Student mental health detection method and device based on multi-modal behavior fusion and dynamic weight adjustment

Abstract

The invention provides a student mental health detection method and device based on multi-modal behavior fusion and dynamic weight adjustment, and relates to the field of mental health detection, and the method comprises the steps: collecting student behavior data in a non-intrusive manner, and carrying out the data fusion; dynamically adjusting weights of physiological signal data, behavior mode data and digital trace data in the student behavior data based on a weight distribution network; constructing a time sequence behavior graph based on the student behavior data after weight distribution; based on the time sequence behavior graph, identifying time sequence behaviors related to the psychological problem through a message passing mechanism by utilizing a graph neural network; and performing psychological state grading based on the time sequence behaviors by utilizing an evaluation model considering the behavior deviation degree and the time sequence mutability, and generating corresponding intervention suggestions according to the graded psychological states. Through multi-modal data fusion, dynamic weight adjustment, time sequence diagram analysis and other innovations, the limitation of single instrument training in the prior art is broken through.

Description

Technical Field

[0001] This application relates to the field of mental health testing technology, and more specifically, to a method and device for testing student mental health based on multimodal behavior fusion and dynamic weight adjustment. Background Technology

[0002] Current methods for assessing student mental health generally suffer from the limitations of "single-device training," making it difficult to achieve comprehensive, dynamic, and personalized evaluations. Traditional questionnaire methods are akin to requiring students to complete "psychological weightlifting" through self-reporting. Relying on subjectively completed scales (such as PHQ-9 and GAD-7), they are not only susceptible to social desirability bias—students may conceal anxiety or depression symptoms due to stigma, leading to data distortion—but also fail to capture dynamic fluctuations in mental states due to a lack of real-time data, such as the accumulated stress before exams or stress responses to sudden events. Meanwhile, physiological signal monitoring methods are like forcing physical fitness tests using professional treadmills. While collecting data such as heart rate variability (HRV) and skin conductance through wearable devices provides objective indicators, they are highly dependent on the devices, have a poor user experience, and students may refuse to use them due to frequent charging, discomfort, or privacy concerns. Furthermore, the correlation between physiological signals and mental states is significantly affected by individual differences; for example, the same HRV value may correspond to completely different emotional states in different students, leading to interpretative ambiguity. More importantly, existing methods mostly use static models, such as training with fixed weights, which cannot adapt to the characteristics of students as they change with age and academic stage. For example, the stressors of freshmen and senior high school students are completely different, but traditional models use a uniform threshold for evaluation, resulting in poor adaptability and high false alarm rate. Summary of the Invention

[0003] The embodiments of this application provide a method and apparatus for detecting student mental health based on multimodal behavior fusion and dynamic weight adjustment, in order to solve the technical problems existing in the prior art.

[0004] Other features and advantages of this application will become apparent from the following detailed description, or may be learned in part from practice of this application.

[0005] According to a first aspect of the embodiments of this application, a method for detecting student mental health based on multimodal behavior fusion and dynamic weight adjustment is provided, including: Student behavior data is collected and fused using a non-invasive method. The student behavior data includes: physiological signal data, behavioral pattern data, and digital trace data. Design a lightweight weight allocation network to dynamically adjust the weights of physiological signal data, behavioral pattern data, and digital trace data in student behavior data; Construct a time-series behavior graph based on student behavior data after weight allocation; Based on the aforementioned temporal behavior map, a graph neural network is used to identify temporal behaviors related to psychological problems through a message passing mechanism. An assessment model is constructed that considers behavioral deviation and temporal abruptness. The assessment model is used to classify psychological states based on the temporal behaviors, and corresponding intervention suggestions are generated based on the classified psychological states.

[0006] In some embodiments of this application, based on the foregoing scheme, physiological signal data is collected in a non-invasive manner, including: The system uses a smart bracelet to monitor heart rate variability and skin conductance in real time, and calculates the intensity of the autonomic nervous system response. The time-domain analysis of heart rate variability is as follows: ; in, The RMSSD value represents the interval between consecutive heartbeats, where N is the number of heartbeats. The lower the RMSSD value, the stronger the sympathetic nerve activity. Frequency domain analysis of skin electrical activity is as follows:

[0007] Wherein, EDA represents skin electrical activity, SCL represents skin conductance level, t represents time, and SCR represents skin conductance response.

[0008] In some embodiments of this application, based on the foregoing scheme, behavioral pattern data is collected in a non-invasive manner, including: Behavioral pattern data is captured through classroom cameras and the campus card system, including: classroom participation, frequency of social interaction, and regularity of daily routines; Establish quantitative models for classroom participation, frequency of social interaction, and regularity of daily routines; The classroom participation measurement model is as follows: ; in, and These are the weighting coefficients; The social interaction frequency quantification model is as follows: ; in, It is the ratio of the time spent traveling with the k-th type of companions to the total social time, used to quantify the proportion of time a certain type of companion occupies in the overall social interaction; The quantitative model for regularity of daily routines is as follows:

[0009] in, Let represent the time of falling asleep recorded for the i-th time, where i represents different times of falling asleep, and n represents the total number of times the time of falling asleep has been recorded. Let be the average time to fall asleep over n attempts. Let be the standard deviation of the time to fall asleep n times.

