Family-school psychological health cooperation method and system

Abstract

The invention relates to the technical field of psychological health intelligent evaluation and collaborative intervention, and discloses a family-school psychological health collaborative method and system.The method comprises the steps that a multi-dimensional data model is built based on the psychological theory, and family-school double-end data collection is achieved through standardized indexes; constructing a multi-dimensional psychological state diagram structure, and realizing home-school data deep fusion by adopting a time sequence dynamic coding method; a time sequence modeling and multi-level risk assessment method is adopted, and a dynamic mental health risk assessment result and an intelligent early warning signal are output; a cross-platform collaborative intervention management method is adopted, and a personalized intervention strategy and a collaborative task allocation scheme are output; and a multi-dimensional effect evaluation and reinforcement learning optimization method is adopted to realize continuous optimization of an intervention strategy. Through artificial intelligence technologies such as the dynamic graph neural network and deep reinforcement learning, intelligent fusion and cooperative intervention of family-school psychological health data are realized, and the early recognition accuracy and intervention effect of psychological health problems are improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent assessment and collaborative intervention technology for mental health, and more specifically, to a method and system for collaborative mental health between home and school. Background Technology

[0002] With social development and changes in the educational environment, students' mental health issues are becoming increasingly prominent. Traditional mental health assessments mainly rely on the subjective judgment of mental health counselors and single-scale tests, lacking a multi-dimensional and dynamic assessment system. Current technologies for mental health assessments often employ static questionnaires or periodic psychological tests, which cannot capture the dynamic changes in students' mental states in real time. Furthermore, assessment results often lag behind the occurrence of actual mental health problems, making early warning and timely intervention difficult.

[0003] Regarding home-school collaboration, families and schools currently tend to focus on students' mental health independently, lacking effective information sharing and coordination mechanisms. Parents and teachers observe students' mental states from different perspectives, resulting in information that is both differentiated and complementary. However, existing technologies cannot effectively integrate the observational data from both families and schools, leading to incomplete and inaccurate mental health assessment information. Furthermore, when students are identified as having mental health risks, the lack of unified intervention strategies and coordination mechanisms between families and schools often results in each school taking its own measures, leading to poor outcomes and potential conflicts.

[0004] Existing technologies lack intelligent, multi-dimensional psychological state modeling methods, failing to comprehensively and accurately describe students' mental health status; they lack technical means for integrating home-school data, hindering the effective integration of multi-source, heterogeneous mental health data; and they lack dynamic risk assessment and early warning mechanisms, making it difficult to promptly detect potential mental health problems. Therefore, there is an urgent need for a technical solution that can achieve intelligent integration of home-school mental health data, dynamic risk assessment, and collaborative intervention. Summary of the Invention

[0005] This invention provides a method and system for home-school mental health collaboration, which solves the technical problems of difficulty in describing students' psychological state and lack of home-school collaboration in related technologies.

[0006] This invention provides a method for home-school collaboration in mental health, comprising the following steps: S1, based on psychological theory, constructs a multi-dimensional data model, uses standardized indicators to collect data from both home and school, and outputs a structured dataset of psychological states; S2, based on the structured dataset of psychological states, constructs a multidimensional psychological state graph structure, and uses a time-series dynamic coding method to achieve deep integration of home and school data, outputting a comprehensive psychological state feature vector; S3, based on a comprehensive psychological state feature vector, uses time series modeling and multi-level risk assessment methods to output dynamic mental health risk assessment results and intelligent early warning signals; S4, based on intelligent early warning signals, adopts a cross-platform collaborative intervention management method to output personalized intervention strategies and collaborative task allocation schemes; S5, based on personalized intervention strategies, employs multi-dimensional effect evaluation and reinforcement learning optimization methods to output intervention effect evaluation reports and strategy optimization suggestions.

[0007] In a preferred embodiment, the construction of the multi-dimensional data model based on psychological theory includes: A multi-dimensional psychological state description framework is adopted, which decomposes students' psychological state into four core dimensions: emotional, behavioral, cognitive, and environmental dimensions; and establishes a standardized indicator system and scoring mechanism for each dimension. Establish a unified 5-level scoring system, and form a dimensional state vector by arranging the scores of each indicator in order.

[0008] In a preferred embodiment, constructing a multidimensional mental state graph structure based on a structured mental state dataset includes: A heterogeneous graph construction method is used to encode students, teachers, parents, and four-dimensional psychological state indicators as nodes and edge relationships in a graph; Create a central node for each student, an observer node for each teacher and parent, and nodes for the four psychological dimensions of emotion, behavior, cognition, and environment. Establish directed observation edges between teachers and students, and between parents and students; establish connection edges between students and the four psychological dimensions; and establish correlation edges between dimensions based on psychological theories. Normalization is used to map node feature values ​​to the 0-1 interval, and all edge weights are normalized to the 0-1 interval to generate a standardized adjacency matrix and weight matrix.

[0009] In a preferred embodiment, the method of time series modeling and multi-level risk assessment based on a comprehensive psychological state feature vector includes: A time-series prediction model is constructed using a long short-term memory network, which simultaneously captures short-term fluctuations and long-term trends in mental states through a multi-layered network structure. A multi-level risk assessment algorithm is used to comprehensively assess mental health risks from three time dimensions: immediate, short-term, and long-term. Immediate risk assessment analyzes the degree of deviation between the current mental state and the individual's historical baseline; short-term risk assessment is based on the prediction of the trend of changes in mental health status in the next few weeks; and long-term risk assessment collects historical risk assessment data and uses time decay weights. An adaptive threshold setting algorithm is adopted to set individualized early warning thresholds based on the distribution characteristics of individual historical risk scores, and generate intelligent early warning signals that include risk type, urgency level and trigger time.

