Student-oriented psychological counseling method based on large model

Abstract

The invention discloses a student-oriented psychological counseling method based on a large model. The method comprises the following steps: S1, acquiring emotion expression text data; s2, establishing an emotional state feature vector based on the emotional intensity; s3, collecting emotional state feature vectors for multiple times in a preset time window; s4, generating an emotion buffering strategy based on a short-term emotion fluctuation monitoring result; s5, guiding the emotional state; s6, storing the emotional state feature data after the emotional state feature data is relieved into an emotional file of the student; s7, based on emotional state feedback data, adopting an adaptive learning algorithm to finely adjust the large model, optimizing the analysis of an emotional fluctuation state and the generation ability of an emotional buffering strategy, and gradually adjusting and optimizing an emotional buffering method; and S8, based on the emotional state feedback data and the emotional change trend of the student, repeatedly executing the steps S1-S7 to form an emotional buffer closed-loop feedback system. According to the invention, a more efficient and more intelligent solution can be provided for student psychological health management.

Description

Technical Field

[0001] The present invention relates to the technical field of psychological counseling, and in particular to a psychological counseling method for students based on a large model. Background Art

[0002] With the continuous development of artificial intelligence technology, the application of big models in natural language processing, emotional computing and personalized interaction has gradually matured, bringing new breakthroughs in the field of mental health. Student psychological counseling, as an important part of mental health education, is directly related to students' mental state management and emotional problem intervention. However, in the traditional student psychological counseling system, emotion recognition and personalized counseling still have many shortcomings, which makes it difficult to meet the diverse mental health needs of current students.

[0003] At present, schools and psychological counseling institutions generally rely on manual psychological counseling to carry out their work. Psychological counselors need to conduct face-to-face interviews or observe and analyze students' words and deeds to judge their psychological state and develop counseling plans. Although the above plans can provide personalized services for specific problems, they have obvious limitations: on the one hand, due to the limited number of psychological counselors, it is difficult to meet the psychological counseling needs of a large number of students in schools; on the other hand, students' emotional expressions often have semantic ambiguity and implicit characteristics. Relying solely on manual judgment is prone to omissions or deviations, affecting the accuracy and timeliness of psychological counseling. In addition, students may have psychological pressure when facing psychological counseling, which makes them unwilling to actively express their true emotions, further increasing the difficulty of manual psychological counseling.

[0004] In recent years, some psychological counseling technologies based on emotional computing have been gradually applied to the field of education. Usually, simple emotion recognition algorithms are used to classify students' emotions and generate corresponding counseling suggestions. These technologies have made up for the shortcomings of manual psychological counseling to a certain extent. However, the existing technologies have exposed the following problems in practical applications: On the one hand, the existing emotion recognition algorithms are mostly based on keyword matching or shallow sentiment analysis, which makes it difficult to accurately identify the internal logic of students' complex emotional expressions, resulting in a lack of reliability in emotion recognition results; on the other hand, the existing psychological counseling system is difficult to provide dynamically adjusted counseling plans based on the personalized needs of different students. It can usually only generate suggestions with strong generality and cannot effectively solve students' individual psychological problems. In addition, the technology lacks a closed-loop feedback mechanism and cannot continuously track and optimize students' emotional states, nor can it improve the system's counseling capabilities through learning accumulation.

[0005] In summary, existing technologies have significant deficiencies in terms of emotion recognition accuracy, personalization of counseling programs, and dynamic optimization capabilities, making it difficult to meet the needs of schools and students for efficient, accurate, and personalized mental health management. These technical deficiencies directly affect the effectiveness and coverage of student psychological counseling, and a new method is urgently needed to solve the above problems. Summary of the invention

[0006] One purpose of the present invention is to propose a psychological counseling method for students based on a large model, which can provide a more efficient and intelligent solution for students' mental health management.

[0007] A psychological counseling method for students based on a large model according to an embodiment of the present invention includes the following steps:

[0008] S1. By collecting text information of students' feedback and performing data preprocessing on the text information, noise information irrelevant to emotions is removed to obtain text data of emotion expression;

[0009] S2. Use the big model to analyze the emotional state of the emotional expression text data, use the emotion recognition algorithm to identify the students' emotional state and its characteristics, and establish the emotional state feature vector based on the emotional intensity;

[0010] S3, by collecting the emotional state feature vector multiple times within the preset time window, calculating the rate of change and fluctuation range of the emotional intensity within the time window, and judging whether the student is in a state of large emotional fluctuation; if the fluctuation range exceeds the preset threshold, it is judged that the student is currently in a state of large emotional fluctuation;

[0011] S4, generating emotional buffering strategies based on short-term emotional fluctuation monitoring results;

[0012] S5. Provide students with guidance on emotional buffering strategies, provide text or audio guidance, and help students calm down in the short term through meditation, mindfulness exercises or deep breathing methods, and guide their emotional state;

[0013] S6. After the emotion buffering guidance, collect the students' emotion feedback data, recalculate the students' emotion state feature vector, and store the alleviated emotion state feature data into the students' emotion file;

[0014] S7. Based on the emotional state feedback data, the adaptive learning algorithm is used to fine-tune the large model, optimize the analysis of emotional fluctuation state and the generation of emotional buffering strategies, and gradually adjust and optimize the emotional buffering method;

[0015] S8. Based on the emotional state feedback data and the students' emotional change trends, S1-S7 are repeatedly executed to form an emotional buffer closed-loop feedback system.

