Education psychological crisis early warning and intervention platform
By constructing an educational psychological crisis early warning and intervention platform, utilizing CRITIC weight analysis and machine learning models, combined with a multimodal knowledge base and multi-round empathic dialogue interaction, the platform solves the problem that traditional systems cannot provide early warning and intervention for psychological crises among Generation Z college students, achieving higher early warning accuracy and user satisfaction.
Patent Information
- Application Number
- CN202510504502.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-20
- Publication Date
- 2025-11-07
AI Technical Summary
Existing mental health assessment systems for college students are unable to provide effective early warning and intervention for mental health crises tailored to the characteristics of Generation Z college students. Traditional models do not fully consider the differences in the growth environment and psychological state of Generation Z.
An educational psychological crisis early warning and intervention platform was constructed, including a data collection module, a data processing and analysis module, an early warning and intervention module, and a user interaction module. It adopts CRITIC weight analysis, machine learning models, and RAG large models, combined with a multimodal mental health knowledge base, and provides multi-terminal access interfaces and multi-round empathetic dialogue interaction. The system response speed is optimized through asynchronous processing technology.
It improves the accuracy of psychological crisis early warning, reduces false alarm rate, enhances user consultation satisfaction and retention rate, ensures user privacy and security, provides personalized intervention suggestions and real-time response, and adapts to the mental health needs of Generation Z college students.
Smart Images
Figure CN120913766A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of psychological early warning and intervention, in particular to an education psychological crisis early warning and intervention platform. BACKGROUND
[0002] With the sound of the clock in 2024, almost all the undergraduate students are the image of Z generation. Compared with the millennium generation, Z generation is more affected by mobile Internet; the time of outdoor activities is shorter; the interpersonal relationship of face-to-face is more sensitive... In short, the growth environment of Z generation is different from that of the millennium generation, and there is a certain difference in psychological condition. The establishment and perfection of the traditional university student psychological crisis early warning and intervention model pay less attention to this point, and even pay more attention to the research of the millennium generation. Therefore, it is necessary to adapt to the development of the times, use big data technology, collect, analyze and process the basic information data of college students, teaching management and student management data information on the basis of traditional college student psychological health evaluation system, extract the main characteristic quantity of college students' psychological health problems according to the primary and secondary relationship of the role played by each aspect and the influence degree of college students, especially Z generation college students' psychological crisis, give different weights, explore to establish a psychological crisis early warning model index system containing traditional evaluation information, student basic information, daily behavior information data, use neural network technology to classify and judge Z generation college students' psychological crisis, and build Z generation college students' psychological crisis intelligent diagnosis model, create more real-time and applicable methods and tools for Z generation college students' psychological health screening analysis, psychological crisis early warning and even later intervention.
[0003] In view of this, an education psychological crisis early warning and intervention platform is provided. SUMMARY
[0004] The purpose of the present application is to provide an education psychological crisis early warning and intervention platform to solve the problem that the existing psychological health evaluation system cannot perform psychological crisis early warning and intervention according to the characteristics of Z generation college students.
[0005] To solve the above technical problems, the present application provides an education psychological crisis early warning and intervention platform, comprising:
[0006] A data acquisition module is used to acquire the basic information, psychological health evaluation data, daily behavior data and teaching management data of the user;
[0007] A data processing and analysis module is connected with the data acquisition module, and performs feature extraction and weight distribution on the collected data based on the CRITIC weight analysis method, and classifies and judges the user's psychological health state in combination with a machine learning model;
[0008] The early warning and intervention module is connected with the data processing and analysis module, integrates a psychological crisis early warning model and a RAG large model, and realizes semantic retrieval and dynamic suggestion generation through construction of a multi-modal psychological health knowledge base.
[0009] The user interaction module is connected with the early warning and intervention module, provides a multi-terminal access interface, includes units of psychological assessment, online psychological counseling, psychological knowledge popularization, anonymous hole, decompression game, personal center function, and multi-round empathetic dialogue interaction, and optimizes system response speed through asynchronous processing technology.