[0010] In some embodiments of this application, based on the foregoing scheme, digital trace data is collected in a non-invasive manner, including: We obtain online learning engagement, homework delay probability, and risk word frequency through campus network logs and learning platform APIs. Online learning engagement is calculated using the focus index formula, which is as follows: ; The probability of job delay is predicted using a logistic regression model; The risk word frequency is obtained through the risk word frequency calculation formula, which is as follows:

[0011] Where W is a preset risk vocabulary, W = {anxiety, insomnia, self-harm}; TF-IDF value is a weighted score of term frequency - inverse document frequency, and I is an indicator function.

[0012] In some embodiments of this application, based on the foregoing scheme, the design of a lightweight weight allocation network, and the dynamic adjustment of the weights of physiological signal data, behavioral pattern data, and digital trace data in student behavior data based on the weight allocation network, includes: Transform individual student characteristics and group behavior into computable vector representations; A lightweight weight allocation network is designed using the TinyML architecture; The vector representation is used as the basis for weight adjustment, and the weights of physiological signal data, behavioral pattern data and digital trace data are dynamically adjusted using the weight allocation network. The dynamic adjustment rules include: increasing the weight of daily routines for senior high school students and suppressing the weight of social activities for introverted students.

[0013] In some embodiments of this application, based on the foregoing scheme, the construction of a time-series behavioral graph based on weighted student behavioral data includes: Based on the weighted physiological signal data, behavioral pattern data, and digital trace data, node features are encoded to form initial node features. Calculate edge weights based on the initial features of the nodes; Edge weights exceeding a set threshold are retained to form a temporal behavior graph.

[0014] In some embodiments of this application, based on the foregoing scheme, the step of identifying temporal behaviors related to psychological problems using a graph neural network through a message passing mechanism based on the temporal behavior graph includes: By using graph neural networks to capture the temporal evolution features of nodes and edges in a temporal behavior graph, abnormal behaviors can be identified, and long-term risk assessment and short-term acute stress detection can be performed. The long-term risk assessment formula is as follows: ; in, To calculate the average social weight of the i-th type over 30 days, Social weight threshold The standard deviation of bedtime Values ​​are taken over 2 hours; The formula for testing short-term acute stress is as follows: ; in, The rate of change of HRV value over 72 hours. To determine the frequency of searches for "anxiety," Keyword = 5 times / day.

[0015] In some embodiments of this application, based on the foregoing scheme, during the process of identifying abnormal behavior, the threshold for anomaly detection is dynamically optimized using group behavior, including: Calculate the median and variance of the group edge weights as a dynamic baseline; The threshold is updated based on the dynamic baseline.

[0016] In some embodiments of this application, based on the foregoing scheme, the step of using the assessment model to classify psychological states based on the temporal behavior and generating corresponding intervention suggestions based on the classified psychological states includes: Establish a baseline of behavioral data for students of the same grade / major as a reference for calculating individual deviations; Individual deviation is calculated based on a baseline of behavioral data, and the psychological state level is determined by combining individual deviation with temporal variability. Differentiated intervention behavior data were matched based on psychological state levels; Feedback and optimization are performed based on behavioral data after intervention.

[0017] According to a second aspect of the embodiments of this application, a student mental health detection device based on multimodal behavior fusion and dynamic weight adjustment is provided, comprising: The acquisition unit is used to acquire student behavior data in a non-invasive manner and perform data fusion. The student behavior data includes: physiological signal data, behavioral pattern data, and digital trace data. A weight allocation unit is used to design a lightweight weight allocation network and dynamically adjust the weights of physiological signal data, behavioral pattern data, and digital trace data in student behavior data based on the weight allocation network. The building unit is used to construct a time-series behavior graph based on student behavior data after weight allocation. The identification unit is used to identify temporal behaviors related to psychological problems based on the temporal behavior map and using a graph neural network through a message passing mechanism. The assessment unit is used to construct an assessment model that considers behavioral deviation and temporal abruptness. The assessment model is used to classify psychological states based on the temporal behaviors, and corresponding intervention suggestions are generated based on the classified psychological states.

[0018] The technical solution of this application, through innovations such as multimodal data fusion, dynamic weight adjustment, and time series graph analysis, breaks through the limitations of the existing technology of "single equipment training" and provides a non-intrusive, real-time, and personalized solution for mental health management in the education field, which has significant technological progress and practical value.

[0019] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0020] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application. It is obvious that the drawings described below are merely some embodiments of this application, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort. In the drawings: Figure 1 A flowchart illustrating a student mental health detection method based on multimodal behavior fusion and dynamic weight adjustment according to an embodiment of this application is shown. Figure 2 A block diagram of a student mental health testing device based on multimodal behavior fusion and dynamic weight adjustment according to an embodiment of this application is shown. Figure 3 A block diagram of an electronic device according to one embodiment of this application is shown; Figure 4 A schematic diagram of the structure of a computer system suitable for implementing the electronic device of the present application is shown. Detailed Implementation

[0021] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided to make this application more comprehensive and complete, and to fully convey the concept of the exemplary embodiments to those skilled in the art.