[0016] Optionally, S1 includes the following sub-steps:

[0017] S11. Collect text information of students’ feedback through the psychological assistant system. The text information includes students’ emotional descriptions. e , psychological stress source T p , Confused Questions T q And other psychological related content o ;

[0018] S12, collect the text information T = {T e ,T p ,T q ,T o}Preprocessing to remove irrelevant punctuation, repeated words, or non-emotion-related information;

[0019] S13, perform word segmentation, part-of-speech tagging and sentiment feature tagging on the denoised text information set T', and extract key text features F i , forming the emotional expression text dataset D:

[0020] D={F1,F2,…,F n};

[0021] Among them, F i is the feature vector of the ith text information, and n is the number of student text information collected.

[0022] Optionally, S2 includes the following sub-steps:

[0023] S21, input the emotion expression text dataset D into the big model, and use the emotion recognition algorithm of the big model to identify each feature vector F i Perform emotional state analysis, map the results of linear transformation into the probability distribution of emotional categories, quantify students' tendencies in different emotional categories, and calculate the emotional probability distribution P i :

[0024]

[0025] in, W j is the emotion category c j The weight vector, b j is the bias term, m is the total number of emotion categories, and p ij Denotes the eigenvector F i Belongs to the emotion category c j The probability of , e is the natural exponent, is a normalization factor so that the sum of the probabilities of all emotion categories is 1;

[0026] S22. According to the emotion probability distribution P iCalculate each emotion category c j Corresponding emotional intensity l j :

[0027] I j =S(p ij )=-β·ln(1-p ij );

[0028] Among them, S(p ij ) is the emotion intensity function, β is the emotion intensity adjustment coefficient, which is used to amplify or reduce the value of emotion intensity to adapt to different application scenarios, and p ij is the eigenvector F i Belongs to the emotion category c j The probability of reflecting the students' emotional tendency in this emotional category, ln is the natural logarithmic function, when p ij When it approaches 1, l j will tend to positive infinity, indicating that the emotional intensity is extremely high, l j is the emotion category c j The emotional intensity, the larger the value, the stronger the student's emotional response in this emotional category;

[0029] S23, each feature vector F i Emotion category c j and its emotional intensity j Combine and construct the student's emotional state feature vector V i .

[0030] Optionally, S3 includes the following sub-steps:

[0031] S31, collect students' T in the preset time window w The multiple emotional state feature vector set V = {V1, V2, …, V t};

[0032] S32, for each emotion category c in the time window j Emotional intensity sequence Calculate its rate of change

[0033]

[0034] in, is the emotion category c j The rate of change of emotion intensity at time point t, Δt is the time interval, and are the emotion categories c at time points t and t-1 respectively j emotional intensity;

[0035] S33, based on the rate of change of emotion intensity Calculate each emotion category c j In the time window T w The fluctuation range A within w j :

[0036]

[0037] Among them, A j is the emotion category c j The fluctuation range, and Respectively represent the emotion category c in the time window j The maximum and minimum values ​​of emotional intensity;

[0038] S34, by changing the fluctuation range A j With the preset threshold A th Compare and judge whether the student is in a state of greater emotional fluctuation:

[0039]

[0040] Among them, the status j Indicates the emotion category c j Fluctuation state, 1 means in a state of greater emotional fluctuation, 0 means not reaching a state of greater fluctuation;

[0041] S35, integrate the fluctuation states of all emotion categories, and make logical judgment S = state 1, state 2, ..., state m}Evaluate the overall emotional fluctuations of students. If at least one emotional category meets A j ≥A th , it is determined that the student is currently in a state of greater emotional fluctuation and this state is stored.

[0042] Optionally, S4 includes the following sub-steps:

[0043] S41, input the student's overall emotional fluctuation state set S and the fluctuation range set of emotional categories, and determine the emotional category set C that needs to be buffered through the logic judgment module b :

[0044]

[0045] Among them, A j is the emotion category c j The fluctuation range, δ 2 is the variance threshold of the fluctuation, which is used to filter out the emotional categories with more drastic emotional fluctuations. th is the preset fluctuation amplitude threshold, is the emotion category c j In the time window T wAverage emotional intensity within:

[0046]

[0047] S42, for each emotion category c to be buffered j ∈C b , combined with the corresponding emotional intensity Mapping functions by buffering strategy Generate a preliminary buffering strategy B that adapts to emotion category and intensity j :

[0048]

[0049] Among them, B j For the emotion category c j The initial buffer strategy, α1, α2, α3 are the emotion intensity thresholds set according to the applicable scope of the buffer strategy;

[0050] S43, inputting the preliminary buffering strategy set into the large model to generate specific buffering content, including text or audio guidance;

[0051] S44, delivering the generated specific buffer content set to the student's psychological assistant interface, and providing buffer guidance for different emotion categories in real time.

[0052] Optionally, S5 includes the following sub-steps:

[0053] S51. Based on the emotion category set determined in S4, the buffer content is dynamically generated for each student in combination with the student's historical interaction data and real-time emotional state:

[0054] Multi-sensory guidance: The meditation scene is presented synchronously through text description and audio background. The audio background contains natural sound effects adapted to emotional categories and guides students' breathing rhythm through prompt sounds.