[0010] In a preferred embodiment, the cross-platform collaborative intervention management method based on intelligent early warning signals includes: Key features of historical successful cases are extracted from the intervention case knowledge base. The similarity between the current student situation and historical cases is calculated, and the case with the highest similarity is selected as the basis for strategy recommendation. Complex intervention strategies are broken down into actionable task units, and different types of intervention activities are identified through strategy content analysis. Intelligent algorithms are used to achieve optimal matching between tasks and execution entities, and each execution entity is evaluated from the dimensions of professional capabilities, available time, workload, and collaboration capabilities.

[0011] In a preferred embodiment, the method of using multi-dimensional effect evaluation and reinforcement learning optimization based on personalized intervention strategies includes: By comparing and analyzing the direct effects of the intervention, we analyzed the mental health status assessment data of students before and after the intervention, and quantified the degree of improvement in each dimension of mental health status. By using Markov decision process modeling, the characteristics of students' mental health status, family environment, and school environment are defined as the state space, and the intervention strategies, intervention intensity, and intervention frequency are constructed as the action space. A reward function is designed based on the intervention effect evaluation results. The state-action value function is learned through deep reinforcement learning algorithms, and the probability distribution of intervention strategy selection is optimized using the policy gradient method. Build personalized mental health profiles to record the entire process from problem discovery to intervention completion, and integrate multi-dimensional data from psychological assessment, intervention strategies, implementation process, and effect evaluation.

[0012] In a preferred embodiment, the deep integration of home-school data using a time-series dynamic coding method includes: A dynamic graph neural network is used to encode the graph structure evolution in the time series. The observed data is divided into time windows according to the timestamp information, and a corresponding graph snapshot is constructed for each time window to form a time series graph sequence. Graph convolution operations are performed on each graph in the temporal graph sequence. Local and global structural information is captured through a multi-layer graph convolutional network to obtain node embedding representations. A recurrent neural network is used to capture the temporal evolution pattern of graph sequences, and an attention mechanism is combined to identify important time points to generate a weighted representation that integrates temporal information. By extracting multi-scale temporal features, different time windows are used to capture short-term change patterns and long-term evolution trends, and positive and negative sample pairs are constructed for comparative loss calculation.

[0013] In a preferred embodiment, the step of using standardized indicators to achieve data collection from both home and school ends includes: Teachers collect data by selecting students whose psychological state needs to be recorded through an open platform, selecting relevant tags for four dimensions—emotion, behavior, cognition, and environment—based on daily observations, scoring each dimension indicator according to a 5-level scoring standard, and using natural language processing technology to parse the supplementary text input by teachers. Parent-side data collection uses contextualized question templates to select and describe observed psychological states, transforming professional terminology into specific questions in everyday life scenarios. It simplifies the scoring process for each dimension of indicators and uses visual elements such as emoticons and color coding to help parents score quickly and accurately. The observation data from teachers and parents were organized into a vector format.

[0014] In a preferred embodiment, the output personalized intervention strategy and collaborative task allocation scheme include: Through the lifecycle, define task states including creation, allocation, execution, pending confirmation, and completion, and formulate state transition rules and triggering conditions; Establish a cross-platform synchronization mechanism to ensure real-time information synchronization and coordination between different platforms; establish a message queue to achieve a unified message format and data transmission; and achieve real-time updates of task status, execution progress, and resource status through multiple synchronization mechanisms. Establish a conflict resolution mechanism and adopt real-time monitoring and multi-strategy solutions for conflicts; The effectiveness and quality of collaborative work are evaluated through a systematic monitoring mechanism, which tracks task completion, time progress, quality progress and resource consumption in real time, assesses the efficiency of home-school collaboration from multiple dimensions, and establishes a problem identification and early warning mechanism.

[0015] In a preferred embodiment, a home-school mental health collaboration system is used to execute the above-described home-school mental health collaboration method, including: The multi-dimensional data model building module is used to build multi-dimensional data models based on psychological theories, use standardized indicators to collect data from both home and school, and output a structured dataset of psychological states. The multidimensional mental state graph construction module is used to construct a multidimensional mental state graph structure based on a structured mental state dataset. It adopts a time-series dynamic coding method to achieve deep integration of home and school data and outputs a comprehensive mental state feature vector. The temporal modeling and risk assessment module is used to output dynamic mental health risk assessment results and intelligent early warning signals based on comprehensive psychological state feature vectors, using time series modeling and multi-level risk assessment methods. The cross-platform collaborative intervention module is used to output personalized intervention strategies and collaborative task allocation schemes based on intelligent early warning signals and cross-platform collaborative intervention management methods. The multi-dimensional assessment and optimization module is used to output intervention effect assessment reports and strategy optimization suggestions based on personalized intervention strategies, employing multi-dimensional effect assessment and reinforcement learning optimization methods.

[0016] The beneficial effects of this invention are as follows: This study applies dynamic graph neural network technology to the fusion of home-school mental health data. By constructing a multidimensional mental state graph structure and temporal dynamic coding, it achieves deep fusion and spatiotemporal feature extraction of heterogeneous mental health data. An intelligent mental health risk early warning mechanism is established using a long short-term memory network combined with a multi-level risk assessment algorithm. Deep reinforcement learning technology is introduced to optimize intervention strategies. Through Markov decision process modeling and policy gradient methods, adaptive optimization of personalized intervention strategies is achieved, improving the pertinence and effectiveness of interventions. This system solves key technical challenges in home-school collaboration on mental health, achieving standardized collection and intelligent integration of mental health data from families and schools, breaking down traditional information silos between home and school. Through cross-platform collaborative intervention management, it establishes a unified task allocation and execution mechanism, improving the efficiency of home-school collaboration. The system can create personalized mental health profiles for each student, realizing intelligent management of the entire process from problem discovery to intervention completion, providing scientific and efficient technical support for education management departments and mental health service institutions, and has broad application prospects and social value. Attached Figure Description

[0017] Figure 1 This is a flowchart of a home-school mental health collaboration method according to the present invention; Figure 2 This is a module diagram of a home-school psychological health collaboration system according to the present invention. Detailed Implementation

[0018] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, some features described in the examples may be combined in other examples.