[0022] Furthermore, the described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. Numerous specific details are provided in the following description to give a thorough understanding of embodiments of this application. However, those skilled in the art will recognize that the technical solutions of this application can be practiced without one or more of the specific details, or other methods, components, apparatuses, steps, etc., can be employed. In other instances, well-known methods, apparatuses, implementations, or operations are not shown or described in detail to avoid obscuring various aspects of this application.

[0023] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0024] The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily have to be performed in the described order. For example, some operations / steps can be broken down, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.

[0025] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0026] The following detailed description of some embodiments of this application will be provided in conjunction with the accompanying drawings. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0027] To address the problems existing in current technologies, this application proposes a student mental health detection method based on multimodal behavior fusion and dynamic weight adjustment. This method first utilizes campus IoT devices to non-invasively collect multi-dimensional behavioral data, including physiological signals (HRV, EDA), behavioral patterns (classroom participation, social frequency, daily routine), and digital traces (online learning duration, search keywords), similar to comprehensively monitoring heart rate, movement accuracy, and dietary records during exercise, avoiding the limitations of a single data source. Second, a dynamic weight allocation model is introduced. Based on individual student characteristics (such as gender, grade, and major) and group baseline data, a lightweight neural network is used to calculate the weights of each dimension in real time on edge devices. For example, a higher weight is assigned to "daily routine" for senior high school students, while a lower weight is assigned to "social interaction" for introverted students. First, it achieves "personalized training programs." Second, it constructs a temporal behavioral map, transforming behavioral data from multiple consecutive days into a dynamic graph structure. Through graph neural networks (GNNs), it analyzes behavioral change trends and captures potential psychological problems. For example, long-term social avoidance may indicate depressive tendencies, while short-term sleep disturbances may be related to acute stress events, overcoming the shortcomings of traditional static models that cannot identify temporal correlations. Finally, based on a comprehensive score of behavioral deviation and mutability, it classifies mental health status into four levels (healthy, warning, mild abnormality, and severe abnormality) and generates graded intervention suggestions, such as pushing meditation courses at the warning level and triggering emergency assistance at the severe abnormality level, achieving "precise rehabilitation guidance."

[0028] See Figure 1 The diagram shows a flowchart of a student mental health detection method based on multimodal behavior fusion and dynamic weight adjustment according to an embodiment of this application.

[0029] like Figure 1 As shown, a student mental health detection method based on multimodal behavior fusion and dynamic weight adjustment is presented, specifically including steps S100 to S500.

[0030] refer to Figure 1 Step S100: Student behavior data is collected and fused in a non-invasive manner. The student behavior data includes: physiological signal data, behavioral pattern data and digital trace data.

[0031] It should be noted that, in this embodiment, physiological signal data is obtained by real-time monitoring of heart rate variability (HRV) and electrical skin activity (EDA); behavioral pattern data is obtained by quantifying classroom participation (including speaking frequency and head-up rate), social interaction frequency (analysis of the number of people in the cafeteria and library), and daily routine regularity (modeling based on bedtime and bedtime); and digital trace data is obtained by tracking online learning duration, homework submission delay rate, and online search keywords (such as psychologically related words like "anxiety" and "insomnia").

[0032] This innovative approach avoids the direct reliance on sensitive psychological questionnaire data in traditional methods. By replacing subjective self-assessment with objective behavioral data, it effectively reduces student resistance while ensuring the continuity and ecological validity of data collection, providing a more reliable and acceptable solution for mental health assessment.

[0033] In some feasible embodiments, based on the aforementioned scheme, physiological signal data are collected in a non-invasive manner, including: The system uses a smart bracelet to monitor heart rate variability and skin conductance in real time, and calculates the intensity of the autonomic nervous system response. The time-domain analysis of heart rate variability is as follows: ; in, The RMSSD value represents the interval between consecutive heartbeats, where N is the number of heartbeats. The lower the RMSSD value, the stronger the sympathetic nerve activity. Frequency domain analysis of skin electrical activity is as follows:

[0034] Wherein, EDA represents skin electrical activity, SCL represents skin conductance level, t represents time, and SCR represents skin conductance response.

[0035] In some feasible embodiments, based on the aforementioned scheme, behavioral pattern data is collected in a non-invasive manner, including: Behavioral pattern data is captured through classroom cameras and the campus card system, including: classroom participation, frequency of social interaction, and regularity of daily routines; Establish quantitative models for classroom participation, frequency of social interaction, and regularity of daily routines; The classroom participation measurement model is as follows: ; in, and These are the weighting coefficients; The social interaction frequency quantification model is as follows: ; in, It is the ratio of the time spent traveling with the k-th type of companions to the total social time, used to quantify the proportion of time a certain type of companion occupies in the overall social interaction; The quantitative model for regularity of daily routines is as follows:

[0036] in, Let represent the time of falling asleep recorded for the i-th time, where i represents different times of falling asleep, and n represents the total number of times the time of falling asleep has been recorded. Let be the average time to fall asleep over n attempts. Let be the standard deviation of the time to fall asleep n times.