[0055] Virtual scene immersion guidance: A dynamic virtual buffering scene is constructed through a psychological assistant system, and students visualize the buffering process through augmented reality devices or screens;

[0056] S52. Provide active interactive meditation guidance:

[0057] Question-and-answer meditation: During meditation, students provide feedback on their emotional state to the system through voice or text, and the system adjusts the meditation guidance content in real time;

[0058] Rhythmic meditation: Dynamically monitor the student's voice or input rhythm, and adjust the student's voice characteristics or speaking speed to audio feedback that matches the meditation rhythm;

[0059] S53. In the process of mindfulness practice, combine dynamic task-based guidance:

[0060] Mindfulness scanning tasks: step-by-step guides students through focusing on sensory feedback from different parts of the body;

[0061] Mindful Observation Task: Helps students develop concentration and relieve negative emotions by guiding them to observe specific external objects;

[0062] S54. In the deep breathing buffering method, various forms of rhythm control are provided:

[0063] Dynamic visual breathing rhythm graph: Through the psychological assistant interface, a dynamic graph that adjusts with breathing is displayed, and students adjust their breathing rhythm according to the contraction and expansion of the graph;

[0064] Biofeedback control: The device monitors students' breathing rate and heart rate in real time, and feeds back to the buffer content, automatically generating breathing rhythm guidance that matches the students' current physiological state.

[0065] Optionally, S7 includes the following sub-steps:

[0066] S71, collecting an emotional state feedback data set, each feedback data F in the emotional state feedback data set i ={(c1,ΔI1),(c2,ΔI2),…,(c m ,ΔI m )} includes the emotion intensity change value of the emotion category set:

[0067]

[0068] in, and Respectively represent the intensity of emotions before and after the buffering activity;

[0069] S72, classify and aggregate the emotional state feedback data set, and construct the emotional buffer effect matrix E, the matrix element E ij The calculation is:

[0070] E ij =w j ΔI j ;

[0071] Among them, E ij Indicates that student i is in emotion category c j The buffering effect on j is the emotion category c j The weight factor is set according to the influence of the emotion category on the overall emotion fluctuation of students;

[0072] S73. Input the emotion buffering effect matrix into the meta-learning model to construct a second-order optimization loss function Fine-tune a large model:

[0073]

[0074] Among them, θ represents the parameters of the large model, represents the large model fine-tuned on the feedback data of student i, α1 is the learning rate, represents the emotion buffer loss of student i:

[0075]

[0076] Among them, ΔI j is the emotion category c j The actual intensity of emotions changes, Changes in the intensity of emotions predicted by the model;

[0077] S74. Dynamically adjust the buffer strategy mapping function according to the optimization results of the meta-learning model Adapt buffering strategies to individual students’ emotional characteristics;

[0078] S75. Use the optimized large model for emotion fluctuation state analysis and emotion buffer strategy generation, and record the improvement of the buffer effect to form a buffer effect optimization cycle.

[0079] The beneficial effects of the present invention are:

[0080] (1) The present invention adopts an adaptive optimization framework based on meta-learning, and dynamically adjusts the model parameters through second-order optimization in the generation of emotional buffer strategies and emotional fluctuation analysis to achieve rapid adaptation to individual emotional states. By constructing a second-order optimization loss function, the meta-learning framework can quickly fine-tune the large model under the condition of a small amount of emotional feedback data, so that it can more accurately capture the complex changing relationship between different student emotion categories. Compared with the traditional emotion recognition model based on fixed parameters, the present invention significantly improves the generalization ability of the model under small sample conditions. In addition, by dynamically adjusting the buffer strategy mapping function, the present invention can optimize the generation of buffer content according to the real-time emotional state of the students. Experimental results show that the buffer strategy adaptation efficiency and emotional fluctuation mitigation rate have been improved by 23% and 18%, respectively.

[0081] (2) The present invention constructs an emotion buffering effect matrix and an emotion fluctuation state feature vector to comprehensively model the multi-dimensional information of emotion intensity, fluctuation amplitude, and time series characteristics, thereby effectively solving the problem of single emotion recognition results in the prior art. Through the time weighting mechanism and the dynamic learning rate adjustment strategy, the model can dynamically adjust the focus at different time periods and capture key features at moments when emotions fluctuate greatly, significantly improving the sensitivity to students' short-term emotion fluctuations and the ability to control long-term emotion trends. Compared with traditional emotion recognition technology, the present invention has outstanding performance in recognition accuracy and time series modeling capabilities, and the overall recognition accuracy of emotion categories has been improved by 21%, and the error range in emotion dynamic prediction has been reduced to about 65% of the traditional method.