[0019] At least one embodiment of the present invention discloses a method for home-school mental health collaboration, such as... Figure 1 As shown, it includes the following steps: S1, based on psychological theory, constructs a multi-dimensional data model, uses standardized indicators to collect data from both home and school, and outputs a structured dataset of psychological states; S11 employs a multi-dimensional psychological state description framework, decomposing students' psychological states into four core dimensions and establishing a standardized indicator system and scoring mechanism for each dimension. Specifically, the emotional dimension includes emotional indicators such as anxiety, depression, anger, fear, and excitement, used to assess changes in students' emotional states; the behavioral dimension includes behavioral indicators such as withdrawal, aggression, inattention, social avoidance, and impulsivity, used to observe students' external behavioral performance; the cognitive dimension includes cognitive indicators such as learning ability, memory, attention, and logical thinking, used to assess students' cognitive function; and the environmental dimension includes environmental factor indicators such as family environment, school environment, and peer relationships, used to analyze the impact of the external environment on students' psychological states. The process involves standardizing the dimensional indicators; selecting key indicators for each dimension based on psychological theory and clinical practice, with selection criteria including representativeness, observability, quantifiability, and relevance to mental health; organizing the selected indicators into corresponding dimensional indicator sets to form a structured indicator system; establishing a unified 5-level scoring system to ensure consistency and comparability of the assessment, where 1 point represents very low level, 2 points represent low level, 3 points represent moderate level, 4 points represent high level, and 5 points represent very high level; and forming corresponding dimensional state vectors by arranging the scores of each indicator in sequence to achieve a numerical representation of mental state. The output includes a standardized mental state data model and a unified scoring standard.

[0020] S12, based on an open platform, acquires psychological state data from teachers and outputs a standardized teacher observation dataset; Data collection for teachers: Teachers select students whose psychological states need to be recorded on a platform (such as DingTalk). The system provides a student list and search function for quick location. Teachers select relevant tags based on daily observations across four dimensions: emotion, behavior, cognition, and environment. The system provides a categorized tag interface that supports multiple selections and quick filtering. Each dimension is scored according to a 5-level scoring standard. The system provides a visual scoring interface and scoring instructions to ensure the accuracy and consistency of the scoring. Teachers can input detailed observation descriptions and supplementary explanations. The system automatically records the precise timestamps of data entry, including the start time, completion time, and last modification time, ensuring the timeliness and traceability of the data. The system automatically maps the labels selected by teachers to the standardized dimension indicators defined in S11; it checks the completeness and validity of the scoring data to ensure that all required fields are completed and meet the requirements of the Level 5 scoring criteria; and it uses natural language processing technology to parse the supplementary text input by teachers, extract key information, and automatically supplement the label data. The processed data is organized into a standardized vector format. The emotion dimension vector contains the score data and weight information of each emotion indicator; the behavior dimension vector contains the score data and observation frequency of each behavior indicator; the cognition dimension vector contains the score data and trend of each cognitive indicator; and the environment dimension vector contains the score data and degree of influence of each environmental indicator. Metadata organization includes auxiliary information such as timestamps and data source identifiers (teacher ID, school ID, etc.). S13, acquire psychological state data from parents and output a parent observation dataset in the same format as the teacher's. Data collection for parents: Parents first confirm the student's identity information to be recorded. The system provides a student information confirmation interface, supporting student selection and management for families with multiple children. Parents select and describe observed psychological states through contextualized question templates. These templates transform professional terminology into specific questions from everyday life scenarios, such as how the child's mood has been at home recently, or whether the child's interaction with the family is normal. Parents perform simplified scoring operations for various indicators. The system provides an intuitive scoring interface, using visual elements such as emoticons and color indicators to help parents score quickly and accurately. Parents can input detailed observation descriptions and family environment information, including specific behavioral performance, emotional changes, and learning status. The system automatically records the precise timestamp of data entry, including entry time, geographical location, and device information, ensuring data integrity and traceability. Ensure data quality and format consistency; convert parents' natural language descriptions into structured data format, use natural language processing technology to identify key information and classify and label it; automatically map the information entered by parents to the standardized dimensional indicators defined in S11; The processed data is organized into a vector format identical to that used by teachers. The emotion dimension vector contains score data for each emotion indicator and specific markers related to the family environment. The behavior dimension vector contains score data for each behavior indicator and specific information from family observations. The cognitive dimension vector contains score data for each cognitive indicator and influencing factors of the family learning environment. The environmental dimension vector contains score data for each environmental indicator and a detailed description of the family environment. Metadata processing includes timestamps, data source identifiers (parent ID, family ID, etc.), quality assessment scores, and other auxiliary data. The output includes a standardized parent observation dataset.

[0021] The final output is a structured dataset of psychological states, including a standardized observation dataset for teachers, containing four-dimensional vectors of emotion, behavior, cognition, and environment, as well as metadata, and a standardized observation dataset for parents, containing four-dimensional vectors of emotion, behavior, cognition, and environment, as well as metadata.

[0022] S2, based on the structured dataset of psychological states, constructs a multidimensional psychological state graph structure, and uses a time-series dynamic coding method to achieve deep integration of home and school data, outputting a comprehensive psychological state feature vector; S21, based on standardized psychological state data, uses a heterogeneous graph construction method to encode multidimensional psychological states into graph structure representations, and outputs standardized graph data containing node features and edge relationships; Based on the standardized psychological state data output by S1, a multi-layered heterogeneous graph structure is constructed, encoding students, teachers, parents, and four-dimensional psychological state indicators as nodes and edge relationships in the graph. Student central node construction: Create a central node for each student, encoding basic information such as student ID, grade, and class, as well as behavioral pattern features extracted from historical observation data, to form the initial representation of the student node; Observer node construction: Observer nodes are created for each teacher and parent. Teacher nodes represent the perspective of observing the school environment, and parent nodes represent the perspective of observing the home environment. Observer feature encoding includes information such as identity information, observation ability assessment, observation frequency, and observation quality. The project constructs psychological dimension nodes; it creates dedicated nodes for the four psychological dimensions of emotion, behavior, cognition, and environment; the emotion dimension node includes information such as emotion intensity, stability, and trend of change; the behavior dimension node includes indicators such as behavior frequency, intensity, and consistency; the cognition dimension node includes cognitive function indicators such as attention level, learning ability, and mental activity; and the environment dimension node includes assessments of family environment quality, school environment adaptability, and social environment status.

