A mental illness pre-diagnosis information processing method and system

Abstract

The application discloses a kind of psychological disease pre-diagnosis information processing method and system, method includes: the multi-modal behavior signal and physiological parameter signal of user are collected, and respectively carry out the differential privacy protection processing based on edge computing and the noise separation processing based on wavelet transform, obtain standardized behavior characteristic sequence and time sequence physiological characteristic vector;Behavior node feature and psychological state marker are extracted from standardized behavior characteristic sequence, and three-dimensional association matrix is constructed in combination with time sequence physiological characteristic vector;Potential risk probability is calculated based on graph neural network model, generates pre-diagnosis grading result, and through gradual protocol fusion objective behavior environment data, output personalized intervention strategy, finally generate comprehensive pre-diagnosis report.The application realizes the accurate modeling of psychological-physiological-behavior dynamic association by dynamic knowledge graph construction and cross-modal fusion analysis, while giving consideration to data privacy protection, effectively improves the accuracy and practicality of mental health monitoring.

Description

Technical Field

[0001] This invention relates to the field of mental health monitoring technology, specifically to a method and system for processing information for the pre-diagnosis of mental illness. Background Technology

[0002] In recent years, mental health monitoring technology has been applied in practice, providing a new technological path for early risk identification by integrating multi-source data from wearable devices and mobile terminals. Existing systems typically use time series analysis to process physiological signals and combine them with machine learning models to analyze user behavior patterns, which has to some extent assisted in the diagnosis of common psychological problems such as depression and anxiety. However, in actual deployment, due to significant differences in time scale and sampling frequency among different modalities, it is difficult to accurately capture the dynamic correlation between behavioral events and physiological responses. On the other hand, privacy protection requirements necessitate strict anonymization of raw data, but traditional encryption methods often disrupt the continuity of behavioral characteristics, making it impossible to effectively extract some diagnostically valuable time-series patterns. These limitations make the system prone to misjudgment in practical applications, either over-interpreting transient abnormalities caused by environmental factors as psychological problems or ignoring the potential risks indicated by weak signals from multiple modalities. Summary of the Invention

[0003] In view of the above problems, the present invention provides a method and system for processing information for the pre-diagnosis of mental illness. By establishing an analytical framework that can both protect user privacy and accurately reflect the dynamic relationship between psychology, physiology and behavior, the problem of the difficulty in effectively integrating and analyzing multi-source heterogeneous data is solved.

[0004] To achieve the above objectives, in a first aspect, this application provides a method for processing information for the pre-diagnosis of mental illness, comprising:

[0005] The system collects multimodal behavioral signals and physiological parameter signals from users. The multimodal behavioral signals include keyboard interaction dynamics, screen touch trajectory features, and environmental acoustic marker features. The physiological parameter signals are indirectly acquired through built-in sensors in the device.

[0006] The multimodal behavioral signal is subjected to a first preprocessing to obtain a standardized behavioral feature sequence. The first preprocessing is configured as differential privacy protection processing based on edge computing. The physiological parameter signal is subjected to a second preprocessing to obtain a temporal physiological feature vector. The second preprocessing is configured as noise separation processing based on wavelet transform.

[0007] Behavioral node features and psychological state markers are extracted from standardized behavioral feature sequences. Behavioral node features, psychological state markers, and temporal physiological feature vectors are then combined to construct a three-dimensional association matrix using a dynamic knowledge graph. The dimensions of the three-dimensional association matrix include behavioral pattern dimension, time dimension, and physiological response dimension.

[0008] Based on the graph neural network model, the potential risk probability is calculated on the three-dimensional correlation matrix to obtain the pre-diagnosis and classification results. The pre-diagnosis and classification results are fused with the user's objective behavioral environment across modalities using a progressive protocol to generate personalized intervention strategies. The pre-diagnosis and classification results are configured as a multi-dimensional output including risk level, development trajectory and vulnerability factors.

[0009] Based on behavioral node characteristics, pre-diagnosis grading results, and personalized intervention strategies, a comprehensive pre-diagnosis report is generated, which includes an individual psychological profile, environmental adaptability assessment, and intervention effect prediction.

[0010] In some embodiments, the multimodal behavioral signal undergoes a first preprocessing step to obtain a standardized behavioral feature sequence. This first preprocessing step is configured as edge-computation-based differential privacy protection processing, including:

[0011] Keystroke interval analysis and pressure intensity analysis are performed on the keyboard interaction dynamics characteristics to generate the first row of feature sub-vectors. Keystroke interval analysis includes calculating the standard deviation of the time difference between adjacent keystrokes, and pressure intensity analysis includes statistically analyzing the kurtosis characteristics of pressure values.

[0012] The sliding trajectory curvature analysis and touch point distribution analysis are performed on the screen touch trajectory characteristics to generate the second row of feature sub-vectors. The sliding trajectory curvature analysis uses cubic spline interpolation to calculate the average curvature, and the touch point distribution analysis includes calculating the spatial autocorrelation index of touch point density.

[0013] The acoustic signature features of the environment are analyzed by acoustic entropy value analysis and silence duration analysis to generate a third row of feature sub-vectors. The acoustic entropy value analysis uses Mel frequency cepstral coefficients to calculate spectral entropy, and the silence duration analysis includes statistically analyzing the duration distribution of silence segments per unit time.

[0014] The first row feature vector, the second row feature vector, and the third row feature vector are subjected to feature fusion processing based on attention weights to obtain the intermediate vector of behavior features.

[0015] The intermediate vector of behavioral features is subjected to noise addition processing based on the Laplacian mechanism. The noise scale parameter of the noise addition processing is dynamically adjusted according to the feature sensitivity to generate a noisy feature vector that meets the preset differential privacy requirements.