[0010] As a further improvement of the technical solution, the basic information includes gender, place of origin, family upbringing, exercise habit, academic record and leave frequency.
[0011] As a further improvement of the technical solution, in the data processing and analysis module, the weight allocation method includes: using the CRITIC algorithm to calculate the information amount and conflict of the features, and determining the weight; the parental upbringing, whether participating in sports and the leave frequency are given a weight higher than the preset value as the core early warning index.
[0012] As a further improvement of the technical solution, the machine learning model filters features significantly related to mental health through chi-square test, and pre-processes data by using positive and mean value methods.
[0013] As a further improvement of the technical solution, the machine learning model is a Lingxin model based on ChatGLM-6B base model fine-tuning, which realizes the functions of empathy guidance, confession mining and professional suggestion generation by training joint single-round long text data set and multi-round dialogue data set.
[0014] As a further improvement of the technical solution, the RAG large model includes a knowledge base, and the knowledge base includes psychological health articles, clinical guidelines, case studies and psychological assessment tools.
[0015] The knowledge base construction method includes: converting the psychological health articles, clinical guidelines, case studies and assessment tool texts into vector representations, and storing them by using the inverted index technology; updating the knowledge base content in real time through regular incremental training, and dynamically adjusting the retrieval strategy combined with user evaluation results and counseling records.
[0016] As a further improvement of the technical solution, the user interaction module includes a front end and a back end, the front end realizes dynamic page rendering by using HTML, CSS and JavaScript; the back end is built based on a Spring Boot framework and a MySQL database, and deploys a RESTful API interface through an Aliyun server.
[0017] As a further improvement of the technical solution, the psychological evaluation unit integrates SCL-90 symptom self-rating scale, calculates the user's mental health score through total score and factor score, and generates a personalized evaluation report and intervention suggestion in combination with the large model; the online psychological counseling unit adopts multi-round progressive dialogue, manages user historical sessions through context caching technology, and sets up a high-risk keyword triggered artificial intervention mechanism; the anonymous confidant unit adopts encryption storage technology to protect user privacy, allows users to record emotional diaries in text form, and links with the psychological counseling unit data; the decompression game unit includes an emotion recognition and stress relief interactive program, and optimizes the game recommendation strategy in combination with user behavior data.
[0018] As a further improvement of the technical solution, the user interaction module further includes an information intercommunication unit, which realizes data intercommunication with the school educational administration system through student ID binding, and synchronizes the psychological abnormal user information to the school management terminal in real time.
[0019] Compared with the prior art, the present application has the following advantages:
[0020] 1. In the education psychological crisis early warning and intervention platform, the core risk factors of the mental health of the Z generation college students are accurately captured by dynamically allocating weights through calculating the characteristic variability and conflict, and the characteristics such as "whether to participate in sports" and "leave frequency" which are significantly related to psychological crisis are screened out, so as to avoid redundant data interference, improve the model warning accuracy, and reduce the false positive rate.
[0021] 2. In the education psychological crisis early warning and intervention platform, based on ChatGLM-6B fine-tuning, single-round long text and multi-round dialogue data sets are fused to support progressive empathy guidance, a vectorized knowledge base containing psychological health articles, clinical guidelines and case studies is constructed, professional suggestions are generated through semantic retrieval, and the user consultation satisfaction is improved.
[0022] 3. In the education psychological crisis early warning and intervention platform, the user diary is encrypted by using encryption storage technology to ensure privacy security, and is linked with the psychological counseling data to assist the model in identifying long-term emotional changes, recommend emotion regulation games in combination with user behavior data, improve the daily use time, and thus improve the user retention rate. BRIEF DESCRIPTION OF DRAWINGS
[0023] Figure 1 The figure is a schematic diagram of the system structure of the present application. DETAILED DESCRIPTION
[0024] With reference to the drawings of the embodiments of the present application, the technical solutions in the embodiments of the present application will be described clearly and completely. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments of the present application, all the secondary embodiments obtained by those skilled in the art without creative work fall within the scope of the present application.