[0082] (3) The present invention constructs an emotion buffering strategy optimization mechanism based on closed-loop feedback, integrates students' real-time emotion feedback into the emotion state feature update and model adaptive learning, and can gradually optimize the accuracy of emotion recognition and the pertinence of the buffering strategy through weighted processing and aggregation analysis of feedback information, thereby realizing dynamic closed-loop optimization of psychological counseling. In the personalized psychological counseling of different students, the present invention can continuously learn students' emotional trigger points and behavioral patterns and dynamically adjust counseling methods, making psychological intervention more targeted and sustainable. Experimental results show that the present invention improves the emotion counseling effect by 26% compared with traditional methods in terms of students' long-term emotion management and personalized support, and significantly improves the psychological counseling satisfaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0083] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0084] Figure 1 The present invention is a flowchart of a method for psychological counseling for students based on a large model. DETAILED DESCRIPTION

[0085] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0086] refer to Figure 1 , a psychological counseling method for students based on a large model, including the following steps:

[0087] S1. By collecting text information of students' feedback and performing data preprocessing on the text information, noise information irrelevant to emotions is removed to obtain text data of emotion expression;

[0088] S2. Use the big model to analyze the emotional state of the emotional expression text data, use the emotion recognition algorithm to identify the students' emotional state and its characteristics, and establish the emotional state feature vector based on the emotional intensity;

[0089] S3, by collecting the emotional state feature vector multiple times within the preset time window, calculating the rate of change and fluctuation range of the emotional intensity within the time window, and judging whether the student is in a state of large emotional fluctuation; if the fluctuation range exceeds the preset threshold, it is judged that the student is currently in a state of large emotional fluctuation;

[0090] S4, generating emotional buffering strategies based on short-term emotional fluctuation monitoring results;

[0091] S5. Provide students with guidance on emotional buffering strategies, provide text or audio guidance, and help students calm down in the short term through meditation, mindfulness exercises or deep breathing methods, and guide their emotional state;

[0092] S6. After the emotion buffering guidance, collect the students' emotion feedback data, recalculate the students' emotion state feature vector, and store the alleviated emotion state feature data into the students' emotion file;

[0093] S7. Based on the emotional state feedback data, the adaptive learning algorithm is used to fine-tune the large model, optimize the analysis of emotional fluctuation state and the generation of emotional buffering strategies, and gradually adjust and optimize the emotional buffering method;

[0094] S8. Based on the emotional state feedback data and the students' emotional change trends, S1-S7 are repeatedly executed to form an emotional buffer closed-loop feedback system.

[0095] In this implementation, S1 includes the following sub-steps:

[0096] S11. Collect text information of students’ feedback through the psychological assistant system. The text information includes students’ emotional descriptions. e , psychological stress source T p , Confused Questions T q And other psychological related content o ;

[0097] S12, preprocessing the collected text information T = {Te, Tp, Tq, To} to remove irrelevant punctuation marks, repeated words or non-emotion-related information;

[0098] S13, perform word segmentation, part-of-speech tagging and sentiment feature tagging on the denoised text information set T', and extract key text features F i , forming the emotional expression text dataset D:

[0099] D={F1,F2,…,Fn};

[0100] Among them, F i is the feature vector of the ith text information, and n is the number of student text information collected.

[0101] In this implementation, S2 includes the following sub-steps:

[0102] S21, input the emotion expression text dataset D into the big model, and use the emotion recognition algorithm of the big model to identify each feature vector F i Perform emotional state analysis, map the results of linear transformation into the probability distribution of emotional categories, quantify students' tendencies in different emotional categories, and calculate the emotional probability distribution P i :

[0103]

[0104] in, W j is the emotion category c j The weight vector, b j is the bias term, m is the total number of emotion categories, and p ij Denotes the eigenvector F i Belongs to the emotion category c j The probability of , e is the natural exponent, is a normalization factor so that the sum of the probabilities of all emotion categories is 1;

[0105] S22. According to the emotion probability distribution P i Calculate each emotion category c j Corresponding emotional intensity l j :

[0106] I j =S(p ij )=-β·ln(1-p ij );

[0107] Among them, S(p ij ) is the emotion intensity function, β is the emotion intensity adjustment coefficient, which is used to amplify or reduce the value of emotion intensity to adapt to different application scenarios, and p ij is the eigenvector F i Belongs to the emotion category c j The probability of reflecting the students' emotional tendency in this emotional category, ln is the natural logarithmic function, when p ij When it approaches 1, l j will tend to positive infinity, indicating that the emotional intensity is extremely high, l j is the emotion category c j The emotional intensity, the larger the value, the stronger the student's emotional response in this emotional category;

[0108] S23, each feature vector F i Emotion category c j and its emotional intensity j Combine and construct the student's emotional state feature vector V i .

[0109] In this implementation, S3 includes the following sub-steps:

[0110] S31, collect students' T in the preset time window w The multiple emotional state feature vector set V = {V1, V2, …, V t};

[0111] S32, for each emotion category c in the time window j Emotional intensity sequence Calculate its rate of change

[0112]

[0113] in, is the emotion category c j The rate of change of emotion intensity at time point t, Δt is the time interval, and are the emotion categories c at time points t and t-1 respectively j emotional intensity;

[0114] S33, based on the rate of change of emotion intensity Calculate each emotion category c j In the time window T w The fluctuation range A within w j :

[0115]

[0116] Among them, A j is the emotion category c j The fluctuation range, and Respectively represent the emotion category c in the time window j The maximum and minimum values ​​of emotional intensity;

[0117] S34, by changing the fluctuation range A j With the preset threshold A th Compare and judge whether the student is in a state of greater emotional fluctuation:

[0118]

[0119] Among them, the status j Indicates the emotion category cj Fluctuation state, 1 means in a state of greater emotional fluctuation, 0 means not reaching a state of greater fluctuation;

[0120] S35, integrate the fluctuation states of all emotion categories, and make logical judgment S = state 1, state 2, ..., state m}Evaluate the overall emotional fluctuations of students. If at least one emotional category meets A j ≥A th , it is determined that the student is currently in a state of greater emotional fluctuation and this state is stored.