[0016] The noisy feature vectors are subjected to Z-score normalization transformation to obtain the normalized behavioral feature sequence.

[0017] In some embodiments, the physiological parameter signal undergoes a second preprocessing to obtain a time-series physiological feature vector. The second preprocessing is configured as a noise separation process based on wavelet transform, including:

[0018] The physiological parameter signal is subjected to multi-scale wavelet decomposition to generate a set of wavelet coefficients containing approximation coefficients and detail coefficients. The multi-scale wavelet decomposition is configured to be performed using orthogonal wavelet basis functions with compact support characteristics.

[0019] Adaptive threshold denoising is performed on the wavelet coefficient set to obtain denoised wavelet coefficients. The adaptive threshold denoising process includes threshold functions at various scales, and the threshold functions at various scales are configured to be determined based on the statistical distribution characteristics of the wavelet coefficients.

[0020] The denoised wavelet coefficients are subjected to signal reconstruction processing to obtain denoised parameter signals. The signal reconstruction processing is configured to filter and retain scale coefficients that reflect physiological rhythm characteristics.

[0021] The denoised parameter signal is processed by time-domain feature extraction to generate a time-series physiological feature vector. The time-domain feature extraction process includes calculating time-varying feature parameters that reflect the activity characteristics of the autonomic nervous system.

[0022] In some embodiments, extracting behavioral node features and mental state markers from a standardized behavioral feature sequence includes:

[0023] High-information-content node detection processing is performed on standardized behavioral feature sequences to identify behavioral node features. The high-information-content node detection processing is configured to perform detection based on the significance level of behavioral feature changes.

[0024] The identified behavioral node features are subjected to psychological state association analysis to extract psychological state markers. The psychological state association analysis is configured to perform feature matching based on a preset psychological-behavioral mapping relationship model.

[0025] Temporal continuity verification is performed on behavioral node features and psychological state markers to ensure the temporal consistency of feature extraction. The temporal continuity verification is configured to perform state transition probability analysis based on a Markov chain model.

[0026] The verified behavioral node features and psychological state markers are processed by feature encoding to generate structured features.

[0027] In some embodiments, a three-dimensional association matrix is ​​constructed using a dynamic knowledge graph by combining behavioral node features, psychological state markers, and temporal physiological feature vectors, including:

[0028] The structured features are encoded using behavioral pattern dimensions to generate behavioral pattern feature vectors.

[0029] Physiological response dimension encoding is performed on the time-series physiological feature vector to generate physiological response feature vector;

[0030] A dynamic relationship is established between the behavioral pattern dimension, the time dimension, and the physiological response dimension. The dynamic relationship is configured to learn multi-dimensional feature interaction based on graph attention network.

[0031] The behavioral pattern feature vector, physiological response feature vector, and their dynamic correlation are represented in a matrix to generate a three-dimensional correlation matrix.

[0032] In some embodiments, the pre-diagnosis and grading results are obtained by calculating the potential risk probability of the three-dimensional correlation matrix based on a graph neural network model, including:

[0033] The three-dimensional correlation matrix is ​​subjected to graph structure modeling to construct a dynamic heterogeneous graph network. The graph structure modeling is configured to model the behavioral pattern dimension, time dimension and physiological response dimension as different types of nodes respectively.

[0034] Multi-hop neighbor feature aggregation is performed on dynamic heterogeneous graph networks to generate node-enhanced feature representations. The multi-hop neighbor feature aggregation is configured to use a gated attention mechanism to achieve cross-dimensional feature propagation.

[0035] Risk probability prediction processing is performed on the node augmentation feature representation to calculate the potential risk probability of each mental illness category. The risk probability prediction processing is configured to be implemented as a multi-label classifier based on the sigmoid activation function.

[0036] The potential risk probability is graded and thresholded to generate a pre-diagnosis grading result. The grading threshold judgment process is configured to set multiple risk thresholds based on clinical diagnostic criteria.

[0037] In some embodiments, a progressive protocol is used to cross-modal fuse pre-diagnosis and grading results with the user's objective behavioral environment to generate personalized intervention strategies, including:

[0038] The pre-diagnosis grading results are processed for risk level analysis to generate pre-diagnosis risk level intervals. The risk level analysis process is configured to perform interval matching based on preset multi-level risk thresholds.

[0039] The objective behavioral environment of the user is processed to extract environmental features and generate an environmental feature vector. The environmental feature extraction process is configured to extract contextual features from user device usage records and application logs.

[0040] The pre-diagnostic risk level intervals and environmental feature vectors are aligned across modally to establish a risk-environment association mapping. The cross-modal alignment process is configured to use a cross-modal attention mechanism to achieve feature space alignment.

[0041] Intervention strategies are generated and processed based on risk-environment correlation mapping, and personalized intervention strategies are output.

[0042] Personalized intervention strategies are encoded to generate an executable sequence of intervention instructions. The strategy encoding process is configured to encapsulate strategy parameters and execution conditions in JSON format.

[0043] In some embodiments, the method further includes:

[0044] Based on the antifragility feedback mechanism, the personalized intervention strategy is dynamically adjusted. When the user's psychological resilience is detected to be improved, negative weight decay is automatically triggered, and an adaptive execution plan is generated that includes the intensity of digital companionship, environmental adjustment gradient, and artificial intervention disguise strategy.

[0045] In some embodiments, dynamically tuning personalized intervention strategies based on an antifragility feedback mechanism includes:

[0046] Collect user response behavior characteristics, and perform psychological resilience assessment on the response behavior characteristics to generate resilience improvement indicators. The psychological resilience assessment includes multi-dimensional analysis based on behavioral pattern change characteristics and physiological parameter stability characteristics.