[0037] In some feasible embodiments, based on the aforementioned scheme, digital trace data is collected in a non-invasive manner, including: We obtain online learning engagement, homework delay probability, and risk word frequency through campus network logs and learning platform APIs. Online learning engagement is calculated using the focus index formula, which is as follows: ; The probability of job delay is predicted using a logistic regression model; The risk word frequency is obtained through the risk word frequency calculation formula, which is as follows:

[0038] Where W is a preset risk vocabulary, W = {anxiety, insomnia, self-harm}; TF-IDF value is a weighted score of term frequency - inverse document frequency, and I is an indicator function.

[0039] It should be noted that the student behavior data collected in step S100 is encrypted with student IDs using a hash algorithm, retaining only group attribute tags such as grade and major; and the data packets are encrypted using the AES-256 protocol before being uploaded to the campus private cloud to avoid leakage of sensitive information.

[0040] In addition, to adapt to rapid changes in psychological state, this embodiment updates data in real time through a sliding window (e.g., 7 days).

[0041] Continue to refer to Figure 1 Step S200: Design a lightweight weight allocation network, and dynamically adjust the weights of physiological signal data, behavioral pattern data and digital trace data in student behavior data based on the weight allocation network.

[0042] It should be noted that step S200 further constructs an intelligent weight allocation mechanism. A lightweight weight allocation network is trained based on historical behavioral data. This network can dynamically adjust the data weights of three dimensions—physiological signals, behavioral patterns, and digital traces—according to individual student characteristics (such as gender, grade, and major) and the group's behavioral baseline. For example, for senior high school students, whose academic pressure is highly correlated with their sleep-wake cycle stability, the system will automatically increase the weight coefficient of the "sleep-wake cycle regularity" dimension. For introverted students, the system reduces the weight of the "social interaction frequency" dimension through personality assessment data or historical social behavior analysis, avoiding misjudgments due to differences in social behavior. In terms of algorithm implementation, a lightweight neural network architecture such as TinyML is adopted, directly deployed on edge devices such as smart bracelets or campus gateways. Personalized weight parameters are generated in real time through localized computation, eliminating the need to upload raw data to the cloud. This ensures data processing efficiency (response latency less than 200ms) and eliminates the risk of data privacy leakage from a technical architecture perspective, achieving a dual innovation of "edge intelligence" and "dynamic adaptation."

[0043] In some feasible embodiments, based on the foregoing scheme, step S200 includes: Step S210: Transform individual student characteristics and group behavior into computable vector representations; Step S220: Using the vector representation as the benchmark for weight adjustment, the weights of physiological signal data, behavioral pattern data, and digital trace data are dynamically adjusted using the weight allocation network. The dynamic adjustment rules include: increasing the weight of daily routines for senior high school students and suppressing the weight of social activities for introverted students.

[0044] For example, the conversion process in step S210 is as follows: Individual characteristic coding: ; in, , , , Equal to one-hot encoding or embedding vectors, and Concat represents vector concatenation.

[0045] Baseline calculation of group behavior: ; in, For the k-th student, the original data for the j-th dimension (e.g., daily routine). Data quality weights (such as equipment reliability coefficients). and These are the population mean and variance, respectively.

[0046] It should be noted that step S210 constructs an interpretable baseline model by integrating static features (gender, grade) with dynamic behavioral data, providing a quantitative basis for subsequent weight adjustments.

[0047] It should be noted that, in this embodiment, the network output layer (Sigmoid activation) of the weight allocation network is designed as follows:

[0048] in, The weight coefficient for the j-th dimension (0≤ ≤1), and These are trainable parameters.

[0049] Weight normalization:

[0050] Ensure that the total weight of the three dimensions (physiological signals, behavioral patterns, and digital traces) is 1.

[0051] This embodiment uses a weight allocation network designed with the TinyML architecture. Through a single-layer neural network and normalization constraints, the number of model parameters is compressed to less than 10KB, which is suitable for the computing resources of the MCU (microcontroller) of the smart bracelet.

[0052] For example, the dynamic weight adjustment rules are as follows: The importance of daily routines for senior high school students should be increased:

[0053] Social weight suppression in introverted students:

[0054] Among them, βgrade and γintro are preset hyperparameters that are optimized through A / B testing using historical data.

[0055] It is understandable that this embodiment combines the rule engine with the neural network output to achieve a hybrid decision-making process of "hard constraints" and "soft adjustments" to avoid model overfitting.

[0056] Continue to refer to Figure 1 Step S300: Construct a time-series behavior graph based on the student behavior data after weight allocation.

[0057] It should be noted that step S300 innovatively constructs a temporal behavioral graph, transforming multi-dimensional data such as physiological signals, behavioral patterns, and digital traces collected from students over multiple consecutive days into a dynamic graph structure. Nodes represent individual students, and edge weights reflect the strength of correlations between different behavioral dimensions (such as the correlation between sleep patterns and classroom participation), forming a dynamic evolutionary network connected by a time axis. This graph can accurately capture abnormal behavioral patterns. For example, when a student suddenly exhibits a sharp drop in social interaction frequency or a reversed day-night sleep schedule, the system automatically marks abnormal fluctuations in edge weights and pinpoints the specific behavioral dimension.