[0025] At present, the traditional college student psychological crisis early warning and intervention model does not construct the growth environment and psychological condition of Z generation college students, but pays more attention to the research of the millennium generation, therefore, it is necessary to utilize the neural network technology to classify and distinguish the psychological crisis of Z generation college students, construct an intelligent diagnosis model of the psychological crisis of Z generation college students, and create a more real-time and applicable method and tool for the psychological health screening analysis, psychological crisis early warning and even later intervention of Z generation college students.
[0026] In view of this, please refer to Figure 1 The purpose of the present application is to provide an education psychological crisis early warning and intervention platform, which comprises:
[0027] A data acquisition module is configured to acquire basic information, psychological health evaluation data, daily behavior data and educational administration data of a user.
[0028] A data processing and analysis module is connected with the data acquisition module, configured to perform feature extraction and weight distribution on the acquired data based on a CRITIC weight analysis method, and perform classification and discrimination on the psychological health state of the user in combination with a machine learning model.
[0029] An early warning and intervention module is connected with the data processing and analysis module, configured to integrate a psychological crisis early warning model and a RAG large model, and realize semantic retrieval and dynamic suggestion generation through construction of a multi-modal psychological health knowledge base.
[0030] A user interaction module is connected with the early warning and intervention module, configured to provide a multi-terminal access interface, including units of psychological evaluation, online psychological counseling, psychological knowledge popularization, anonymous hole, decompression game, personal center function and multi-round empathetic dialogue interaction, and optimize the system response speed through asynchronous processing technology.
[0031] In the education psychological crisis early warning and intervention platform, the core risk factors of the mental health of Generation Z college students are accurately captured by calculating the variability and conflict of the characteristics, dynamically allocating weights, screening characteristics such as "whether to participate in sports" and "leave frequency" that are significantly related to psychological crisis, avoiding redundant data interference, improving the accuracy of model warning, and reducing the false positive rate; at the same time, based on ChatGLM-6B fine-tuning, combining single-round long text and multi-round dialogue data sets, supporting progressive empathy guidance, building a vectorized knowledge base containing mental health articles, clinical guidelines, and case studies, and generating professional advice through semantic retrieval to improve user consultation satisfaction.
[0032] Considering that the gender ratio (more men than women) of Generation Z college students may affect group psychological characteristics, such as men being more inclined to hide emotional problems and requiring targeted intervention; students in remote areas (such as the University of Jieyang campus) are more at risk of psychological crisis due to limited access to resources and lack of social support; parental rearing style is a core factor affecting mental health; whether to participate in sports is significantly related to mental health, and students with low exercise frequency have a higher rate of psychological abnormalities; academic pressure is a common inducement for psychological problems among Generation Z college students, and academic performance can be quantitatively related to psychological state in combination with grade point average; high leave frequency is highly related to psychological abnormalities and is a key indicator of the early warning model; therefore, in the data collection module, the user's basic information includes gender, place of origin, family upbringing, exercise habits, academic performance, and leave frequency;
[0033] Due to the objectivity and scientificity of the CRITIC algorithm, the dispersion degree of the characteristics reflects the ability of Generation Z college students to distinguish between psychological states, and the negative correlation between characteristics avoids redundant information interference, and after weight normalization, high information content characteristics are given higher priority, therefore, in the data processing and analysis module, the weight allocation method includes: using the CRITIC algorithm to calculate the information content and conflict of the characteristics to determine the weight;
[0034] The CRITIC algorithm process is as follows:
[0035] Step 1: Calculate the standard deviation (variability) and correlation coefficient matrix (conflict) of each characteristic.
[0036] Step 2: Information content = variability x (1 - average correlation coefficient), such as "parental rearing style" information content = 2.013 x 17.841 = 35.911.
[0037] Step 3: Weight normalization, formula: weight = information content / total information content x 100%.