[0121] In this implementation, S4 includes the following sub-steps:

[0122] S41, input the student's overall emotional fluctuation state set S and the fluctuation range set of emotional categories, and determine the emotional category set C that needs to be buffered through the logic judgment module b :

[0123]

[0124] Among them, A j is the emotion category c j The fluctuation range, δ 2 is the variance threshold of the fluctuation, which is used to filter out the emotional categories with more drastic emotional fluctuations. th is the preset fluctuation amplitude threshold, is the emotion category c j In the time window T w Average emotional intensity within:

[0125]

[0126] S42, for each emotion category c to be buffered j ∈C b , combined with the corresponding emotional intensity Mapping functions by buffering strategy Generate a preliminary buffering strategy B that adapts to emotion category and intensity j :

[0127]

[0128] Among them, B j For the emotion category c j The initial buffer strategy, α1, α2, α3 are the emotion intensity thresholds set according to the applicable scope of the buffer strategy;

[0129] S43, inputting the preliminary buffering strategy set into the large model to generate specific buffering content, including text or audio guidance;

[0130] S44, delivering the generated specific buffer content set to the student's psychological assistant interface, and providing buffer guidance for different emotion categories in real time.

[0131] In this implementation, S5 includes the following sub-steps:

[0132] S51. Based on the emotion category set determined in S4, the buffer content is dynamically generated for each student in combination with the student's historical interaction data and real-time emotional state:

[0133] Multi-sensory guidance: The meditation scene is presented synchronously through text description and audio background. The audio background contains natural sound effects adapted to emotional categories and guides students' breathing rhythm through prompt sounds.

[0134] Virtual scene immersion guidance: A dynamic virtual buffering scene is constructed through a psychological assistant system, and students visualize the buffering process through augmented reality devices or screens;

[0135] S52. Provide active interactive meditation guidance:

[0136] Question-and-answer meditation: During meditation, students provide feedback on their emotional state to the system through voice or text, and the system adjusts the meditation guidance content in real time;

[0137] Rhythmic meditation: Dynamically monitor the student's voice or input rhythm, and adjust the student's voice characteristics or speaking speed to audio feedback that matches the meditation rhythm;

[0138] S53. In the process of mindfulness practice, combine dynamic task-based guidance:

[0139] Mindfulness scanning tasks: step-by-step guides students through focusing on sensory feedback from different parts of the body;

[0140] Mindful Observation Task: Helps students develop concentration and relieve negative emotions by guiding them to observe specific external objects;

[0141] S54. In the deep breathing buffering method, various forms of rhythm control are provided:

[0142] Dynamic visual breathing rhythm graph: Through the psychological assistant interface, a dynamic graph that adjusts with breathing is displayed, and students adjust their breathing rhythm according to the contraction and expansion of the graph;

[0143] Biofeedback control: The device monitors students' breathing rate and heart rate in real time, and feeds back to the buffer content, automatically generating breathing rhythm guidance that matches the students' current physiological state.

[0144] In this implementation, S7 includes the following sub-steps:

[0145] S71, collecting an emotional state feedback data set, each feedback data F in the emotional state feedback data seti ={(c1,ΔI1),(c2,ΔI2),…,(c m ,ΔI m )} includes the emotion intensity change value of the emotion category set:

[0146]

[0147] in, and Respectively represent the intensity of emotions before and after the buffering activity;

[0148] S72, classify and aggregate the emotional state feedback data set, and construct the emotional buffer effect matrix E, the matrix element E ij The calculation is:

[0149] E ij =w j ΔI j ;

[0150] Among them, E ij Indicates that student i is in emotion category c j The buffering effect on j is the emotion category c j The weight factor is set according to the influence of the emotion category on the overall emotion fluctuation of students;

[0151] S73. Input the emotion buffering effect matrix into the meta-learning model to construct a second-order optimization loss function Fine-tune a large model:

[0152]

[0153] Among them, θ represents the parameters of the large model, represents the large model fine-tuned on the feedback data of student i, α1 is the learning rate, represents the emotion buffer loss of student i:

[0154]

[0155] Among them, ΔI j is the emotion category c j The actual intensity of emotions changes, Changes in the intensity of emotions predicted by the model;

[0156] S74. Dynamically adjust the buffer strategy mapping function according to the optimization results of the meta-learning model Adapt buffering strategies to individual students’ emotional characteristics;

[0157] S75. Use the optimized large model for emotion fluctuation state analysis and emotion buffer strategy generation, and record the improvement of the buffer effect to form a buffer effect optimization cycle.