[0047] The current personalized intervention strategy is obtained and its utility is evaluated to generate a strategy utility score. The utility evaluation process is configured to combine short-term effect indicators and long-term trend indicators for comprehensive calculation.

[0048] Based on the resilience enhancement index and strategy utility score, the strategy parameters are optimized, and tuning instructions are generated. The strategy parameter optimization is configured to be implemented using a dynamic weight adjustment algorithm.

[0049] The parameter tuning instructions are updated to personalize the intervention strategy, generating an adaptive execution plan.

[0050] In a second aspect, the present invention also provides a mental illness pre-diagnosis information processing system, applicable to the mental illness pre-diagnosis information processing method described in the first aspect. The system includes: a signal acquisition unit, a signal processing unit, and a pre-diagnosis analysis unit. The signal acquisition unit is used to acquire multimodal behavioral signals and physiological parameter signals of the user. The multimodal behavioral signals include keyboard interaction dynamics features, screen touch trajectory features, and environmental acoustic marker features. The signal processing unit is used to perform a first preprocessing on the multimodal behavioral signals to obtain a standardized behavioral feature sequence. The first preprocessing is configured as differential privacy protection processing based on edge computing. The system also performs a second preprocessing on the physiological parameter signals to obtain a temporal physiological feature vector. The second preprocessing is configured as noise separation processing based on wavelet transform. The pre-diagnosis analysis unit is used to... Behavioral node features and psychological state markers are extracted from standardized behavioral feature sequences. These features, along with temporal physiological feature vectors, are used to construct a three-dimensional association matrix using a dynamic knowledge graph. The dimensions of this matrix include behavioral pattern, time, and physiological response. A graph neural network model is used to calculate the potential risk probability of the three-dimensional association matrix, yielding a pre-diagnosis and grading result. A progressive protocol is employed to fuse the pre-diagnosis and grading result with the user's objective behavioral environment across modalities, generating personalized intervention strategies. The pre-diagnosis and grading result is configured as a multi-dimensional output including risk level, development trajectory, and vulnerability factors. Based on the behavioral node features, pre-diagnosis and grading result, and personalized intervention strategies, a comprehensive pre-diagnosis report is generated, including an individual psychological profile, environmental adaptability assessment, and intervention effect prediction.

[0051] Unlike existing technologies, the present invention provides a method and system for processing information for the pre-diagnosis of mental illness. The method includes: collecting multimodal behavioral signals and physiological parameter signals from users, and performing differential privacy protection processing based on edge computing and noise separation processing based on wavelet transform, respectively, to obtain standardized behavioral feature sequences and temporal physiological feature vectors; extracting behavioral node features and psychological state markers from the standardized behavioral feature sequences, and constructing a three-dimensional correlation matrix of behavioral patterns, time dimensions, and physiological response dimensions by combining the temporal physiological feature vectors; calculating potential risk probabilities based on a graph neural network model, generating pre-diagnosis grading results, and fusion of objective behavioral environment data through a progressive protocol to output personalized intervention strategies; finally generating a comprehensive pre-diagnosis report including individual psychological profiles, environmental adaptability assessments, and intervention effect predictions. This invention achieves accurate modeling of the dynamic correlation between psychology, physiology, and behavior through dynamic knowledge graph construction and cross-modal fusion analysis, while also protecting data privacy, effectively improving the accuracy and practicality of mental health monitoring.

[0052] The above description of the invention is merely an overview of the technical solution of this application. In order to enable those skilled in the art to better understand the technical solution of this application and to implement it based on the description and drawings, and to make the above-mentioned objectives and other objectives, features and advantages of this application easier to understand, the following description is provided in conjunction with the specific embodiments and drawings of this application. Attached Figure Description

[0053] The accompanying drawings are only used to illustrate the principles, implementation methods, applications, features, and effects of specific embodiments of the present invention and other related contents, and should not be considered as limitations on this application.

[0054] In the accompanying drawings of the instruction manual:

[0055] Figure 1 This is a flowchart illustrating steps S101 to S105 of the processing method described in the specific implementation embodiment;

[0056] Figure 2 This is a flowchart illustrating steps S201 to S206 of the processing method described in the specific implementation embodiment;

[0057] Figure 3 This is a schematic diagram of the processing system described in a specific embodiment.

[0058] The reference numerals in the accompanying drawings of the above specification are as follows:

[0059] 1. Processing system;

[0060] 11. Signal acquisition unit;

[0061] 12. Signal processing unit;

[0062] 13. Pre-diagnosis analysis unit. Detailed Implementation

[0063] To illustrate the possible application scenarios, technical principles, implementable specific solutions, and achievable objectives and effects of this application in detail, the following description, in conjunction with the listed specific embodiments and accompanying drawings, provides a detailed explanation. The embodiments described herein are merely illustrative of the technical solutions of this application and are therefore intended to limit the scope of protection of this application.

[0064] In this document, the term "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The term "embodiment" appearing in various places throughout the specification does not necessarily refer to the same embodiment, nor does it specifically limit its independence or connection with other embodiments. In principle, in this application, as long as there are no technical contradictions or conflicts, the technical features mentioned in each embodiment can be combined in any way to form corresponding implementable technical solutions.

[0065] Unless otherwise defined, the technical terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the use of related terms herein is merely for the purpose of describing particular embodiments and is not intended to limit this application.

[0066] In the description of this application, the term "and / or" is used to describe the logical relationship between objects, indicating that three relationships can exist. For example, A and / or B means: A exists, B exists, and A and B exist simultaneously. Additionally, the character " / " in this document generally indicates that the preceding and following objects have an "or" logical relationship.