[0058] In some feasible embodiments, based on the aforementioned scheme, the construction of a time-series behavioral graph based on the weighted student behavioral data includes: Based on the weighted physiological signal data, behavioral pattern data, and digital trace data, node features are encoded to form initial node features. Calculate edge weights based on the initial features of the nodes; Edge weights exceeding a set threshold are retained to form a temporal behavior graph.

[0059] For example, the node feature encoding formula is as follows: ; in, The physiological signal at time t (e.g., mean HRV). For behavioral patterns (such as social frequency). Digital traces (such as the frequency of search keywords) are spliced ​​together to form the initial features of nodes.

[0060] Edge weight calculation: ; Where Corr is the Pearson correlation coefficient, which calculates the correlation between the behavioral changes of nodes i and j at time t; The time decay factor (λ=0.1) is used to reduce the weight of historical data.

[0061] Animated GIF update:

[0062] Only associations with edge weights exceeding the threshold θ=0.3 are retained to form a sparse dynamic graph.

[0063] Understandably, by using time decay factors and dynamic edge pruning, noise interference can be reduced, recent behavioral correlations can be highlighted, and the timeliness of the map can be improved.

[0064] Continue to refer to Figure 1Step S400: Based on the temporal behavior map, use a graph neural network to identify temporal behaviors related to psychological problems through a message passing mechanism.

[0065] Understandably, this step combines the temporal modeling capabilities of graph structures and graph neural networks to achieve "space-time" dual-dimensional feature extraction, breaking through the limitations of isolated analysis in traditional methods.

[0066] In some feasible embodiments, based on the foregoing scheme, the step of identifying temporal behaviors related to psychological problems using a graph neural network through a message passing mechanism based on the temporal behavior graph includes: By using graph neural networks to capture the temporal evolution features of nodes and edges in a temporal behavior graph, abnormal behaviors can be identified, and long-term risk assessment and short-term acute stress detection can be performed. The long-term risk assessment formula is as follows: ; in, To calculate the average social weight of the i-th type over 30 days, Social weight threshold The standard deviation of bedtime Values ​​are taken over 2 hours; The formula for testing short-term acute stress is as follows: ; in, The rate of change of HRV value over 72 hours. To determine the frequency of searches for "anxiety," Keyword = 5 times / day.

[0067] For example, the recognition process of a graph neural network includes: Node state update (message passing mechanism):

[0068] in, It is the ReLU activation function. and For trainable parameters, Let i be the set of neighbors of node i.

[0069] Temporal feature aggregation:

[0070] By aggregating the temporal features of nodes over the past T=7 days using a graph neural network, the final behavioral representation is generated. .

[0071] Anomaly edge detection:

[0072] The reconstruction error between the predicted and actual edge weights is calculated, and the error exceeds a threshold. If the value is greater than 0.5, it is marked as an anomaly.

[0073] In some feasible embodiments, based on the aforementioned scheme, during the process of identifying abnormal behavior, the threshold for anomaly detection is dynamically optimized using group behavior, including: Calculate the median and variance of the group edge weights as a dynamic baseline; The threshold is updated based on the dynamic baseline.

[0074] For example, the dynamic baseline calculation formula is as follows: ; Threshold update rules: ; in, =0.05 is the learning rate. =0.2 is the lower limit of the threshold to prevent the detection standard from being relaxed too much.

[0075] Understandably, this step uses an online learning mechanism to adapt the threshold to changes in group behavior, avoiding false negatives or missed positives caused by fixed thresholds and improving system robustness.

[0076] Continue to refer to Figure 1 Step S500: Construct an assessment model that considers behavioral deviation and temporal abruptness, use the assessment model to classify psychological states based on the temporal behaviors, and generate corresponding intervention suggestions based on the classified psychological states.

[0077] In some feasible embodiments, based on the foregoing scheme, the step of using the assessment model to classify psychological states based on the temporal behavior and generating corresponding intervention suggestions based on the classified psychological states includes: Establish a baseline of behavioral data for students of the same grade / major as a reference for calculating individual deviations; Individual deviation is calculated based on a baseline of behavioral data, and the psychological state level is determined by combining individual deviation with temporal variability. Differentiated intervention behavior data were matched based on psychological state levels; Feedback and optimization are performed based on behavioral data after intervention.

[0078] For example, the baseline is calculated as follows: ; Where k∈{social frequency, daily routine regularity, mean HRV, ...} represents the behavioral dimensions. Let N be the behavioral data of student i at time t, and N be the number of students in the same group.

[0079] Dynamic baseline update:

[0080] Using exponential moving average ( The baseline is updated using the attenuation coefficient to balance the stability of historical data with the sensitivity to new data.

[0081] It is understandable that a dynamic baseline is used to adapt to seasonal changes in group behavior (such as fluctuations in schedules during exam weeks) and to avoid misjudgments caused by a fixed baseline.

[0082] Behavioral deviation calculation:

[0083] in, Weighting coefficients for information entropy in each dimension to highlight the discriminative power of low-variance dimensions (such as HRV); Let be the Mahalanobis distance of student i, reflecting the degree of deviation from long-term behavioral patterns.