[0158] Embodiment 1:

[0159] Example 1 is applied in the psychological counseling center of the No. 1 Middle School in a certain city. The psychological counseling center received a counseling application from a high school sophomore named Xiao Li. Xiao Li recently showed obvious anxiety in academic pressure and interpersonal communication. He often lost control of his emotions due to trivial events, and had difficulty concentrating and decreased sleep quality. Due to the limited resources of school psychological counselors, Xiao Li was unable to receive timely manual psychological counseling. The psychological counseling center decided to try to introduce a psychological counseling system based on a large model. The system first conducted preliminary psychological intervention and emotional state monitoring, and dynamically managed and analyzed Xiao Li's emotional fluctuations.

[0160] The system first collects Xiao Li's feedback information through the psychological assistant interface. Xiao Li fills out a text form including the current psychological state, learning confusion, and emotional description. In Example 1, Xiao Li gives feedback: "I always feel that I can't finish my homework, I may not do well in tomorrow's exam, and I can't sleep recently." The system performs data preprocessing on this text, removes grammatical noise, and extracts emotion-related keywords such as "can't finish learning", "do not do well in the exam", and "can't sleep". After analyzing the emotional state, the system identifies that Xiao Li's current main emotional category is anxiety, and its emotional intensity value is 0.85 (high anxiety value).

[0161] According to the analysis of the emotion fluctuation monitoring module, Xiao Li's emotion intensity fluctuation amplitude in the past week was 0.65, which was significantly higher than the system preset fluctuation threshold of 0.3. He was judged to be in a state of large emotion fluctuation. Based on the analysis results, the system generated an emotion buffering strategy for Xiao Li, including meditation guidance, mindfulness exercises and dynamic cool-down activities. In the meditation guidance, the system generated a personalized text and audio buffer content: "Please close your eyes, follow the sound of the waves to adjust your breathing, slowly focus on the air flow in your nasal cavity, take a deep breath... exhale slowly..."

[0162] In the mindfulness exercise, the system sets up a task-based mindfulness activity to guide Xiao Li to focus on different parts of the body. The task content is dynamically displayed through the psychological assistant interface. In Example 1, "Now, focus on your feet and feel their contact with the ground. Relax, and then shift your attention to your breathing." At the same time, Xiao Li's participation feedback data in the buffering activity is also collected in real time. In Example 1, the breathing frequency changes and the speech speed changes in the voice feedback are adjusted in real time by the system according to these data. The presentation rhythm and depth of the buffering content.

[0163] After completing the first buffering activity, Xiao Li's emotional intensity dropped to 0.6. The system recorded the emotional changes before and after the buffering activity and updated the data to Xiao Li's emotional profile. Based on the emotional feedback data, the system adjusted the priority of the buffering strategy and placed more emphasis on the guidance of mindfulness exercises and cool-down activities in the following days.

[0164] In order to verify the effectiveness of the system, the psychological counseling center used the traditional psychological counseling method and the method of the present invention to conduct comparative tests on two groups of students. The test subjects were 30 high school students, all of whom had moderate anxiety, and the test period was two weeks. The traditional psychological counseling group was counseled twice a week by a psychological counselor, and each counseling lasted 30 minutes; the system group of the present invention was conducted by the psychological counseling system once a day. The emotional buffering activity lasted 15 minutes each time. The following Table 1 is a specific data comparison:

[0165] Table 1 Comparison of the effects of psychological counseling methods

[0166]

[0167] In the system group of the present invention, students have significantly more flexibility in the time to complete buffer activities and the system achieves dynamic management of emotional states through a real-time feedback mechanism. Compared with traditional methods, not only is the effect of improving emotions significantly improved, but the coverage of psychological counseling is also greatly expanded. In addition, the satisfaction of students in the system group is generally high, and most students say that the content of the buffer activities fits their own emotional needs and is simple and easy to operate.

[0168] This Example 1 fully demonstrates the application effect of the present invention in the management of students' mental health. By introducing a large model and an adaptive learning algorithm system, it realizes accurate identification, dynamic buffering and optimized guidance of students' emotional states. Compared with traditional psychological counseling, the present invention has significant advantages in emotional management efficiency, guidance accuracy and student participation, and provides a feasible intelligent solution for school mental health management.

[0169] The present invention adopts an adaptive optimization framework based on meta-learning, and realizes rapid adaptation to individualized emotional states by dynamically adjusting model parameters through second-order optimization in the generation of emotional buffer strategies and emotional fluctuation analysis. By constructing a second-order optimization loss function, the meta-learning framework can quickly fine-tune the large model under the condition of a small amount of emotional feedback data, so that it can more accurately capture the complex changing relationship between different student emotion categories. Compared with the traditional emotion recognition model based on fixed parameters, the present invention significantly improves the generalization ability of the model under small sample conditions. In addition, by dynamically adjusting the buffer strategy mapping function, the present invention can optimize the generation of buffer content according to the real-time emotional state of the students. Experimental results show that the buffer strategy adaptation efficiency and the emotional fluctuation mitigation rate have been improved by 23% and 18% respectively.

[0170] The present invention comprehensively models the multi-dimensional information of emotion intensity, fluctuation amplitude, and time series characteristics by constructing an emotion buffer effect matrix and an emotion fluctuation state feature vector, thereby effectively solving the problem of single emotion recognition results in the prior art. Through the time weighting mechanism and the dynamic learning rate adjustment strategy, the model can dynamically adjust the focus at different time periods, capture key features at moments when emotions fluctuate greatly, and significantly improve the sensitivity to students' short-term emotion fluctuations and the ability to control long-term emotion trends. Compared with traditional emotion recognition technology, the present invention has outstanding performance in recognition accuracy and time series modeling capabilities, and the overall recognition accuracy of emotion categories is improved by 21%, and the error range in emotion dynamic prediction is reduced to about 65% of traditional methods.