[0067] In this application, terms such as “first” and “second” are used only to distinguish one entity or operation from another, and do not necessarily require or imply any actual quantity, hierarchy or order relationship between these entities or operations.

[0068] Without further limitations, the use of terms such as “comprising,” “including,” “having,” or other similar open-ended expressions in this application is intended to cover non-exclusive inclusion, which does not exclude the presence of additional elements in a process, method, or product that includes the stated elements, such that a process, method, or product that includes a list of elements may include not only those defined elements but also other elements not expressly listed, or elements inherent to such a process, method, or product.

[0069] As understood in the Examination Guidelines, in this application, expressions such as "greater than," "less than," and "exceeding" are understood to exclude the stated number; expressions such as "above," "below," and "within" are understood to include the stated number. Furthermore, in the description of the embodiments in this application, "multiple" means two or more (including two), and similar expressions related to "multiple" are also understood in this way, such as "multiple groups" and "multiple times," unless otherwise explicitly specified.

[0070] In the description of the embodiments of this application, the space-related expressions used, such as "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "vertical," "top," "bottom," "inner," "outer," "clockwise," "counterclockwise," "axial," "radial," and "circumferential," indicate the orientation or positional relationship based on the orientation or positional relationship shown in the specific embodiments or drawings. They are only for the purpose of describing the specific embodiments of this application or for the reader's understanding, and do not indicate or imply that the device or component referred to must have a specific position, a specific orientation, or be constructed or operated in a specific orientation. Therefore, they should not be construed as limitations on the embodiments of this application.

[0071] The processor described in the embodiments of this application can be implemented by hardware, firmware, software, or a combination thereof. It can be a circuit, one or more of an application-specific integrated circuit (ASIC), a digital signal processor (DSP), a digital signal processing device (DSPD), a programmable logic device (PLD), a field-programmable gate array (FPGA), a central processing unit (CPU), a controller, a microcontroller, or a microprocessor. It also includes other physical, biological, or chemical structures that can implement the same or equivalent functions as the processors listed above, such as biological neurons, quantum computing units, DNA computing units, etc., so that the processor can execute some or all of the steps in the computer program or method involved in the various embodiments of this application, or any combination of the steps mentioned therein.

[0072] The computer program involved in the embodiments can be stored in a computer device readable storage medium, which includes, but is not limited to, disks, magnetic tapes, magnetic cards, floppy disks, flash memory, optical disks, optical cards, read-only memory (ROM), random access memory (RAM), erasable programmable ROM (EPROM), and electrically erasable programmable ROM (EEPROM), etc., and also includes other biological, physical, or chemical structures that can achieve the same or equivalent functions as the storage media listed above, such as DNA, RNA, proteins, and other units with information storage capabilities. In specific embodiments, the storage medium involved can be one of the above-mentioned media types, or a combination of the above-mentioned media types. In different embodiments, the computer program involved in the embodiments can be centrally stored in a single medium, or distributed and stored in multiple media. The memory containing the computer device readable storage medium can be non-volatile memory or random access memory. These computer device readable storage media can be built into the device, or can be connected to the device involved in the embodiments as an external device or part of an external device. In some embodiments, the memory having a computer device readable storage medium is deployed locally; in other embodiments, the memory may be deployed remotely from the processor, for example, as a network-attached memory accessed via RF circuitry or an external port and a communication network, wherein the communication network may be the Internet, one or more intranets, a local area network (LAN), a wide area network (WLAN), a storage area network (SAN), or a suitable combination thereof, as long as computer device access to the memory is enabled. Furthermore, the computer program involved in the embodiments may be stored in plaintext / ciphertext form, or it may be designed as training data, integrated and recombined through model training and implicitly stored in the parameter states of a deep neural network or other machine learning model.

[0073] Please see Figure 1 In a first aspect, this embodiment provides a method for processing information for the pre-diagnosis of mental illness, including:

[0074] S101. Collect the user's multimodal behavioral signals and physiological parameter signals. The multimodal behavioral signals include keyboard interaction dynamics, screen touch trajectory features and environmental acoustic marker features. The physiological parameter signals are indirectly obtained through the device's built-in sensors.

[0075] S102. Perform a first preprocessing on the multimodal behavioral signal to obtain a standardized behavioral feature sequence. The first preprocessing is configured as differential privacy protection processing based on edge computing. Perform a second preprocessing on the physiological parameter signal to obtain a temporal physiological feature vector. The second preprocessing is configured as noise separation processing based on wavelet transform.

[0076] S103. Extract behavioral node features and psychological state markers from the standardized behavioral feature sequence, and construct a three-dimensional association matrix by combining the behavioral node features, psychological state markers and temporal physiological feature vectors through a dynamic knowledge graph. The dimensions of the three-dimensional association matrix include behavioral pattern dimension, time dimension and physiological response dimension.

[0077] S104. Based on the graph neural network model, the potential risk probability is calculated on the three-dimensional correlation matrix to obtain the pre-diagnosis and classification results. The pre-diagnosis and classification results are fused with the user's objective behavioral environment across modalities using a progressive protocol to generate personalized intervention strategies. The pre-diagnosis and classification results are configured as a multi-dimensional output containing risk level, development trajectory and vulnerability factors.

[0078] S105. Generate a comprehensive pre-diagnosis report that includes individual psychological profiles, environmental adaptability assessments, and intervention effect predictions based on behavioral node characteristics, pre-diagnosis grading results, and personalized intervention strategies.