[0023] Observational relationship construction: Directed observational edges are established between teachers and students, and between parents and students. The edge weights are calculated based on observation frequency and data quality, reflecting the closeness of the observer's attention to the student and the credibility of the observation results.

[0024] Psychological state association edges are constructed; association edges are established between students and four psychological dimensions, as well as correlation edges between dimensions based on psychological theories. The Pearson correlation coefficient is used to quantify the strength of the associations between dimensions for weight calculation.

[0025] A consistency assessment edge is constructed; an observation consistency connection is established between teachers and parents, and the weight is calculated based on the degree of consistency between home and school observation results to assess the reliability of the observation data.

[0026] Graph structure standardization processing; node feature standardization, using Min-Max normalization to map node feature values ​​to the [0,1] interval to ensure the comparability of features of different types of nodes; using the mean fill strategy for missing features; edge weight standardization, normalizing all edge weights to the [0,1] interval, pruning edges with too small weights (threshold 0.1), generating standardized adjacency matrix and weight matrix; Output a standardized heterogeneous graph structure, which is a complete graph structure including the set of nodes, the set of edges, the adjacency matrix, and the weight matrix; S22, based on a standardized heterogeneous graph structure, uses a dynamic graph neural network method for temporal encoding and outputs a dynamic graph representation containing temporal evolution information; Based on the constructed heterogeneous graph structure, a dynamic graph neural network method is used to encode the graph structure evolution in the time series and capture the dynamic change pattern of psychological state.

[0027] Based on the timestamp information output by S1, the observed data is divided into time windows, and a corresponding graph snapshot is constructed for each time window to form a time series graph sequence. The node status is updated according to the observed data at each time point, and the edge weights are dynamically adjusted based on data quality and observation frequency. Through standardized operations such as graph structure alignment, feature dimension unification, and missing data processing, the graph structure at different time points is ensured to be consistent and comparable. Dynamic graph neural network encoding is employed. Each graph in the temporal graph sequence undergoes graph convolution operations, capturing local and global structural information through a multi-layer graph convolutional network to obtain node embedding representations. A recurrent neural network is used to capture the temporal evolution patterns of the graph sequence, combined with an attention mechanism to identify important time points, generating a weighted representation that integrates temporal information. Dynamic graph feature fusion is implemented through multi-scale temporal feature extraction, using different time windows to capture short-term change patterns and long-term evolution trends. Positive and negative sample pairs are constructed for comparative loss calculation, improving the discriminative and generalization capabilities of the features. The output is a comprehensive psychological state feature vector containing spatiotemporal information and multimodal fusion, including a deep feature representation of home-school integration.

[0028] S3, based on a comprehensive psychological state feature vector, uses time series modeling and multi-level risk assessment methods to output dynamic mental health risk assessment results and intelligent early warning signals; S31, based on the comprehensive psychological state feature vector, uses a long short-term memory network for time series modeling and outputs the prediction results of psychological state change trends; Based on the comprehensive psychological state feature vector output by S2, a time series modeling method is used to analyze the dynamic changes in mental health status and predict future trends in mental health status.

[0029] Time series data organization: Historical psychological state characteristic data are arranged in chronological order to form continuous time series data. Missing data are interpolated to ensure the integrity and continuity of the time series.

[0030] Feature weighting: Based on the importance and stability of different psychological dimensions, features are dynamically weighted to improve the model's sensitivity to changes in key features.

[0031] Deep learning modeling: A time-series prediction model is constructed using a long short-term memory network. Through a multi-layer network structure, it simultaneously captures short-term fluctuations and long-term trends in mental states, thereby achieving accurate prediction of future mental health status.

[0032] Trend analysis and anomaly detection: Analyze the changing trends and abnormal patterns in the prediction results, identify the development direction, magnitude of change and periodic patterns of mental health status, and detect abnormal changes by comparing with an individual's historical baseline.

[0033] Outputs trend prediction results for mental health status, including predicted mental health status data for future time points, trend analysis reports, and abnormal change detection results.

[0034] S32, based on the trend prediction results of mental health status, adopts a multi-level risk assessment algorithm to output immediate, short-term and long-term mental health risk assessment results; Based on the trend prediction results of mental health status output by S31, a multi-level risk assessment algorithm is adopted to comprehensively assess mental health risks from three time dimensions: immediate, short-term and long-term.

[0035] Real-time risk assessment: Analyze the degree of deviation between the current psychological state and the individual's historical baseline, assess the consistency of home-school observation results, integrate the baseline deviation risk and the observation consistency risk, generate real-time risk assessment results and determine the risk level.

[0036] Short-term risk assessment: Based on the forecast of changes in mental health status over the next few weeks, identify abnormal patterns and potential risk points in the forecast trends, analyze the fluctuation characteristics and stability of mental health status in the short term, and determine the short-term risk level.

[0037] Long-term risk assessment: Collect historical risk assessment data, use time decay weighting to highlight the importance of recent risks, calculate cumulative long-term risk indicators, identify long-term trend patterns of changes in mental health status, integrate historical risk accumulation and long-term trends, and determine the long-term risk level.