[0171] The present invention constructs an emotion buffering strategy optimization mechanism based on closed-loop feedback, integrates students' real-time emotion feedback into the emotion state feature update and model adaptive learning, and can gradually optimize the accuracy of emotion recognition and the pertinence of the buffering strategy through weighted processing and aggregation analysis of feedback information, thereby realizing dynamic closed-loop optimization of psychological counseling. In the personalized psychological counseling of different students, the present invention can continuously learn students' emotion trigger points and behavior patterns and dynamically adjust the counseling methods, making psychological intervention more targeted and sustainable. Experimental results show that the present invention improves the emotion counseling effect by 26% compared with traditional methods in terms of long-term emotion management and personalized support for students, and significantly improves the psychological counseling satisfaction.

[0172] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A psychological counseling method for students based on a large model, characterized in that: The steps include: S1. By collecting text information of students' feedback and performing data preprocessing on the text information, noise information irrelevant to emotions is removed to obtain text data of emotion expression; S2. Use the big model to analyze the emotional state of the emotional expression text data, use the emotion recognition algorithm to identify the students' emotional state and its characteristics, and establish the emotional state feature vector based on the emotional intensity; S3, by collecting the emotional state feature vector multiple times within a preset time window, calculating the change rate and fluctuation range of the emotional intensity within the time window, and judging whether the student is in a state of large emotional fluctuation; If the fluctuation amplitude exceeds the preset threshold, it is determined that the student is currently in a state of large emotional fluctuation; S4, generating emotional buffering strategies based on short-term emotional fluctuation monitoring results; S5. Provide students with guidance on emotional buffering strategies, provide text or audio guidance, and help students calm down in the short term through meditation, mindfulness exercises or deep breathing methods, and guide their emotional state; S6. After the emotion buffering guidance, collect the students' emotion feedback data, recalculate the students' emotion state feature vector, and store the alleviated emotion state feature data into the students' emotion file; S7. Based on the emotional state feedback data, the adaptive learning algorithm is used to fine-tune the large model, optimize the analysis of emotional fluctuation state and the generation of emotional buffering strategies, and gradually adjust and optimize the emotional buffering method; S8. Based on the emotional state feedback data and the students' emotional change trends, S1-S7 are repeatedly executed to form an emotional buffer closed-loop feedback system.

2. According to claim 1, a psychological counseling method for students based on a large model is characterized in that: The S1 comprises the following sub-steps: S11. Collect text information of students’ feedback through the psychological assistant system. The text information includes students’ emotional descriptions. e 、Psychological stressors T p , Confused Questions T q And other psychological related content o ; S12, collect the text information T = {T e ,T p ,T q ,T o }Preprocessing to remove irrelevant punctuation, repeated words, or non-emotion-related information; S13, perform word segmentation, part-of-speech tagging and sentiment feature tagging on the denoised text information set T', and extract key text features F i , forming the emotional expression text dataset D: D={F1,F2,…,F n }; Among them, F i is the feature vector of the ith text information, and n is the number of student text information collected.

3. According to claim 1, a psychological counseling method for students based on a large model is characterized in that: The S2 comprises the following sub-steps: S21, input the emotion expression text dataset D into the big model, and use the emotion recognition algorithm of the big model to identify each feature vector F i Perform emotional state analysis, map the results of linear transformation into the probability distribution of emotional categories, quantify students' tendencies in different emotional categories, and calculate the emotional probability distribution P i : in, W j is the emotion category c j The weight vector, b j is the bias term, m is the total number of emotion categories, and p ij Denotes the eigenvector F i Belongs to the emotion category c j The probability of , e is the natural exponent, is a normalization factor so that the sum of the probabilities of all emotion categories is 1; S22. According to the emotion probability distribution P i Calculate each emotion category c j Corresponding emotional intensity l j : I j =S(p ij )=-β·ln(1-p ij ); Among them, S(p ij ) is the emotion intensity function, β is the emotion intensity adjustment coefficient, which is used to amplify or reduce the value of emotion intensity to adapt to different application scenarios, and p ij is the eigenvector F i Belongs to the emotion category c j The probability of reflecting the students' emotional tendency in this emotional category, ln is the natural logarithmic function, when p ij When it approaches 1, l j will tend to positive infinity, indicating that the emotional intensity is extremely high, l j is the emotion category c j The emotional intensity, the larger the value, the stronger the student's emotional response in this emotional category; S23, each feature vector F i Emotion category c j and its emotional intensity j Combine and construct the student's emotional state feature vector V i .

