he does have suicidal tendencies, then we construct a statistical analysis model to find the dependent variable.
[0066] Among them, the correlation analysis adopts Pearson rank correlation, assuming A is the user corpus, B is the label value, and r is the correlation coefficient:
[0067]
[0068] Select r> above a set threshold (e.g., 0.4) to construct a label value.
[0069] Among them, taking the suicide model construction as an example, the construction method of the statistical analysis model is described:
[0070] s1: Label screening: extract labels that may be related to the possibility of suicide through literature and data analysis, a total of n labels.
[0071] s2: Data cleaning. The criteria for selecting each piece of data are that the missing value is less than n / 2, the counselor has manually interviewed the suicidal ideation, and the most recent data was updated within six months.
[0072] s3: Statistical modeling - model building, construct a binary logistics regression model, where suicidal ideation is the dependent variable, Y, the presence of suicidal ideation is 1, and the absence of suicidal ideation is 0, X1 X2X3...X n are the relevant label values of suicide, which are also binary variables. The formula can be:
[0073]
[0074] s4: Statistical Modeling - Parameter Estimation, using Maximum Likelihood Estimation (MLE) to estimate model parameters,
[0075] The likelihood function is the product of the probabilities of all observations:
[0076]
[0077] The logarithmic function is:
[0078]
[0079] The maximum value of the logarithmic function is solved by the gradient descent method to obtain the estimated values of the parameters.
[0080] s5: Model testing. The chi-square test is used to judge the goodness of model fitting, and the Wald statistic is used to test the significance of the regression coefficients. Through the significance of the regression coefficients, we obtained a total of 17 independent variables X, including label values such as being impulsive, having low social support, having high depressive mood, and being self-centered.
[0081] s6: Model evaluation. Therefore, our suicide model is
[0082] P(likelihood of suicide = 1 | X) = 1 / (1 + exp(-(0.3241 + 0.23 * suicidal tendency + 0.082 * high sense of loneliness +... + 0.1423 * low support from friends))). Substitute the test data. If P < 0.5, it is 0; otherwise, it is 1. Compare the two groups of data to obtain data such as the accuracy rate and recall rate of the model.
[0083] It can be understood that the above method for constructing the statistical analysis model can also be used for constructing predictive labels such as mental health (yes / no) and academic achievement (high / low). Since it is difficult to obtain accurate label values for these predictive labels through forms such as user questions and answers, the label values of these predictive labels can be obtained by constructing a statistical analysis model between these predictive labels and other relevant labels.
[0084] ③ Construct label collection means. In addition to using labels for internal product functions, provide means for active label assignment, such as initiating conversations with users, collecting corpus data, recommending assessment tools, and recommending function utilities.
[0085] ④ Construct a user information table to fix the user and their labels on one table for query and call.
[0086] (2) Process of collecting main complaints
[0087] Purpose: To accurately identify the main psychological problems faced by the user currently.
[0088] Operation: Construct a professional psychological counselor agent to collect the current psychological problems of students. It is necessary to collect three pieces of information from the user: "what's the problem", "what's the feeling", and "what's the thought".
[0089] (3) Establish a relationship model
[0090] Purpose: Based on the psychological portrait and the user's main complaints, construct a user problem occurrence network model to deeply understand the user's psychological state and the root causes of problems, providing a basis for subsequent counseling and intervention.
[0091] Operation: Collect user psychology in the early stage and build a relationship network based on the user's main complaints: the relationship network mainly follows the (stress source-trait-psychological problem-behavior) model to understand the causes of students' troubles and predict the subsequent behaviors. The specific operations are as follows:
[0092] ① Node definition: Each specific problem of the user is defined as a node in the network, and the user's main complaint is also defined as a new node. Node vi∈V represents the specific problem of the user.
[0093] ② Network construction: Identify the associations and interactions between problems and construct a problem network map. Let G = (V, E) be a network graph, where V is a set of nodes, E is a set of edges, and the edge eij∈E represents the user's specific problem.
[0094] ③ Relationship mapping: determine the relationship between nodes, such as causal relationship, trigger relationship, etc., and represent these relationships in the form of lines in the network model.
[0095] Through the LLM model, the relationship between nodes is identified and the problem set V = {v1, v2, v3...vn} is determined;
[0096] The relationship between nodes is mainly determined by the partial correlation coefficient, which can also be customized;
[0097] The thickness of the edge represents the strength of the relationship, and the color of the edge represents whether the relationship is positive or negative;
[0098] The nodes are connected through edges, and each edge is defined as eij. That is, if there are 10 nodes, there can be at most n*(n-1) / 2 edges. If the correlation coefficient is not considered, there can be a total of at most 45 edges.