[0038] Adaptive threshold setting: Individualized thresholds are set based on the distribution characteristics of an individual's historical risk score, and the thresholds are dynamically adjusted based on the changing patterns of an individual's mental health status and external factors. The final risk level is determined by integrating the risk assessment results from the three levels.

[0039] Outputs real-time mental health risk assessment results, including the current mental health risk level and probability; short-term mental health risk assessment results, including the future short-term mental health risk development trend and risk level; long-term mental health risk assessment results, including long-term mental health risk development patterns and risk level predictions; multi-level comprehensive risk reports, including comprehensive risk assessment reports integrating three time levels; and individualized risk threshold parameters, including individualized risk judgment thresholds set for individual characteristics. S33, based on the results of multi-level risk assessment, adopts an adaptive threshold setting algorithm to output intelligent early warning signals and early warning levels; Based on the multi-level risk assessment results output by S32, an adaptive threshold setting algorithm is adopted to dynamically adjust the early warning threshold and generate intelligent early warning signals.

[0040] Adaptive threshold setting: Analyze the distribution characteristics of an individual's historical risk score, use the quantile method to set an individualized early warning threshold, dynamically adjust the threshold according to factors such as individual development stage, life events, and environmental changes, and verify the accuracy of the threshold through historical data.

[0041] Multi-level threshold integration: Establish corresponding early warning threshold systems for immediate, short-term, and long-term risk assessments, assign weight coefficients to risk thresholds at different time levels, and form a unified early warning threshold system.

[0042] Intelligent early warning signal generation: The results of risk assessment at each level are compared with the corresponding thresholds to evaluate indicators such as the degree of exceeding the threshold, duration, and frequency. The early warning level is determined based on the degree to which the risk exceeds the threshold, and an early warning signal is generated that includes the risk type, urgency level, trigger time, risk description, and recommended measures.

[0043] Prioritize early warning signals; assess the urgency, scope of impact, and resource requirements of different risk signals, and prioritize the early warning signals accordingly.

[0044] Early warning level classification and response strategy generation; establish a four-level early warning level system, formulate judgment criteria and response time requirements for each level, identify the specific risk types that trigger early warnings, and match corresponding response strategies according to risk types and early warning levels.

[0045] Output intelligent early warning signals, including warning signals with risk type, urgency level and trigger time; early warning level classification results, based on the classification and judgment results of the four-level early warning system; early warning response strategy suggestions, specific response strategies and handling suggestions for different early warning levels; early warning priority ranking, priority ranking and handling order of multiple early warning signals.

[0046] S4, based on intelligent early warning signals, adopts a cross-platform collaborative intervention management method to output personalized intervention strategies and collaborative task allocation schemes; Based on the intelligent early warning signals output by S3, a cross-platform collaborative intervention management method is adopted to generate personalized intervention strategies and realize the intelligent allocation of collaborative tasks.

[0047] S41, Personalized intervention strategy generation; strategy recommendation based on case-based reasoning; extracting key features of historical successful cases from an intervention case knowledge base, including student characteristics, risk characteristics, environmental characteristics, and intervention characteristics. Calculating the similarity match between the current student's situation and historical cases using a multi-dimensional similarity calculation algorithm, considering the importance weights of different features. Relevant cases are retrieved and ranked based on the similarity calculation results, selecting the case with the highest similarity as the basis for strategy recommendation, and making adaptive adjustments according to the current student's specific situation.

[0048] S42, Multi-dimensional Strategy Optimization: Based on the risk type in the warning signals, match corresponding intervention strategy categories, and develop specialized intervention methods for different risk types such as emotional, behavioral, cognitive, social, and academic risks. Adjust the intensity and frequency of intervention strategies according to the warning level, establishing a four-level warning system corresponding to intervention intensity. Optimize strategy design by combining students' personality traits, learning abilities, and family environment, and assess the availability and feasibility of resources required for strategy implementation.

[0049] S43, Intelligent Decomposition of Collaborative Tasks; Application of Task Decomposition Algorithms; Decomposing complex intervention strategies into executable task units, identifying different types of intervention activities through strategy content analysis, refining them into independent executable tasks, and clarifying task boundaries and responsibilities. Assessing the execution difficulty of each task unit, analyzing the dependencies between tasks, and determining the task execution order and time estimation through dependency graph construction and critical path analysis.

[0050] Task attribute definition: Define detailed attribute information for each task unit, assign the most suitable executor to the task based on professional matching principles, and clarify task objectives, operational steps, and standards. Set completion deadlines based on task urgency, complexity, and dependencies, and develop an evaluation system that includes completion standards, quality standards, and acceptance criteria.

[0051] S44 enables cross-platform collaborative management; it applies optimal task allocation algorithms; and achieves optimal matching between tasks and execution entities through intelligent algorithms. Each execution entity is evaluated based on dimensions such as professional competence, available time, workload, and collaboration ability to create a competence profile. A matching score is calculated based on factors such as skill matching, time matching, and workload matching. An optimization algorithm is employed to maximize matching, balance workload, and minimize completion time, achieving globally optimal allocation under multiple constraints.

[0052] A cross-platform synchronization mechanism is established to ensure real-time information synchronization and coordination across different platforms. A message queue is established to achieve a unified message format and data transmission, and multiple synchronization mechanisms are used to achieve real-time updates of task status, execution progress, and resource status. Multiple notification mechanisms are designed, and various notification methods are used for customized notifications. A conflict resolution mechanism is established, employing real-time monitoring and multi-strategy solutions for various conflicts.

[0053] S45, Task Lifecycle Management; Application of the State Machine Pattern; Managing the entire lifecycle of a task through the state machine pattern, defining task states including creation, assignment, execution, pending confirmation, and completion, and formulating state transition rules and triggering conditions. Establishing a state monitoring mechanism to monitor state changes and transitions in real time. Establishing an exception handling process, employing multiple handling strategies and recovery mechanisms for various exceptions.