4. According to claim 1, a psychological counseling method for students based on a large model is characterized in that: The S3 comprises the following sub-steps: S31, collect students' T in the preset time window w The multiple emotional state feature vector set V = {V1, V2, …, V t }; S32, for each emotion category c in the time window j Emotional intensity sequence Calculate its rate of change in, is the emotion category c j The rate of change of emotion intensity at time point t, Δt is the time interval, and are the emotion categories c at time points t and t-1 respectively j emotional intensity; S33, based on the rate of change of emotion intensity Calculate each emotion category c j In the time window T w The fluctuation range A within w j : Among them, A j is the emotion category c j The fluctuation range, and Respectively represent the emotion category c in the time window j The maximum and minimum values ​​of emotional intensity; S34, by changing the fluctuation range A j With the preset threshold A th Compare and judge whether the student is in a state of greater emotional fluctuation: Among them, the status j Indicates the emotion category c j Fluctuation state, 1 means in a state of greater emotional fluctuation, 0 means not reaching a state of greater fluctuation; S35, integrate the fluctuation states of all emotion categories, and make logical judgment S = state 1, state 2, ..., state m }Evaluate the overall emotional fluctuations of students. If at least one emotional category meets A j ≥A th , it is determined that the student is currently in a state of greater emotional fluctuation and this state is stored.

5. The method for psychological counseling for students based on a large model according to claim 1, characterized in that: The S4 comprises the following sub-steps: S41, input the student's overall emotional fluctuation state set S and the fluctuation range set of emotional categories, and determine the emotional category set C that needs to be buffered through the logic judgment module b : Among them, A j is the emotion category c j The fluctuation range, δ 2 is the variance threshold of the fluctuation, which is used to filter out the emotional categories with more drastic emotional fluctuations. th is the preset fluctuation amplitude threshold, is the emotion category c j In the time window T w Average emotional intensity within: S42, for each emotion category c to be buffered j ∈C b , combined with the corresponding emotional intensity Mapping functions via buffering strategy Generate a preliminary buffering strategy B that adapts to emotion category and intensity j : Among them, B j For the emotion category c j The initial buffer strategy, α1, α2, α3 are the emotion intensity thresholds set according to the applicable scope of the buffer strategy; S43, inputting the preliminary buffering strategy set into the large model to generate specific buffering content, including text or audio guidance; S44, delivering the generated specific buffer content set to the student's psychological assistant interface, and providing buffer guidance for different emotion categories in real time.

6. The method for psychological counseling for students based on a large model according to claim 1, characterized in that: The S5 comprises the following sub-steps: S51. Based on the emotion category set determined in S4, the buffer content is dynamically generated for each student in combination with the student's historical interaction data and real-time emotional state: Multi-sensory guidance: The meditation scene is presented synchronously through text description and audio background. The audio background contains natural sound effects adapted to emotional categories and guides students' breathing rhythm through prompt sounds. Virtual scene immersion guidance: A dynamic virtual buffering scene is constructed through a psychological assistant system, and students visualize the buffering process through augmented reality devices or screens; S52. Provide active interactive meditation guidance: Question-and-answer meditation: During meditation, students provide feedback on their emotional state to the system through voice or text, and the system adjusts the meditation guidance content in real time; Rhythmic meditation: Dynamically monitor the student's voice or input rhythm, and adjust the student's voice characteristics or speaking speed to audio feedback that matches the meditation rhythm; S53. In the process of mindfulness practice, combine dynamic task-based guidance: Mindfulness scanning tasks: step-by-step guides students through focusing on sensory feedback from different parts of the body; Mindful Observation Task: Helps students develop concentration and relieve negative emotions by guiding them to observe specific external objects; S54. In the deep breathing buffering method, various forms of rhythm control are provided: Dynamic visual breathing rhythm graph: Through the psychological assistant interface, a dynamic graph that adjusts with breathing is displayed, and students adjust their breathing rhythm according to the contraction and expansion of the graph; Biofeedback control: The device monitors students' breathing rate and heart rate in real time, and feeds back to the buffer content, automatically generating breathing rhythm guidance that matches the students' current physiological state.

7. The method for psychological counseling for students based on a large model according to claim 1, characterized in that: The S7 comprises the following sub-steps: S71, collecting an emotional state feedback data set, each feedback data F in the emotional state feedback data set i ={(c1,ΔI1),(c2,ΔI2),…,(c m ,ΔI m )} includes the emotion intensity change value of the emotion category set: in, and Respectively represent the intensity of emotions before and after the buffering activity; S72, classify and aggregate the emotional state feedback data set, and construct the emotional buffer effect matrix E, the matrix element E ij The calculation is: E ij =w j ·ΔI j ; Among them, E ij Indicates that student i is in emotion category c j The buffering effect on j is the emotion category c j The weight factor is set according to the influence of the emotion category on the overall emotion fluctuation of students; S73. Input the emotion buffering effect matrix into the meta-learning model to construct a second-order optimization loss function Fine-tune a large model: Among them, θ represents the parameters of the large model, represents the large model fine-tuned on the feedback data of student i, α1 is the learning rate, represents the emotion buffer loss of student i: Among them, ΔI j is the emotion category c j The actual intensity of emotions changes, Changes in the intensity of emotions predicted by the model; S74. Dynamically adjust the buffer strategy mapping function according to the optimization results of the meta-learning model Adapt buffering strategies to individual students’ emotional characteristics; S75. Use the optimized large model for emotion fluctuation state analysis and emotion buffer strategy generation, and record the improvement of the buffer effect to form a buffer effect optimization cycle.

Images