[0099] ④ Pattern recognition: Use the centrality analysis algorithm in graph theory to identify patterns and key influencing factors in the problem network.
[0100] Determine the importance of nodes through optimized degree centrality analysis;
[0101] Degree centrality measures the number of connections to a node, but in practice, degree centrality is too simple and can easily miss some important nodes. Therefore, we add the partial correlation coefficients of the direct connections of each node to construct a new degree centrality algorithm formula. Change from d(vi) to
[0102]
[0103] Where R = the sum of partial correlation coefficients, and rij is the coefficient of each edge.
[0104] Build a relationship network: visualize the constructed network through network visualization tools.
[0105] (4) Confirmation of coaching goals
[0106] Purpose: To jointly determine the specific goals of the coaching with the user and ensure that the coaching process has a clear direction.
[0107] Operation: Based on the user's main complaint and psychological portrait, LLM guides the user to clarify the counseling goals and formulate corresponding counseling plans.
[0108] (5) Recommended intervention plan
[0109] Purpose: To provide users with personalized psychological intervention programs.
[0110] Operation: Based on the user's counseling goals and psychological portrait, LLM recommends appropriate intervention plans, such as cognitive behavioral therapy, relaxation training, etc.
[0111] (6) Evaluation of intervention effects
[0112] Objective: To evaluate the effect of psychological intervention program.
[0113] Operation: After a period of coaching, the system evaluates the effectiveness of the intervention program based on user feedback.
[0114] Specific application examples:
[0115] User Background
[0116] Name: Zhang San
[0117] Age: 25
[0118] Occupation: Postgraduate
[0119] Mental state: I feel anxious and stressed recently due to the upcoming graduation thesis defense and employment pressure.
[0120] Counseling Process
[0121] Step 1: Label generation for psychological profile model
[0122] Operation: Zhang San makes a preliminary registration through the “Lingben Partner” system, filling in basic information and a self-assessment of his current mental state.
[0123] System response: LLM generates psychological portrait labels based on the information provided by Zhang San, such as "anxiety tendency", "academic pressure", and "career development anxiety".
[0124] Step 2: Collection of complaints
[0125] Operation: The system communicates with Zhang San through natural language processing technology to collect his main complaints.
[0126] User input: [User]: "I've been feeling anxious lately. I'm worried that I won't be able to defend my thesis, and I'm also worried that I won't be able to find a good job."
[0127] System Response: [Model]: “I understand how you’re feeling, and many people have similar concerns when facing important moments. Let’s see how we can address these issues together.”
[0128] Step 3: Build a relationship model
[0129] Action: LLM builds a trusting relationship with John by showing empathy and providing positive feedback.
[0130] System Response: [Model]: “It’s normal to feel anxious when faced with this kind of stress. I’m here to help you find ways to reduce your anxiety.”
[0131] Step 4: Confirmation of coaching goals
[0132] Operation: LLM guides Zhang San to clarify the tutoring goals.
[0133] System Response: [Model]: "To help you reduce your anxiety, we can set some specific coaching goals. For example, improve the efficiency of paper preparation or develop an active career preparation plan."
[0134] Step 5: Recommend intervention options
[0135] Operation: Based on Zhang San’s psychological profile and counseling goals, LLM recommends an intervention plan.
[0136] System response: [Model]: "Based on your situation, I suggest you try some relaxation training, such as deep breathing exercises and mindfulness meditation. At the same time, we can develop a detailed defense preparation plan and career planning."
[0137] Step 6: Intervention effect evaluation and optimization
[0138] Operation: After a period of counseling, systematically evaluate the effectiveness of the intervention program.
[0139] User feedback: [User]: “I find meditation helpful, but I’m still making slow progress on my paper.”
[0140] System response: "I noticed that meditation helps you. For essay writing, we can adjust the strategy, such as breaking down the task, setting small goals, and taking it step by step."
[0141] Step 7: Continued follow-up and support
[0142] Action: LLM regularly tracks Zhang San's mental state and provides ongoing support.
[0143] System Tracking: [Model]: “It’s been two weeks since we last spoke, and I wanted to get an update on how you’re doing. Is your meditation practice still going? How’s your paper coming along?”
[0144] In an embodiment of the present invention, a psychological counseling system based on an LLM model is provided, which includes the following modules: a psychological portrait construction module, which is used to construct a user psychological portrait based on collected user-related data through a pre-constructed psychological portrait model; a main complaint collection module, which is used to collect the user's main complaint information through interaction with the user through a human-computer interaction module; a relationship model establishment module, which is used to construct a user problem occurrence network model based on the user's psychological portrait and the main complaint information, understand the causes of the user's troubles and predict the subsequent possible behaviors, so as to provide a basis for subsequent counseling and intervention; a counseling goal confirmation and intervention plan recommendation module, which is used to guide the user to determine the corresponding counseling goal and recommend a suitable intervention plan based on the problem relationship pattern determined by the relationship model establishment module.