[0038] Meanwhile, to ensure the CRITIC weight focuses on key indicators, the machine learning model filters features significantly related to mental health through the chi-square test and pre-processes data using normalization and mean value methods; that is, assigning a higher weight than the preset value (e.g., ≥10%) to parental rearing style, participation in sports, and leave frequency as core early warning indicators; filtering out significantly related features (e.g., "leave frequency" X 2 = 28.537, p = 0.000) through the chi-square test (<0.01), thus ensuring the CRITIC weight focuses on key indicators;
[0039] Through CRITIC weight allocation and chi-square test, the model's sensitivity to core indicators such as "parental rearing style" and "leave frequency" is improved, enhancing the accuracy of early warning. The weight allocation results (e.g., "parental rearing style" weight) explicitly quantify feature contributions, assisting school authorities in developing targeted intervention strategies (e.g., family communication coaching). Meanwhile, dynamic weight adjustment (combined with incremental data updates) allows the model to adapt to Z generation behavior changes, such as the impact of leave frequency fluctuations during the epidemic on mental state, which can quickly respond to the early warning system. Furthermore, intervention suggestions driven by high-weight features (e.g., exercise habits) such as recommended exercise plans reduce the proportion of ineffective interventions and improve school resource utilization.
[0040] Since existing Lingxin models are primarily based on foreign language counseling scenarios, Chinese is vast and profound, and existing Lingxin models are difficult to understand. Considering that ChatGLM-6B is a large open-source model optimized specifically for Chinese, it supports long text understanding and generation, can accurately capture emotional expressions in the Chinese context (such as "I understand your feelings" and other natural responses), and meets the language needs of the counseling scenario. Meanwhile, ChatGLM-6B supports full-parameter instruction fine-tuning, allowing the model's performance to be optimized using domain data (psychological counseling corpus), avoiding the computational cost of training from scratch. Therefore, the machine learning model is a Lingxin model based on ChatGLM-6B base model fine-tuning, which is trained through joint single-round long text dataset and multi-round dialogue dataset to realize empathy guidance, confession mining, and professional suggestion generation functions. The single-round long text dataset is 150,000 instructions and 500,000 high-quality answers, focusing on deep psychological analysis and suggestion generation (such as depression intervention strategies). The multi-round dialogue dataset is generated through ChatGPT and GPT-4 for 1 million rounds of dialogue, simulating progressive interaction in real counseling (such as "Can you tell me more about your feelings?"). By combining single-round and multi-round data (a total of 1.2 million samples), the balance between empathy guidance and professional advice capabilities is achieved, avoiding user fatigue caused by pure single-round training (long replies) and the weakening of multi-round data suggestions, and improving the naturalness of the conversation.
[0041] To enhance the professionalism and authority of the recommendations, the RAG large model includes a knowledge base that includes mental health articles, clinical guidelines, case studies, and psychological assessment tools. Mental health articles cover the causes, symptoms, and self-help strategies for common mental health problems such as depression and anxiety, providing a broad theoretical basis for the model (such as the principles of cognitive behavioral therapy). Clinical guidelines integrate international authoritative guidelines (such as the WHO mental health intervention standard) to ensure that the recommendations meet medical standards (such as the referral process for patients with severe depression). Case studies include real counseling cases (such as the intervention record of anxiety caused by academic pressure), helping the model learn coping strategies for complex scenarios. Psychological assessment tools embed standardized scales such as SCL-90 and PHQ-9 to support dynamic assessment of user mental state (such as SCL-90 factor score > 3 points triggering high-risk warning).
[0042] Considering that traditional keyword matching cannot capture semantic associations (such as the deep connection between "low mood" and "depression symptoms"), and the complex and variable terminology in the field of mental health, precise semantic understanding is required. Therefore, the knowledge base construction method includes: converting the text of mental health articles, clinical guidelines, case studies, and assessment tools into vector representations, such as using BERT for vector conversion, using inverted index technology for storage, such as using tools such as Faiss to achieve efficient retrieval, thereby improving the accuracy of semantic retrieval and improving the recall rate of high-risk cases.