[0079] In step S101, multimodal behavioral signals refer to feature data obtained through users' daily digital interaction behaviors. Keyboard interaction dynamics features characterize operational characteristics such as key pressure and interval time when typing; screen touch trajectory features reflect touch screen behavior patterns such as swiping and clicking; and environmental acoustic marker features collect environmental sound characteristics through the device's microphone. Preferably, physiological parameter signals are indirectly obtained through sensors such as accelerometers and gyroscopes built into the smart terminal, including physiological indicators such as heart rate variability and skin conductance. These signals collectively constitute the basic data source for psychological state assessment.

[0080] In step S102, preferably, the first preprocessing adopts an edge computing architecture to achieve differential privacy protection. Privacy and data availability are balanced by performing two-stage processing of noise addition and feature extraction locally on the user terminal: first, Laplace noise that meets the differential privacy requirements is added to the original behavioral signal, and then a standardized behavioral feature sequence is generated by an anti-noise feature extraction algorithm (such as local feature aggregation based on sliding window). The staged processing ensures that the noise addition only affects the original signal and preserves the continuity of the feature level. At the same time, the edge computing architecture ensures that the original data does not leave the terminal device.

[0081] Preferably, the noise parameters are dynamically adjusted according to the behavioral feature type: the keyboard interaction dynamics feature adopts a low privacy budget to protect the sensitive keystroke mode, the screen touch trajectory feature adopts a medium noise to preserve the overall shape of the gesture, and the environmental acoustic marker feature adopts a segmented noise addition strategy to distinguish the speech content from the environmental noise.

[0082] In step S103, behavioral node features refer to key behavioral event markers extracted from standardized behavioral feature sequences, while psychological state markers are combinations of behavioral features reflecting specific psychological states. The dynamic knowledge graph establishes a three-dimensional association between discrete behavioral node features, psychological state markers, and continuous temporal physiological feature vectors through graph structure modeling. The behavioral pattern dimension encodes the behavioral feature type, the time dimension aligns with the timestamps of multi-source data, and the physiological response dimension records the corresponding physiological change magnitude. This association method accurately captures the dynamic coupling relationship between psychology, physiology, and behavior.

[0083] In step S104, preferably, the progressive protocol implements cross-modal analysis through a three-level fusion architecture: the first level uses a dynamic time warping algorithm to align the timestamps of behavioral events (i.e., standardized behavioral feature sequences) and physiological responses (i.e., temporal physiological feature vectors) in the time dimension; the second level uses a graph attention mechanism to calculate the spatial correlation weights between behavioral node features and temporal physiological feature vectors to generate an intermodal correlation matrix; the third level introduces an environmental context gateway to perform decision-level fusion of the pre-diagnosis and grading results with the objective behavioral environment (including environmental metadata such as device usage time and interactive application type).

[0084] Preferably, the environmental context gateway employs a gating mechanism to filter out irrelevant environmental factors, retaining only environmental features significantly related to psychological state (such as the correlation between frequent nighttime use of social media and anxiety scores). The resulting personalized intervention strategy includes suggestions for environmental adaptation. This hierarchical fusion mechanism avoids the risk of privacy breaches of the original environmental data while ensuring the environmental relevance of the intervention strategy.

[0085] In step S105, the individual psychological profile integrates behavioral node characteristics and pre-diagnosis grading results to construct a user psychological characteristic file. Environmental adaptability assessment analyzes the stability of the user's psychological state under different environments. Intervention effect prediction models the probability of effectiveness of different intervention measures based on historical data. The comprehensive pre-diagnosis report presents the multi-dimensional analysis results in a visual manner, providing objective evidence for professional diagnosis.

[0086] This embodiment achieves accurate assessment of mental health status through a complete technical chain encompassing multi-source data acquisition, privacy protection processing, dynamic correlation modeling, and intelligent analysis and decision-making. First, it ensures data quality and privacy security; then, it establishes a dynamic model reflecting the intrinsic connection between psychology and physiology; and finally, it outputs assessment results with clinical reference value. This embodiment places particular emphasis on protecting data privacy without sacrificing feature information and achieves the organic integration of multi-dimensional data through knowledge graphs, overcoming the limitations of fragmented data analysis.

[0087] Please see Figure 2In some embodiments, the multimodal behavioral signal undergoes a first preprocessing step to obtain a standardized behavioral feature sequence. This first preprocessing is configured as edge-computation-based differential privacy protection processing, including:

[0088] S201. Perform keystroke interval analysis and pressure intensity analysis on the keyboard interaction dynamics characteristics to generate the first row of feature sub-vectors. The keystroke interval analysis includes calculating the standard deviation of the time difference between adjacent keystrokes, and the pressure intensity analysis includes statistically analyzing the kurtosis characteristics of the pressure values.

[0089] S202. Perform sliding trajectory curvature analysis and touch point distribution analysis on the screen touch trajectory features to generate a second row of feature sub-vectors. The sliding trajectory curvature analysis uses cubic spline interpolation to calculate the average curvature. The touch point distribution analysis includes calculating the spatial autocorrelation index of touch point density.

[0090] S203. Perform acoustic entropy analysis and silence duration analysis on the environmental acoustic marker features to generate a third row of feature sub-vectors. The acoustic entropy analysis uses Mel frequency cepstral coefficients to calculate the spectral entropy. The silence duration analysis includes statistically analyzing the duration distribution of silence segments per unit time.

[0091] S204. Perform attention weight-based feature fusion processing on the first row feature sub-vector, the second row feature sub-vector and the third row feature sub-vector to obtain the intermediate behavior feature vector.

[0092] S205. Perform noise addition processing on the intermediate vector of behavioral features based on the Laplacian mechanism. The noise scale parameter of the noise addition processing is dynamically adjusted according to the feature sensitivity to generate a noisy feature vector that meets the preset differential privacy requirements.