[0145] Furthermore, the psychological portrait model is constructed through the following steps: S1: define the psychological portrait label system; S2: based on the labels in the psychological portrait label system, construct the following multiple types of data collection means and limit the value range of their labels: literature collection means: set questions and option tables corresponding to different labels, and assign corresponding label values to each option; data analysis means: based on expert experience, construct a labeling agent for matching user corpus with label values; statistical modeling means: for predictive class labels, by screening multiple labels related to the predictive class labels, construct a statistical analysis model of the predictive class labels, so as to determine the predictive class label values through the statistical analysis model; S3: construct label collection means; S4: construct a user information table to fix users and their corresponding labels on a table.
[0146] Furthermore, the statistical modeling approach constructs a statistical analysis model through the following steps:
[0147] (1) Screen out n labels related to the construction of prediction class labels through literature collection and data analysis;
[0148] (2) Perform data cleaning;
[0149] (3) Construct a binary logistics regression model, where X1, X2, X3…Xn are the selected relevant label values, Y is the predicted class label value, β0, β1, β2…β n are model parameters;
[0150]
[0151] (4) Model parameter estimation: use maximum likelihood estimation to estimate model parameters:
[0152]
[0153] Its logarithmic function is:
[0154]
[0155] The maximum value of the logarithmic function is solved by the gradient descent method to obtain the estimated value of the parameter;
[0156] (5) Model testing: using the chi-square test to determine the goodness of model fit and using the Wald statistic to test the significance of the regression coefficient;
[0157] (6) Model evaluation: evaluate the model performance through test data.
[0158] Furthermore, in the data analysis method, the correspondence between the user's usage time and the frequency of users clicking on internal resources and the relevant psychological label values is also constructed.
[0159] Furthermore, in step S3, when constructing the label collection means, in addition to using labels for internal product functions, an active label assignment means is also provided.
[0160] Furthermore, when constructing a user psychological portrait based on the collected user-related data, when obtaining the text content generated by the user using the product function, the label value of the user is constructed based on the labeling agent through correlation analysis, wherein the correlation analysis adopts the Pearson rank correlation calculation formula, assuming that A is the obtained text content, that is, the user corpus, B is the label value, and r is the correlation coefficient:
[0161]
[0162] And when r>set threshold, construct the corresponding label value.
[0163] Furthermore, the relationship model building module constructs a user problem occurrence network model according to the following steps: T1 node definition: define each specific problem of the user as a node in the network, and define the user's main complaint as a new node, where the node vi∈V represents the user's specific problem; T2 network construction: identify the association and interaction between problems, and construct a problem network graph, where G=(V,E) is set as a network graph, V is a set of nodes, E is a set of edges, and the edge eij∈E represents the user's specific problem; T3 relationship mapping: determine the relationship between nodes, and represent these relationships in the network model in the form of lines; T4 pattern recognition: use the centrality analysis algorithm in graph theory to identify patterns and key influencing factors in the problem network.
[0164] Furthermore, in step T3 relationship mapping, the LLM model is used to identify the relationship between nodes and determine the problem set V = {v1, v2, v3...vn}; the relationship between nodes is mainly determined by the partial correlation coefficient or is manually defined; the thickness of the edge represents the strength of the relationship, and the color of the edge represents whether the relationship is positively correlated or negatively correlated; the connection between nodes is connected through edges, and each edge is defined as eij.
[0165] Furthermore, in step T4 pattern recognition, the importance of the nodes is determined through optimized degree centrality analysis, and the constructed network is visualized through a network visualization tool.
[0166] Furthermore, the system also includes the following modules: an intervention effect evaluation module: used to evaluate the effect of the intervention plan based on the user feedback results after psychological counseling.
[0167] Beneficial technical effects of the present invention:
[0168] Improved the experience and effect of psychological counseling: Based on user psychological portraits and problem patterns, provide users with personalized psychological intervention plans, including cognitive behavioral therapy, mindfulness therapy, emotion regulation techniques, etc., which are more in line with user needs and improve intervention effects; use natural language processing technology to achieve natural interaction with users, including voice recognition, semantic understanding, sentiment analysis, etc., to improve user experience and make it easier for users to accept and trust the system.