[0084] Temporal abrupt change detection:

[0085] Statistical analysis of the number of daily abrupt changes in behavioral data over 7 days (e.g., a decrease in social frequency exceeding 60%, a sharp drop in HRV value, etc.). These are the mutation thresholds for each dimension.

[0086] Overall rating and classification:

[0087] Among them, Norm( For min-max normalization, and Map to the interval [0,1].

[0088] Understandably, in this step, Mahalanobis distance is used instead of Euclidean distance: the correlation between behavioral dimensions is considered to avoid double scoring of highly correlated dimensions (such as daily routines and classroom participation).

[0089] Mutation Dimension Counting: Quantifying the breadth of acute psychological fluctuations, rather than the magnitude of a single dimension, to improve the ability to identify complex crises.

[0090] It should be noted that matching differentiated intervention behavior data based on psychological state levels can balance accuracy and resource efficiency.

[0091] An example of an intervention resource allocation model is as follows:

[0092] Where R = {meditation courses, counselor appointments, psychological counseling, emergency assistance} represents the intervention resource set. Resource costs (such as time and manpower). The expected intervention effect (predicted using a regression model based on historical data). =1.2 is the effect guarantee coefficient.

[0093] Dynamic priority sorting:

[0094] For students facing multiple risk factors simultaneously (such as family changes + academic pressure), through (Value 0~1) Increase intervention priority.

[0095] Understandably, interventions can bring about the following beneficial effects: Cost-effectiveness optimization: Avoiding the over-allocation of high-cost resources (such as psychological counseling) to low-risk students, the experiment showed that resource utilization increased by 37%.

[0096] Superimposed risk compensation: to prevent high-risk cases from being missed due to a single indicator failing to reach the threshold, and to ensure comprehensive intervention coverage.

[0097] For example, the feedback on the intervention effect is as follows: The hierarchical model and intervention strategies were optimized by analyzing post-intervention behavioral data feedback.

[0098] Intervention effectiveness evaluation:

[0099] Where θeffective = 0.3 is the effective intervention threshold. Mark successful intervention cases.

[0100] Model parameter update:

[0101] For the successful intervention case set S, the weights of each behavioral dimension are adjusted using gradient ascent. , =0.01 is the learning rate.

[0102] Understandably, feedback steps can help the tiered model and intervention strategies adapt to changes in student behavior patterns (such as fluctuations during the initial adjustment period for new students) and strengthen the focus on intervention-sensitive dimensions (such as the weight of social frequency, which may be significantly increased due to meditation courses).

[0103] It should be noted that step S500, based on the dynamic analysis results of the temporal behavior map, constructs a four-level mental health status assessment system, subdividing students' mental states into four levels: healthy, warning, mild abnormality, and severe abnormality. The assessment model comprehensively considers two core indicators: behavioral deviation and temporal abrupt change. Behavioral deviation is quantified by calculating the standard deviation of individual student behavioral data (such as social frequency and regularity of daily routine) compared to the baseline of the same grade / major group, reflecting the degree of deviation from long-term behavioral patterns. Temporal abrupt change is assessed based on the sudden change in behavioral patterns within 7 days (such as a single-day decrease in social interaction frequency exceeding 60% or a HRV value below the threshold for 3 consecutive days), assessing acute psychological fluctuations. The system automatically determines the level of mental state based on the weighted composite score of the two indicators (deviation accounts for 60%, and abrupt change accounts for 40%). For example, if a student's long-term social node weight is 1.5 standard deviations lower than the group baseline (high deviation) and their daily routine is reversed within 72 hours (strong abrupt change), they are judged as severely abnormal.

[0104] For different levels of anxiety, precise, tiered intervention plans are generated: students at the early warning level will receive personalized notifications, including 10-15 minute meditation sessions (such as mindfulness breathing exercises) or reminders for proactive consultations with counselors; mild cases trigger guidance for scheduling campus psychological counseling services; and severe cases immediately initiate emergency assistance procedures, including access to a 24-hour psychological hotline, collaboration with psychiatrists at the university hospital, and notification mechanisms for parents. This tiered response mechanism breaks through the traditional, crude "report any alarm" model. Through differentiated intervention strategies, it avoids over-intervention for healthy students (experiments show a 41% increase in intervention acceptance rate) while ensuring timely assistance for high-risk students (intervention response time for severe cases is shortened to within 15 minutes), achieving a balance between precision and humanization in mental health management.

[0105] The following describes an embodiment of the apparatus described in this application, which can be used to execute a student mental health detection method based on multimodal behavior fusion and dynamic weight adjustment as described in the above embodiments of this application. For details not disclosed in the apparatus embodiments of this application, please refer to the embodiments of the method described in the above applications.

[0106] Reference Figure 2 As shown, a student mental health monitoring device 200 based on multimodal behavior fusion and dynamic weight adjustment according to an embodiment of this application includes: The acquisition unit 201 is used to acquire student behavior data in a non-invasive manner and perform data fusion. The student behavior data includes: physiological signal data, behavior pattern data and digital trace data. The weight allocation unit 202 is used to design a lightweight weight allocation network and dynamically adjust the weights of physiological signal data, behavioral pattern data and digital trace data in student behavior data based on the weight allocation network. Construction unit 203 is used to construct a time-series behavior graph based on student behavior data after weight allocation; The identification unit 204 is used to identify temporal behaviors related to psychological problems based on the temporal behavior map and using a graph neural network through a message passing mechanism. The assessment unit 205 is used to construct an assessment model that considers behavioral deviation and temporal abruptness, use the assessment model to classify psychological states based on the temporal behavior, and generate corresponding intervention suggestions based on the classified psychological states.