[0054] Collaboration effectiveness monitoring; evaluating the effectiveness and quality of collaborative work through a systematic monitoring mechanism. Real-time tracking of task completion, time progress, quality progress, and resource consumption to assess the efficiency of home-school collaboration from multiple dimensions. Establishing a problem identification and early warning mechanism, employing various methods to identify different types of problems and provide early warning information. Generating optimization suggestions, including process optimization and resource optimization.

[0055] It outputs personalized intervention strategy solutions, comprehensive intervention strategies customized for specific student situations; collaborative task allocation lists, detailed task breakdowns and responsibility assignment lists; cross-platform synchronous configurations, technical configuration solutions supporting multi-platform collaborative work; task execution schedules, task execution plans including key time nodes; and collaborative workflow diagrams, visual collaborative workflows and responsibility division diagrams.

[0056] S5, based on personalized intervention strategies, employs multi-dimensional effect evaluation and reinforcement learning optimization methods to output intervention effect evaluation reports and strategy optimization suggestions; Based on the intervention strategy execution results output by S4, a multi-dimensional effect evaluation and reinforcement learning optimization method are used to comprehensively evaluate the intervention effect and continuously optimize the strategy.

[0057] S51, Multi-dimensional Intervention Effectiveness Evaluation; Direct Effect Evaluation; This involves evaluating the direct effects and degree of improvement of the intervention through comparative analysis. It analyzes students' mental health status assessment data before and after the intervention, including indicators such as emotional state, cognitive ability, behavioral performance, and social skills. The degree of improvement in each dimension of mental health status is quantified, and the significance of the improvement is verified using statistical methods. Based on the degree of improvement, the intervention effect is categorized into four levels: significant, moderate, slight, and ineffective.

[0058] Stability assessment: This involves analyzing the stability and persistence of the intervention's effects. Data on changes in mental health status were collected during a 3-6 month observation period following the intervention, and the volatility and stability of the data were analyzed. Stability was assessed using the coefficient of variation (COP), calculated as the ratio of the standard deviation to the mean multiplied by 100%. The stability level of the intervention's effect was determined based on the magnitude of the COP.

[0059] Durational effectiveness assessment involves long-term follow-up analysis to determine the duration and decay patterns of the intervention's effects. Mental health status data is analyzed 6-12 months or longer after the intervention to calculate the effective duration, half-life, and complete decay time of the intervention's effects. The types and patterns of decay of the intervention's effects are identified, and the persistence level of the intervention's effectiveness is assessed based on the duration and decay pattern.

[0060] Evaluation of the effectiveness of collaboration: This involves analyzing the role and effect of home-school collaboration in the intervention. It analyzes changes in the consistency of mental health awareness between home and school before and after the intervention, assessing the frequency, quality, and level of cooperation of both parties. It also evaluates the improvement in communication between home and school, calculating the incremental effect and additional benefits of home-school collaboration compared to unilateral intervention.

[0061] S52, Policy Optimization Based on Reinforcement Learning; Markov Decision Process Modeling; This section describes the intervention decision problem by establishing a mathematical model. The student's mental health characteristics, family environment, and school environment are defined as the state space, with states represented in vector form, each dimension corresponding to a state feature. The available intervention strategies, intervention intensity, and intervention frequency are constructed as the action space. A reward function is designed based on the intervention effect evaluation results, constructed by weighting and combining effect rewards, efficiency rewards, safety rewards, and satisfaction rewards. The state transition probability is estimated based on historical data, calculated by the ratio of the number of times a specific action is performed from a specific state to the target state to the total number of times the action is performed in that state.

[0062] Application of reinforcement learning algorithms: Optimal intervention strategies are learned through specific algorithmic implementations. An experience replay mechanism is established to store and reuse historical intervention experience data. A state-action value function is learned using deep reinforcement learning algorithms. The loss function is defined by the expected value of the squared difference between the maximum value of the next state (after discounting the current reward) and the current state action value. The policy gradient method is used to optimize the probability distribution of intervention strategy selection. The policy gradient is calculated by taking the expectation of the product of the logarithmic gradient of the policy function and the state-action value function.

[0063] Strategy optimization and iteration: The effectiveness of intervention strategies is improved through continuous evaluation and refinement. The performance of the current intervention strategy under different conditions is assessed, identifying directions for improvement and key areas for optimization. Key parameters of the strategy recommendation algorithm are adjusted to enhance performance, and the effectiveness of the optimized strategy is verified through simulations and real-world case studies.

[0064] S53, Personalized Portfolio Generation and Knowledge Accumulation: The personalized mental health portfolio construction process establishes a complete personal portfolio through systematic data collection and organization; the complete intervention process records the entire process and key nodes from problem discovery to intervention completion, including key information from stages such as problem discovery, assessment and diagnosis, strategy formulation, implementation, and effect evaluation. Multi-dimensional data integration: Data from psychological assessment, intervention strategies, implementation process, and effect evaluation are integrated using methods such as data standardization, data cleaning, and data correlation. Timeline portfolio generation: Structured personal mental health portfolios are generated according to a timeline, including key information such as time nodes, status changes, and intervention measures. Portfolio quality control: Ensures the completeness, accuracy, and consistency of portfolio data.

[0065] The knowledge pattern extraction and accumulation process extracts and accumulates valuable knowledge patterns by analyzing historical cases. Successful case feature extraction extracts key features and success factors from successful intervention cases, using methods such as statistical analysis, association analysis, and cluster analysis to identify case characteristics, strategy characteristics, process characteristics, and effect characteristics. Failed case lesson summary analyzes the causes and lessons learned from failed cases to formulate avoidance strategies. General pattern recognition identifies general intervention patterns and rules across cases using methods such as pattern mining, machine learning, and statistical modeling. Knowledge base updates and maintenance add new knowledge patterns to the case knowledge base and perform regular maintenance.