[0043] Due to the rapid iteration of knowledge in the field of mental health (such as newly released clinical guidelines), static knowledge bases are prone to become outdated and cannot adapt to the dynamic needs of Generation Z. At the same time, user evaluation results (such as SCL-90 factor scores) and counseling records (such as high-frequency confession keywords) reflect individualized needs, and the retrieval priority needs to be dynamically optimized. Through regular incremental training, the knowledge base content is updated in real time, for example, every quarter, incremental data is extracted from new literature and user logs to update vector indexes, and retrieval strategies are dynamically adjusted based on user evaluation results (such as SCL-90 scores) and counseling high-frequency keywords, such as weighting the recall priority of related documents based on user evaluation results (such as SCL-90 scores) and counseling high-frequency keywords.
[0044] The user interaction module includes a front end and a back end. Considering the preference of Generation Z college students for multi-terminal (PC / mobile terminal) access, the interface needs to be adaptive and the interaction needs to be smooth. Therefore, the front end uses HTML, CSS and JavaScript to realize dynamic page rendering, so that the page loading speed is fast, the interface is adaptive and meets the aesthetic effect of college students. Considering the requirements of high concurrency access and data security, the back end is built based on the Spring Boot framework and MySQL database, supporting high QPS concurrent requests, low response delay, rapid deployment and system stability. Therefore, the RESTful API interface is deployed through the Aliyun server, making the service availability almost perfect, and supporting elastic scaling and real-time monitoring.
[0045] In order to standardize and comprehensiveness of mental health assessment, SCL-90 symptom self-rating scale technology is adopted in the psychological assessment unit. The user's mental health score is calculated by total score and factor score (such as depression and anxiety factor), which can accurately quantify the psychological state and improve the accuracy of abnormal detection.
[0046] Considering that traditional evaluation reports lack personalized guidance, the RAG large model combined with knowledge base retrieval technology is adopted in the psychological assessment unit to generate personalized evaluation reports and intervention suggestions (such as recommending CBT exercises), which improves user satisfaction (NPS) and improves suggestion matching degree.
[0047] Considering that traditional single-round dialogue is difficult to guide users to deep confession, the multi-round progressive dialogue management technology based on Lingxin model is adopted in the online psychological counseling unit. Through context caching technology (such as Redis technology), the user's historical conversation context is cached, which improves the average dialogue round and increases the user's initiative to confess.
[0048] Considering that emergency psychological crisis needs immediate intervention, the high-risk keyword real-time detection and artificial intervention triggering technology is adopted in the online psychological counseling unit. Through the NLP rule engine and school API linkage, the response time of high-risk users is shortened and the efficiency of crisis event handling is improved.
[0049] Due to user privacy protection and psychological data sensitivity, encryption storage technology (such as AES encryption algorithm and SSL transmission technology) is adopted in the anonymous confession unit to perform end-to-end encryption storage of user emotion diary, thereby reducing the risk of data leakage and improving user trust. Emotion diary needs to assist psychological counseling for accurate intervention, so natural language processing (NLP) keyword extraction technology is adopted in the anonymous confession unit, combined with data linkage analysis in the psychological counseling unit, to improve the accuracy of emotion trend identification and shorten the response time of high-risk user warning.
[0050] Considering that the traditional decompression tool lacks dynamic adaptation and user behavior feedback, therefore, in the decompression game unit, an emotion recognition model (such as CNN facial expression analysis) and user behavior data mining technology are adopted, combined with a collaborative filtering algorithm to optimize the recommendation strategy, improve the game recommendation matching degree, and improve the user stress relief efficiency; at the same time, considering that the interactive program needs to improve the user participation, therefore, in the decompression game unit, real-time stress feedback technology (such as heart rate monitoring API) and dynamic difficulty adjustment algorithm are adopted, which prolongs the average game time of the user and improves the emotional relaxation score;
[0051] Since the psychological abnormal user needs to be intervened by the school in time, therefore, the user interaction module also includes an information intercommunication unit, in which a student ID binding and a teaching management system API interface technology are adopted, the high-risk results of psychological evaluation and the counseling records are synchronized in real time through the RESTful API, the response time of the school management end is shortened; at the same time, since the student privacy and data security requirements, in the information intercommunication unit, HTTPS encryption transmission and permission hierarchical control technology are adopted, the synchronized data is end-to-end encrypted and permission isolated, the information leakage risk is reduced, and the compliance rate of the school data access is improved.