[0093] S206. Perform Z-score standardization transformation on the noisy feature vector to obtain the standardized behavioral feature sequence.

[0094] It should be noted that these steps are all completed locally on the user's terminal device. The original multimodal behavioral signal data is destroyed immediately after processing, and the noise scale parameter is inversely proportional to the user's privacy authorization level.

[0095] In step S201, the keystroke interval analysis of keyboard interaction dynamics adopts a dynamic baseline adjustment strategy, establishes a personalized reference threshold based on the user's historical interaction patterns, and automatically triggers the sensitivity adjustment of the feature extraction algorithm when a sudden change in keystroke rhythm is detected in real time; the pressure intensity analysis combines the device hardware characteristics to normalize the pressure value, establishes a pressure-time differential curve to identify the force fluctuation pattern with psychological state indication significance, and enhances the behavioral representation ability by capturing the nonlinear change pattern of force during keystroke.

[0096] In step S202, preferably, the sliding trajectory curvature analysis introduces a speed-weighted correction factor to achieve adaptive sampling rate processing, dynamically adjusts the spline interpolation density according to the touch point movement speed, and assigns higher weight to low-speed touch points when calculating the cubic spline interpolation curvature to capture hesitant gesture features, ensuring that both fast sliding and fine operation can be accurately modeled; preferably, the touch point distribution analysis uses an improved Moran's I exponent to calculate spatial autocorrelation, and adapts to touch point distribution evaluation under different screen sizes by setting a dynamic neighborhood radius.

[0097] In step S203, the voiceprint entropy analysis uses frequency band weighted entropy calculation based on a psychoacoustic model, focusing on enhancing the entropy contribution of the human voice sensitive frequency band; the silence duration analysis identifies effective silence segments through a hidden Markov model, filtering out pseudo-silent intervals caused by equipment operation noise.

[0098] In step S204, during the feature fusion stage, the calculation of attention weights can incorporate cross-modal correlation constraints. First, feature importance is assessed within each modality, then cross-modal weight allocation is performed, creating a dual attention focus. This ensures that the weight allocation of keyboard interaction dynamics features, screen touch trajectory features, and environmental acoustic marker features satisfies prior clinical knowledge for psychological state assessment. For example, in depressive tendency analysis, the fusion weight of acoustic features can be appropriately increased. This mechanism can better capture the changes in the relative importance of various behavioral features under different psychological states.

[0099] In step S205, the noise addition process employs an adaptive Laplace mechanism, with the feature sensitivity grading standard as follows: keystroke interval features have the highest sensitivity, followed by pressure intensity features, then touch curvature features, and finally acoustic entropy features have the lowest sensitivity. Preferably, the noise scale parameter of the keystroke interval feature is determined through simulated attack experiments, effectively resisting privacy inferences based on time-series analysis; noise injection for the touch curvature feature is completed in the frequency domain, preserving the overall pattern of spatial distribution; noise addition for the acoustic entropy feature uses segmented processing guided by speech activity detection, implementing strong noise protection only for non-speech segments. Furthermore, the noise scale parameter decays exponentially according to the privacy level, ensuring stronger protection for highly sensitive features.

[0100] In step S206, the Z-score normalization transformation process uses the user's historical behavior data as a benchmark. It calculates the mean and variance of each feature through a sliding window for dynamic normalization, eliminating the baseline offset caused by device differences, while preserving the user's unique behavior pattern features and avoiding behavior pattern distortion caused by over-standardization.

[0101] This embodiment achieves precise analysis of behavioral signals and synergistic optimization of privacy protection through multi-level feature processing. Keyboard interaction dynamics features employ dynamic baseline adjustment and pressure normalization to capture the psychological implications of input behavior; screen touch trajectory features utilize speed-weighted curvature analysis and improved spatial autocorrelation index to achieve multi-scale gesture modeling; environmental acoustic marker features enhance the accuracy of human voice frequency band analysis based on a psychoacoustic model. The feature fusion stage employs a dual attention mechanism, dynamically weighting each modality feature according to clinical priors. Privacy protection utilizes a hierarchical differential privacy strategy, implementing temporal analysis resistance processing for keystroke interval features, adding noise to touch curvature features in the frequency domain, and segmented protection for acoustic features using voice activity detection. Finally, dynamic normalization driven by user historical data eliminates device differences while preserving individual behavioral pattern characteristics.

[0102] This embodiment achieves high-fidelity conversion of multimodal behavioral signals in an edge computing environment, solving the problems of individual differences and noise interference in mobile device signal acquisition. The hierarchical privacy protection mechanism not only meets strict privacy requirements but also preserves the psychological discrimination information of behavioral characteristics, providing a privacy-compliant and clinically interpretable data foundation for subsequent knowledge graph construction.

[0103] In some embodiments, the physiological parameter signal undergoes a second preprocessing to obtain a time-series physiological feature vector. The second preprocessing is configured as a noise separation process based on wavelet transform, including:

[0104] The physiological parameter signal is subjected to multi-scale wavelet decomposition to generate a set of wavelet coefficients containing approximation coefficients and detail coefficients. The multi-scale wavelet decomposition is configured to be performed using orthogonal wavelet basis functions with compact support characteristics.

[0105] Adaptive threshold denoising is performed on the wavelet coefficient set to obtain denoised wavelet coefficients. The adaptive threshold denoising process includes threshold functions at various scales, and the threshold functions at various scales are configured to be determined based on the statistical distribution characteristics of the wavelet coefficients.

[0106] The denoised wavelet coefficients are subjected to signal reconstruction processing to obtain denoised parameter signals. The signal reconstruction processing is configured to filter and retain scale coefficients that reflect physiological rhythm characteristics.