[0066] The case similarity matching optimization process improves the accuracy and efficiency of case recommendations by refining the matching algorithm. Dynamic feature weight adjustment adjusts the weights of case features based on feedback from the intervention, employing machine learning and statistical analysis to achieve adaptive weight optimization. The similarity calculation algorithm is optimized to improve matching accuracy, including distance metric optimization, similarity function improvement, and feature processing optimization. Case retrieval efficiency is improved by refining the case retrieval algorithm to enhance both retrieval efficiency and accuracy. A recommendation quality assessment mechanism and feedback loop are established to evaluate the quality of case recommendations.

[0067] S54, Feedback Mechanism and Continuous Improvement: A multi-faceted feedback collection process is used to obtain comprehensive improvement information through systematic feedback collection; Student Subjective Feelings: Collecting students' subjective feelings and evaluations of the intervention process and its effects, using methods such as questionnaires and interviews; Parent Satisfaction Survey: Surveying parents' satisfaction with the intervention strategy and collaborative process, using methods such as satisfaction scales and telephone surveys; Teacher Professional Evaluation: Collecting teachers' professional evaluation opinions on the intervention's effectiveness and operability, using methods such as professional scales and expert meetings; Mental Health Expert Review: Inviting mental health experts to conduct a professional review of the intervention plan and its effects, using methods such as expert review and peer review.

[0068] The system performance optimization process improves the overall system performance through systematic performance analysis and optimization. Algorithm performance analysis identifies the performance and optimization potential of each core algorithm, using methods such as performance testing and stress testing to analyze metrics such as algorithm efficiency, accuracy, and stability. System response speed optimization enhances user experience by improving system response speed, including query optimization, caching optimization, and algorithm optimization. Prediction accuracy improvement is achieved through algorithm optimization and data quality enhancement, employing advanced methods such as ensemble learning, deep learning, and transfer learning for model optimization, feature engineering, and parameter tuning. Resource utilization efficiency optimization utilizes computing and storage resources through techniques such as resource scheduling and load balancing.

[0069] Outputs include a multi-dimensional intervention effectiveness evaluation report, encompassing a comprehensive assessment of direct effects, stability, sustainability, and synergy; personalized mental health profiles, providing a structured, complete personal mental health record and developmental trajectory; strategy optimization suggestions, including reinforcement learning-based suggestions and improvement plans for intervention strategies; knowledge model updates, extracting new knowledge models and updating the case knowledge base; system performance optimization plans, providing performance optimization suggestions and implementation schemes for each module of the system; and a continuous improvement plan, a system continuous improvement plan and timeline developed based on feedback results.

[0070] In one embodiment of the present invention, a home-school mental health collaboration system is provided, such as... Figure 2 As shown, it includes: The multi-dimensional data model building module is used to build multi-dimensional data models based on psychological theories, use standardized indicators to collect data from both home and school, and output a structured dataset of psychological states. The multidimensional mental state graph construction module is used to construct a multidimensional mental state graph structure based on a structured mental state dataset. It adopts a time-series dynamic coding method to achieve deep integration of home and school data and outputs a comprehensive mental state feature vector. The temporal modeling and risk assessment module is used to output dynamic mental health risk assessment results and intelligent early warning signals based on comprehensive psychological state feature vectors, using time series modeling and multi-level risk assessment methods. The cross-platform collaborative intervention module is used to output personalized intervention strategies and collaborative task allocation schemes based on intelligent early warning signals and cross-platform collaborative intervention management methods. The multi-dimensional assessment and optimization module is used to output intervention effect assessment reports and strategy optimization suggestions based on personalized intervention strategies, employing multi-dimensional effect assessment and reinforcement learning optimization methods.

[0071] The embodiments of the present invention have been described above. However, the embodiments are not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make more equivalent embodiments under the guidance of the present embodiments, and all of them are within the protection scope of the present embodiments.

Claims

1. A method for collaborative mental health care between home and school, characterized in that, Includes the following steps: S1, based on psychological theory, constructs a multi-dimensional data model, uses standardized indicators to collect data from both home and school, and outputs a structured dataset of psychological states; S2, based on the structured dataset of psychological states, constructs a multidimensional psychological state graph structure, and uses a time-series dynamic coding method to achieve deep integration of home and school data, outputting a comprehensive psychological state feature vector; S3, based on a comprehensive psychological state feature vector, uses time series modeling and multi-level risk assessment methods to output dynamic mental health risk assessment results and intelligent early warning signals; S4, based on intelligent early warning signals, adopts a cross-platform collaborative intervention management method to output personalized intervention strategies and collaborative task allocation schemes; S5, based on personalized intervention strategies, employs multi-dimensional effect evaluation and reinforcement learning optimization methods to output intervention effect evaluation reports and strategy optimization suggestions.

2. The method for home-school collaboration in mental health according to claim 1, characterized in that, The multi-dimensional data model constructed based on psychological theory includes: A multi-dimensional psychological state description framework is adopted, which decomposes students' psychological state into four core dimensions: emotional, behavioral, cognitive, and environmental dimensions; and establishes a standardized indicator system and scoring mechanism for each dimension. Establish a unified 5-level scoring system, and form a dimensional state vector by arranging the scores of each indicator in order.

3. The method for home-school collaboration in mental health according to claim 1, characterized in that, The construction of the multidimensional mental state diagram structure based on the structured dataset of mental states includes: A heterogeneous graph construction method is used to encode students, teachers, parents, and four-dimensional psychological state indicators as nodes and edge relationships in a graph; Create a central node for each student, an observer node for each teacher and parent, and nodes for the four psychological dimensions of emotion, behavior, cognition, and environment. Establish directed observation edges between teachers and students, and between parents and students; establish connection edges between students and the four psychological dimensions; and establish correlation edges between dimensions based on psychological theories. Normalization is used to map node feature values ​​to the 0-1 interval, and all edge weights are normalized to the 0-1 interval to generate a standardized adjacency matrix and weight matrix.