[0169] It can be understood that the above embodiments are only for illustrating the technical concept and features of the present invention, and their purpose is to enable ordinary technicians in the field to understand the content of the present invention and implement it. It cannot be used to limit the protection scope of the present invention. All equivalent changes or modifications made according to the spirit of the present invention should be included in the protection scope of the present invention.
Claims
1. A psychological counseling system based on the LLM model, characterized in that: Includes the following modules: A psychological portrait construction module is used to construct a user psychological portrait based on the collected user-related data through a pre-built psychological portrait model; A main complaint collection module is used to collect the user's main complaint information through interaction with the user through the human-computer interaction module; The relationship model building module is used to build a user problem occurrence network model based on the user's psychological portrait and main complaint information, understand the causes of user troubles and predict the subsequent possible behaviors, so as to provide a basis for subsequent counseling and intervention; The tutoring goal confirmation and intervention plan recommendation module is used to guide users to determine the corresponding tutoring goals and recommend appropriate intervention plans based on the problem relationship pattern determined by the relationship model establishment module.
2. The system according to claim 1, characterized in that The construction of a psychological portrait model includes the following steps: S1: Define the psychological portrait labeling system; S2: Based on the labels in the psychological portrait label system, construct the following types of data collection methods and limit the value range of their labels: Literature collection method: set up questions and option tables corresponding to different labels, and assign corresponding label values to each option. Data analysis method: Based on expert experience, a labeling agent is constructed to match user corpus with label values. Statistical modeling means: for the prediction class label, by screening multiple labels related to the prediction class label, constructing a statistical analysis model of the prediction class label, so as to determine the prediction class label value through the statistical analysis model; S3: Construct label collection means; S4: Build a user information table to fix users and their corresponding tags in a table.
3. The system according to claim 2, characterized in that Statistical modeling methods use the following steps to build a statistical analysis model: (1) Screen out n labels related to the construction of prediction class labels through literature collection and data analysis; (2) Perform data cleaning; (3) Construct a binary logistics regression model, where X1, X2, X3…Xn are the selected relevant label values, Y is the predicted class label value, β0, β1, β2…β n are model parameters; (4) Model parameter estimation: use maximum likelihood estimation to estimate model parameters: Its logarithmic function is: The maximum value of the logarithmic function is solved by the gradient descent method to obtain the estimated value of the parameter; (5) Model testing: using the chi-square test to determine the goodness of model fit and using the Wald statistic to test the significance of the regression coefficient; (6) Model evaluation: evaluate the model performance through test data.
4. The system according to claim 3, characterized in that In the data analysis method, the correspondence between the user's usage time and the frequency of users clicking on internal resources and the relevant psychological label values is also constructed.
5. The system according to claim 4, characterized in that In step S3, when constructing the label collection means, in addition to using labels for internal product functions, active label assignment means are also provided.
6. The system according to claim 5, characterized in that When constructing a user psychological portrait based on the collected user-related data, when obtaining the text content generated by the user using the product function, the label value of the user is constructed based on the labeling agent through correlation analysis. The correlation analysis adopts the Pearson rank correlation calculation formula, assuming that A is the obtained text content, that is, the user corpus, B is the label value, and r is the correlation coefficient: When r> sets the threshold, construct the corresponding label value.
7. The system according to any one of claims 1 to 6, characterized in that: The relationship model building module builds a user problem occurrence network model according to the following steps: T1 Node definition: Each specific problem of the user is defined as a node in the network, and the user's main complaint is also defined as a new node, where the node vi∈V represents the specific problem of the user; T2 Network construction: Identify the associations and interactions between problems and construct a problem network graph, where G = (V, E) is a network graph, V is a set of nodes, E is a set of edges, and the edge eij∈E represents the user's specific problem; T3 Relationship Mapping: Determine the relationships between nodes and represent these relationships in the form of lines in the network model; T4 Pattern Recognition: Use centrality analysis algorithms from graph theory to identify patterns and key influencing factors in problem networks.
8. The system according to claim 7, characterized in that In step T3 relationship mapping, Through the LLM model, the relationship between nodes is identified and the problem set V = {v1, v2, v3...vn} is determined; The relationship between nodes is mainly determined by the partial correlation coefficient or manually defined; The thickness of the edge represents the strength of the relationship, and the color of the edge represents whether the relationship is positive or negative; The nodes are connected through edges, and each edge is defined as eij.
9. The system according to claim 8, characterized in that In step T4 pattern recognition, the importance of nodes is determined through optimized degree centrality analysis, and the constructed network is visualized through network visualization tools.
10. The system according to claim 1, characterized in that The system also includes the following modules: Intervention effect evaluation module: used to evaluate the effectiveness of the intervention plan based on user feedback after psychological counseling.