[0107] The denoised parameter signal is processed by time-domain feature extraction to generate a time-series physiological feature vector. The time-domain feature extraction process includes calculating time-varying feature parameters that reflect the activity characteristics of the autonomic nervous system.

[0108] In this embodiment, physiological parameter signals can be collected by biosensors built into the wearable device. The sampling frequency is dynamically adjusted according to the target physiological characteristics. For example, a higher sampling rate is used for heart rate variability analysis, while the sampling requirement is appropriately reduced for body temperature monitoring. In multi-scale wavelet decomposition processing, the selection of wavelet basis functions considers the waveform characteristics of specific physiological signals. For electrocardiogram signals with obvious oscillatory characteristics, wavelet bases with good symmetry are preferred, while for transient signals such as skin conductance response, wavelet bases with better localization characteristics are more suitable.

[0109] The adaptive threshold denoising process employs a dynamic adjustment mechanism. The thresholds for each scale are determined not only based on the statistical distribution of wavelet coefficients but also through real-time optimization incorporating signal quality assessment results. When significant motion artifacts are detected, the threshold strength is automatically increased, while during periods of stable signal quality, the threshold is appropriately relaxed to preserve more physiological details. The signal reconstruction process incorporates prior knowledge of physiological characteristic frequency bands. By establishing a frequency band feature library of typical physiological rhythms, it intelligently selects the scale coefficients to be retained, prioritizing frequency bands related to respiratory rhythm and heart rate variability.

[0110] Preferably, the temporal feature extraction process employs a multi-granularity analysis method, which captures transient changes in physiological parameters through short-term window analysis while maintaining long-term trend characteristics. The selection of feature parameters considers the needs of different application scenarios; for stress assessment, the focus is on sympathetic nerve activity indicators, while for sleep quality analysis, more attention is paid to feature parameters dominated by the parasympathetic nervous system.

[0111] This embodiment achieves real-time processing through an edge computing architecture, providing high-quality physiological feature representations for subsequent behavioral analysis while ensuring privacy and security. Through a dynamic optimization mechanism based on prior physiological features, it adapts to the differences in physiological characteristics among different users, improving the accuracy and reliability of feature extraction.

[0112] In some embodiments, extracting behavioral node features and mental state markers from a standardized behavioral feature sequence includes:

[0113] High-information-content node detection processing is performed on standardized behavioral feature sequences to identify behavioral node features. The high-information-content node detection processing is configured to perform detection based on the significance level of behavioral feature changes.

[0114] The identified behavioral node features are subjected to psychological state association analysis to extract psychological state markers. The psychological state association analysis is configured to perform feature matching based on a preset psychological-behavioral mapping relationship model.

[0115] Temporal continuity verification is performed on behavioral node features and psychological state markers to ensure the temporal consistency of feature extraction. The temporal continuity verification is configured to perform state transition probability analysis based on a Markov chain model.

[0116] The verified behavioral node features and psychological state markers are processed by feature encoding to generate structured features.

[0117] In this embodiment, the high-information-content node detection process employs a multimodal fusion saliency evaluation strategy. This strategy not only analyzes the statistical changes of individual behavioral features but also comprehensively considers the collaborative change patterns of multi-dimensional features. Specifically, when asynchronous changes in motion intensity and interaction frequency are detected, the system automatically increases the saliency threshold to avoid false detections. Conversely, for periods of synchronous feature changes, the threshold is appropriately lowered to capture subtle behavioral shifts. This dynamic adjustment mechanism allows node detection to adapt to differences in user behavior while maintaining sensitivity to key behavioral changes.

[0118] The psychological state association analysis process employs a hierarchical matching mechanism. First, coarse-grained screening determines the broad categories of psychological states. Then, specific markers are matched based on fine-grained features. This process introduces an attention mechanism to dynamically weight the mapping relationship model, assigning higher weights to psychological state dimensions highly correlated with the current behavioral node, thereby improving the accuracy of psychological state marker extraction.

[0119] The temporal continuity verification process further optimizes the application of Markov chain models. It not only analyzes the transition probabilities of adjacent nodes, but also establishes a long-term dependency model and captures the medium- and long-term evolution patterns of behavior by introducing a sliding window mechanism.

[0120] Feature encoding employs an adaptive encoding strategy, automatically adjusting the encoding granularity based on the temporal density of behavioral node features and psychological state markers. Fine-grained encoding is used in feature-dense regions to preserve details, while coarse-grained encoding is used in sparse regions to improve processing efficiency. Furthermore, semantic similarity metrics can be introduced during the encoding process to map nodes that are close in distance in the behavioral feature space to adjacent regions in the encoding space.

[0121] This embodiment accurately captures key changes in behavioral features through a dynamically adjusted saliency detection mechanism, and improves the extraction accuracy of psychological state markers by combining hierarchical attention matching. Temporal continuity verification employs a sliding window mechanism to analyze the evolution of behavioral patterns, ensuring temporal consistency in feature extraction. An adaptive encoding strategy automatically adjusts the encoding granularity based on feature distribution and maintains the topological structure of the feature space through semantic similarity. This scheme significantly improves the representational ability of behavioral node features and psychological state markers, providing a structured feature representation with temporal relevance and semantic consistency for psychological state analysis, enabling subsequent analysis processes to more accurately reflect the correlation between user behavior patterns and psychological states.

[0122] In some embodiments, a three-dimensional association matrix is ​​constructed using a dynamic knowledge graph by combining behavioral node features, psychological state markers, and temporal physiological feature vectors, including:

[0123] The structured features are encoded using behavioral pattern dimensions to generate behavioral pattern feature vectors.