[0052] The above shows and describes the basic principles, main features and advantages of the present application. Those skilled in the art should understand that the present application is not limited to the above-mentioned embodiments, and the above-mentioned embodiments and descriptions in the specification are only preferred examples of the present application and are not intended to limit the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the present application. The scope of protection of the present application is defined by the appended claims and their equivalents.
Claims
1. An education psychological crisis early warning and intervention platform, characterized in that, The application relates to a mental health monitoring system, comprising: a data collection module for collecting basic information, mental health evaluation data, daily behavior data and educational administration data of a user; a data processing and analysis module connected with the data collection module, which performs feature extraction and weight distribution on the collected data based on a CRITIC weight analysis method, and classifies the mental health state of the user in combination with a machine learning model; a warning and intervention module connected with the data processing and analysis module, which integrates a psychological crisis warning model and a RAG large model, and realizes semantic retrieval and dynamic suggestion generation through the construction of a multi-modal mental health knowledge base; a user interaction module connected with the warning and intervention module, which provides a multi-terminal access interface, including a psychological evaluation, online psychological counseling, psychological knowledge popularization, anonymous hole, decompression game, personal center function and multi-round empathetic dialogue interaction unit, and optimizes the system response speed through asynchronous processing technology.
2. The education psychological crisis early warning and intervention platform according to claim 1, characterized in that: The basic information includes gender, place of origin, family upbringing, exercise habit, academic record and leave frequency.
3. The education psychological crisis early warning and intervention platform according to claim 1, characterized in that, In the data processing and analysis module, the weight distribution method comprises: calculating the information amount and conflict of the features by using a CRITIC algorithm to determine the weight; and giving the parental upbringing, participation in sports and leave frequency a weight higher than a preset value as a core warning index.
4. The education psychological crisis early warning and intervention platform according to claim 1, characterized in that: The machine learning model screens features significantly related to mental health through a chi-square test, and pre-processes the data by using a positive direction and mean value method.
5. The education psychological crisis early warning and intervention platform according to claim 4, characterized in that, The machine learning model is a Lingxin model based on the fine-tuning of a ChatGLM-6B base model, which realizes the functions of empathy guidance, confession mining and professional suggestion generation by training a single round of long text data set and a multi-round dialogue data set.
6. The education psychological crisis early warning and intervention platform according to claim 1, characterized in that: The RAG large model comprises a knowledge base, and the knowledge base comprises mental health articles, clinical guidelines, case studies and psychological evaluation tools. The knowledge base construction method comprises: converting the mental health articles, clinical guidelines, case studies and evaluation tool texts into vector representations and storing them by using an inverted index technology; and updating the knowledge base content in real time through regular incremental training, and dynamically adjusting the retrieval strategy in combination with the user evaluation results and counseling records.
7. The education psychological crisis early warning and intervention platform according to claim 1, characterized in that: The user interaction module comprises a front end and a back end, the front end realizes dynamic page rendering by using HTML, CSS and JavaScript; and the back end is built based on a Spring Boot framework and a MySQL database, and deploys a RESTful API interface through an Aliyun server.
8. The education psychological crisis early warning and intervention platform according to claim 1, characterized in that: The psychological assessment unit integrates the SCL-90 symptom self-rating scale, calculates the user's mental health score through the total score and factor score, and generates a personalized evaluation report and intervention suggestion in combination with a large model; The online psychological counseling unit adopts a multi-round progressive dialogue, manages user historical sessions through context caching technology, and sets up a high-risk keyword triggered artificial intervention mechanism; The anonymous confidant unit adopts encryption storage technology to protect user privacy, allows users to record emotional diaries in text form, and is linked with the psychological counseling unit data; The decompression game unit includes an emotion recognition and stress relief interactive program, and optimizes the game recommendation strategy in combination with user behavior data.
9. The education psychological crisis early warning and intervention platform according to claim 1, characterized in that: The user interaction module also includes an information intercommunication unit, which realizes data intercommunication with the school educational administration system through student ID binding, and synchronizes the information of the psychological abnormal user to the school management end in real time.