[0107] like Figure 3 As shown, this application embodiment also provides an electronic device 300, including a memory 310, a processor 320, and a computer program 311 stored in the memory 310 and executable on the processor. When the processor 320 executes the computer program 311, it implements the steps of the above-mentioned student mental health detection method based on multimodal behavior fusion and dynamic weight adjustment.

[0108] Since the electronic device described in this embodiment is the device used to implement the student mental health detection device based on multimodal behavior fusion and dynamic weight adjustment in the embodiments of this application, those skilled in the art can understand the specific implementation method and various variations of the electronic device in this embodiment based on the method described in the embodiments of this application. Therefore, how the electronic device implements the method in the embodiments of this application will not be described in detail here. Any device used by those skilled in the art to implement the method in the embodiments of this application is within the scope of protection of this application.

[0109] In practice, when the computer program 311 is executed by the processor, it can implement any of the embodiments corresponding to the first aspect.

[0110] Figure 4 A schematic diagram of the structure of a computer system suitable for implementing the electronic device of the present application is shown.

[0111] It should be noted that, Figure 4 The computer system 400 of the electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.

[0112] like Figure 4As shown, the computer system 400 includes a Central Processing Unit (CPU) 401, which can perform various appropriate actions and processes based on a program stored in Read-Only Memory (ROM) 402 or a program loaded from Storage Unit 408 into Random Access Memory (RAM) 403, such as performing the methods described in the above embodiments. The RAM 403 also stores various programs and data required for system operation. The CPU 401, ROM 402, and RAM 403 are interconnected via a bus 404. An Input / Output (I / O) interface 405 is also connected to the bus 404.

[0113] The following components are connected to I / O interface 405: an input section 406 including a keyboard, mouse, etc.; an output section 407 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; a storage section 408 including a hard disk, etc.; and a communication section 409 including a network interface card such as a LAN (Local Area Network) card, modem, etc. The communication section 409 performs communication processing via a network such as the Internet. A drive 410 is also connected to I / O interface 405 as needed. A removable medium 411, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on drive 410 as needed so that computer programs read from it can be installed into storage section 408 as needed.

[0114] Specifically, according to embodiments of this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 409, and / or installed from removable medium 411. When the computer program is executed by central processing unit (CPU) 401, it performs various functions defined in the system of this application.

[0115] It should be noted that the computer-readable medium shown in the embodiments of this application can be a computer-readable signal medium, a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such transmitted data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. The computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to wireless, wired, etc., or any suitable combination thereof.

[0116] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. Each block in a flowchart or block diagram may represent a module, segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0117] The units described in the embodiments of this application can be implemented in software or hardware, and the described units can also be located in a processor. The names of these units do not necessarily limit the specific unit itself.

[0118] In another aspect, this application also provides a computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the student mental health detection method based on multimodal behavior fusion and dynamic weight adjustment described in the above embodiments.

[0119] In another aspect, this application also provides a computer-readable medium, which may be included in the electronic device described in the above embodiments; or it may exist independently and not assembled into the electronic device. The computer-readable medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to implement the student mental health detection method based on multimodal behavior fusion and dynamic weight adjustment described in the above embodiments.

[0120] It should be noted that although several modules or units for the device used to perform actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to the embodiments of this application, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0121] Through the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of this application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, touch terminal, or network device, etc.) to execute the methods according to the embodiments of this application.

[0122] Other embodiments of this application will readily conceive of by those skilled in the art upon consideration of the specification and practice of the embodiments disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. It should be understood that this application is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A student mental health assessment method based on multimodal behavior fusion and dynamic weight adjustment, characterized in that, include: Student behavior data is collected and fused using a non-invasive method. The student behavior data includes: physiological signal data, behavioral pattern data, and digital trace data. Design a lightweight weight allocation network to dynamically adjust the weights of physiological signal data, behavioral pattern data, and digital trace data in student behavior data; Construct a time-series behavior graph based on student behavior data after weight allocation; Based on the aforementioned temporal behavior map, a graph neural network is used to identify temporal behaviors related to psychological problems through a message passing mechanism. An assessment model is constructed that considers behavioral deviation and temporal abruptness. The assessment model is used to classify psychological states based on the temporal behaviors, and corresponding intervention suggestions are generated based on the classified psychological states.

2. The method according to claim 1, characterized in that, Physiological signal data are collected using non-invasive methods, including: The system uses a smart bracelet to monitor heart rate variability and skin conductance in real time, and calculates the intensity of the autonomic nervous system response. The time-domain analysis of heart rate variability is as follows: ； in, The interval between consecutive heartbeats is the interval between consecutive heartbeats of the (i+1)th time, N is the number of heartbeats, and the lower the RMSSD value, the stronger the sympathetic nerve activity; Frequency domain analysis of skin electrical activity is as follows: Wherein, EDA represents skin electrical activity, SCL represents skin conductance level, t represents time, SCR represents skin conductance response, and T represents the time window length for skin electrical activity analysis.