4. The method for home-school collaboration in mental health according to claim 1, characterized in that, The method based on comprehensive psychological state feature vectors, employing time series modeling and multi-level risk assessment, includes: A time-series prediction model is constructed using a long short-term memory network, which simultaneously captures short-term fluctuations and long-term trends in mental states through a multi-layered network structure. A multi-level risk assessment algorithm is used to comprehensively assess mental health risks from three time dimensions: immediate, short-term, and long-term. Immediate risk assessment analyzes the degree of deviation between the current mental state and the individual's historical baseline; short-term risk assessment is based on the prediction of the trend of changes in mental health status in the next few weeks; and long-term risk assessment collects historical risk assessment data and uses time decay weights. An adaptive threshold setting algorithm is adopted to set individualized early warning thresholds based on the distribution characteristics of individual historical risk scores, and generate intelligent early warning signals that include risk type, urgency level and trigger time.

5. The method for home-school collaboration in mental health according to claim 1, characterized in that, The cross-platform collaborative intervention and management method based on intelligent early warning signals includes: Key features of historical successful cases are extracted from the intervention case knowledge base. The similarity between the current student situation and historical cases is calculated, and the case with the highest similarity is selected as the basis for strategy recommendation. Complex intervention strategies are broken down into actionable task units, and different types of intervention activities are identified through strategy content analysis. Intelligent algorithms are used to achieve optimal matching between tasks and execution entities, and each execution entity is evaluated from the dimensions of professional capabilities, available time, workload, and collaboration capabilities.

6. The method for home-school collaboration in mental health according to claim 1, characterized in that, The personalized intervention strategy, employing multi-dimensional effect evaluation and reinforcement learning optimization methods, includes: By comparing and analyzing the direct effects of the intervention, we analyzed the mental health status assessment data of students before and after the intervention, and quantified the degree of improvement in each dimension of mental health status. By using Markov decision process modeling, the characteristics of students' mental health status, family environment, and school environment are defined as the state space, and the intervention strategies, intervention intensity, and intervention frequency are constructed as the action space. A reward function is designed based on the intervention effect evaluation results. The state-action value function is learned through deep reinforcement learning algorithms, and the probability distribution of intervention strategy selection is optimized by using the policy gradient method. Build personalized mental health profiles to record the entire process from problem discovery to intervention completion, and integrate multi-dimensional data from psychological assessment, intervention strategies, implementation process, and effect evaluation.

7. The method for home-school collaboration in mental health according to claim 1, characterized in that, The method of using time-series dynamic coding to achieve deep integration of home-school data includes: A dynamic graph neural network is used to encode the graph structure evolution in the time series. The observed data is divided into time windows according to the timestamp information, and a corresponding graph snapshot is constructed for each time window to form a time series graph sequence. Graph convolution operations are performed on each graph in the temporal graph sequence. Local and global structural information is captured through a multi-layer graph convolutional network to obtain node embedding representations. A recurrent neural network is used to capture the temporal evolution pattern of graph sequences, and an attention mechanism is combined to identify important time points to generate a weighted representation that integrates temporal information. By extracting multi-scale temporal features, different time windows are used to capture short-term change patterns and long-term evolution trends, and positive and negative sample pairs are constructed for comparative loss calculation.

8. The method for home-school collaboration in mental health according to claim 1, characterized in that, The adoption of standardized indicators to achieve data collection from both home and school ends includes: Teachers collect data by selecting students whose psychological state needs to be recorded through an open platform, selecting relevant tags for four dimensions—emotion, behavior, cognition, and environment—based on daily observations, scoring each dimension indicator according to a 5-level scoring standard, and using natural language processing technology to parse the supplementary text input by teachers. Parent-side data collection uses contextualized question templates to select and describe observed psychological states, transforming professional terminology into specific questions in everyday life scenarios. It simplifies the scoring process for each dimension of indicators and uses visual elements such as emoticons and color coding to help parents score quickly and accurately. The observation data from teachers and parents were organized into a vector format.

9. A method for home-school collaboration in mental health according to claim 1, characterized in that, The output personalized intervention strategy and collaborative task allocation scheme include: Through the lifecycle, define task states including creation, allocation, execution, pending confirmation, and completion, and formulate state transition rules and triggering conditions; Establish a cross-platform synchronization mechanism to ensure real-time information synchronization and coordination between different platforms; establish a message queue to achieve a unified message format and data transmission; and achieve real-time updates of task status, execution progress, and resource status through multiple synchronization mechanisms. Establish a conflict resolution mechanism and adopt real-time monitoring and multi-strategy solutions for conflicts; The effectiveness and quality of collaborative work are evaluated through a systematic monitoring mechanism, which tracks task completion, time progress, quality progress and resource consumption in real time, assesses the efficiency of home-school collaboration from multiple dimensions, and establishes a problem identification and early warning mechanism.

10. A home-school mental health collaboration system, used to execute a home-school mental health collaboration method as described in any one of claims 1-9, characterized in that, include: The multi-dimensional data model building module is used to build multi-dimensional data models based on psychological theories, use standardized indicators to collect data from both home and school, and output a structured dataset of psychological states. The multidimensional mental state graph construction module is used to construct a multidimensional mental state graph structure based on a structured mental state dataset. It adopts a time-series dynamic coding method to achieve deep integration of home and school data and outputs a comprehensive mental state feature vector. The temporal modeling and risk assessment module is used to output dynamic mental health risk assessment results and intelligent early warning signals based on comprehensive psychological state feature vectors, using time series modeling and multi-level risk assessment methods. The cross-platform collaborative intervention module is used to output personalized intervention strategies and collaborative task allocation schemes based on intelligent early warning signals and cross-platform collaborative intervention management methods. The multi-dimensional assessment and optimization module is used to output intervention effect assessment reports and strategy optimization suggestions based on personalized intervention strategies, employing multi-dimensional effect assessment and reinforcement learning optimization methods.

Images