3. The method according to claim 1, characterized in that, Behavioral pattern data is collected in a non-invasive manner, including: Behavioral pattern data is captured through classroom cameras and the campus card system, including: classroom participation, frequency of social interaction, and regularity of daily routines; Establish quantitative models for classroom participation, frequency of social interaction, and regularity of daily routines; The classroom participation measurement model is as follows: ； in, and These are the weighting coefficients; The social interaction frequency quantification model is as follows: ； in, It is the ratio of the time spent traveling with the k-th type of companions to the total social time, used to quantify the proportion of time spent by a certain type of companion in overall social interaction. The total number of peer types represents the number of peer categories involved in students' social interactions. For peer type indexes, the value range is 1, 2, ... This is used to iterate through all companion types; The quantitative model for regularity of daily routines is as follows: in, Let represent the time of falling asleep recorded for the i-th time, where i represents different times of falling asleep, and n represents the total number of times the time of falling asleep has been recorded. Let be the average time to fall asleep over n attempts. Let be the standard deviation of the time to fall asleep n times.

4. The method according to claim 1, characterized in that, Digital trace data is collected using non-invasive methods, including: We obtain online learning engagement, homework delay probability, and risk word frequency through campus network logs and learning platform APIs. Online learning engagement is calculated using the focus index formula, which is as follows: ； The probability of job delay is predicted using a logistic regression model; The risk word frequency is obtained through the risk word frequency calculation formula, which is as follows: Where W is a preset risk vocabulary, W = {anxiety, insomnia, self-harm}; the TF-IDF value is a weighted score of term frequency - inverse document frequency, and I is an indicator function. It refers to a single word in a pre-defined risk term database.

5. The method according to claim 1, characterized in that, The lightweight weighted network design dynamically adjusts the weights of physiological signal data, behavioral pattern data, and digital trace data in student behavior data, including: Transform individual student characteristics and group behavior into computable vector representations; A lightweight weight allocation network is designed using the TinyML architecture; The vector representation is used as the basis for weight adjustment, and the weights of physiological signal data, behavioral pattern data and digital trace data are dynamically adjusted using the weight allocation network. The dynamic adjustment rules include: increasing the weight of daily routines for senior high school students and suppressing the weight of social activities for introverted students.

6. The method according to claim 5, characterized in that, The construction of a time-series behavioral graph based on weighted student behavior data includes: Based on the weighted physiological signal data, behavioral pattern data, and digital trace data, node features are encoded to form initial node features. Calculate edge weights based on the initial features of the nodes; Edge weights exceeding a set threshold are retained to form a temporal behavior graph.

7. The method according to claim 6, characterized in that, The method of identifying temporal behaviors related to psychological problems based on the temporal behavior graph and using a graph neural network through a message passing mechanism includes: By using graph neural networks to capture the temporal evolution features of nodes and edges in a temporal behavior graph, abnormal behaviors can be identified, and long-term risk assessment and short-term acute stress detection can be performed. The long-term risk assessment formula is as follows: ； in, To calculate the average social weight of the i-th type over 30 days, Social weight threshold The standard deviation of bedtime Values ​​are taken over 2 hours; The formula for testing short-term acute stress is as follows: ； in, The rate of change of HRV value over 72 hours. To determine the frequency of searches for "anxiety," Keyword = 5 times / day.

8. The method according to claim 7, characterized in that, In the process of identifying anomalous behavior, the threshold for anomaly detection is dynamically optimized using group behavior, including: Calculate the median and variance of the group edge weights as a dynamic baseline; The threshold is updated based on the dynamic baseline.

9. The method according to claim 7, characterized in that, The step of using the assessment model to classify psychological states based on the temporal behavior, and generating corresponding intervention suggestions based on the classified psychological states, includes: Establish a baseline of behavioral data for students of the same grade / major as a reference for calculating individual deviations; Individual deviation is calculated based on a baseline of behavioral data, and the psychological state level is determined by combining individual deviation with temporal variability. Differentiated intervention behavior data were matched based on psychological state levels; Feedback and optimization are performed based on behavioral data after intervention.

10. A student mental health monitoring device based on multimodal behavior fusion and dynamic weight adjustment, characterized in that, include: The acquisition unit is used to acquire student behavior data in a non-invasive manner and perform data fusion. The student behavior data includes: physiological signal data, behavioral pattern data, and digital trace data. A weight allocation unit is used to design a lightweight weight allocation network and dynamically adjust the weights of physiological signal data, behavioral pattern data, and digital trace data in student behavior data based on the weight allocation network. The building unit is used to construct a time-series behavior graph based on student behavior data after weight allocation. The identification unit is used to identify temporal behaviors related to psychological problems based on the temporal behavior map and using a graph neural network through a message passing mechanism. The assessment unit is used to construct an assessment model that considers behavioral deviation and temporal abruptness. The assessment model is used to classify psychological states based on the temporal behaviors, and corresponding intervention suggestions are generated based on the classified psychological states.

Images