[0124] Physiological response dimension encoding is performed on the time-series physiological feature vector to generate physiological response feature vector;

[0125] A dynamic correlation is established between the behavioral pattern dimension, the time dimension, and the physiological response dimension. The time dimension features are directly obtained by extracting timestamp information from the standardized behavioral feature sequence and the temporal physiological feature vector. The dynamic correlation is configured to perform multi-dimensional feature interaction learning based on a graph attention network.

[0126] The behavioral pattern feature vector, physiological response feature vector, and their dynamic correlation are represented in a matrix to generate a three-dimensional correlation matrix.

[0127] In this embodiment, the behavioral pattern dimension encoding of structured features can be understood as a feature space transformation process for structured behavioral features. This process is implemented through a deep feature extraction network, mapping the original behavioral node features to a low-dimensional vector space with semantic consistency. The physiological response dimension encoding process converts the temporal physiological feature vectors into a representation that matches the behavioral pattern features. This process uses a temporal convolutional network to capture multi-scale features of physiological signals. The time dimension features are obtained directly from the timestamp information extracted from the standardized behavioral feature sequence and the temporal physiological feature vector, ensuring strict alignment of each dimension feature on the time axis.

[0128] Dynamic association relationships are achieved through multi-dimensional feature interaction learning via a graph attention network. This network is configured to adaptively calculate the association weights between the behavioral pattern dimension, the temporal dimension, and the physiological response dimension. The network learns feature interaction patterns in different semantic spaces in parallel through a multi-head attention mechanism, ultimately fusing the outputs of each attention head to form a stable association relationship representation. The generation of the three-dimensional association matrix employs tensor concatenation and compression techniques to integrate behavioral pattern feature vectors, physiological response feature vectors, and their dynamic association relationships into a matrix structure with clear physical meaning.

[0129] This embodiment achieves a unified representation of behavioral features, psychological states, and physiological responses by constructing a three-dimensional correlation matrix. Specifically, the behavioral pattern dimension encoding preserves the semantic information of the original behavioral sequence, the physiological response dimension encoding extracts key features of physiological signals, and the establishment of dynamic correlations reveals the potential connections between cross-modal features. This embodiment automatically learns the complex mapping relationships between each dimension using deep learning technology, providing a structured multimodal feature representation foundation for subsequent analysis.

[0130] In some embodiments, the pre-diagnosis and grading results are obtained by calculating the potential risk probability of the three-dimensional correlation matrix based on a graph neural network model, including:

[0131] The three-dimensional correlation matrix is ​​subjected to graph structure modeling to construct a dynamic heterogeneous graph network. The graph structure modeling is configured to model the behavioral pattern dimension, time dimension and physiological response dimension as different types of nodes respectively.

[0132] Multi-hop neighbor feature aggregation is performed on dynamic heterogeneous graph networks to generate node-enhanced feature representations. The multi-hop neighbor feature aggregation is configured to use a gated attention mechanism to achieve cross-dimensional feature propagation.

[0133] Risk probability prediction processing is performed on the node augmentation feature representation to calculate the potential risk probability of each mental illness category. The risk probability prediction processing is configured to be implemented as a multi-label classifier based on the sigmoid activation function.

[0134] The potential risk probability is graded and thresholded to generate a pre-diagnosis grading result. The grading threshold judgment is configured to set multi-level risk thresholds based on clinical diagnostic criteria. The pre-diagnosis grading result is then subjected to interpretable correlation processing with the original three-dimensional correlation matrix to generate a risk factor contribution analysis report. The interpretable correlation processing is configured to be implemented using a graph attention weight backtracking method.

[0135] In this embodiment, the construction of a dynamic heterogeneous graph network can be understood as modeling the behavioral pattern dimension, time dimension, and physiological response dimension in the three-dimensional association matrix as graph nodes with different types of features. This processing is achieved through heterogeneous network embedding technology, where the connection relationship between nodes of different dimensions is determined by the feature correlation in the original three-dimensional association matrix. Multi-hop neighbor feature aggregation processing is used to achieve cross-dimensional feature propagation, and a gated attention mechanism is used to dynamically adjust the information transmission intensity of neighbor nodes with different hop counts, ensuring effective information interaction between distant nodes.

[0136] Risk probability prediction is implemented using a multi-label classifier based on the sigmoid activation function, which simultaneously outputs the potential risk probabilities of multiple mental illness categories. Preferably, the multi-level risk thresholds used in the grading threshold judgment process are set according to clinical diagnostic criteria, with specific threshold divisions comprehensively considering factors such as symptom severity and intervention urgency. Interpretable association processing is achieved through a graph attention weight backtracking method. Specifically, this method traces the contribution of each dimension node to the final pre-diagnosis grading result along the feature propagation path, generating a risk factor contribution analysis report with causal relationships.

[0137] This embodiment achieves deep fusion analysis of multi-dimensional features by constructing a dynamic heterogeneous graph network. The gated attention mechanism effectively solves the information decay problem in long-distance dependency modeling of traditional graph neural networks, and the multi-label classifier adapts to the clinical characteristics of comorbidity in mental illnesses. Risk factor contribution analysis not only provides pre-diagnosis results but also reveals the specific influence paths of various behavioral patterns and physiological indicators on diagnostic conclusions, providing interpretable decision-making basis for subsequent intervention measures. This embodiment achieves end-to-end mapping from multi-dimensional features to clinical diagnosis while maintaining computational efficiency.

[0138] In some embodiments, a progressive protocol is used to cross-modal fuse pre-diagnosis and grading results with the user's objective behavioral environment to generate personalized intervention strategies, including:

[0139] The pre-diagnosis grading results are processed for risk level analysis to generate pre-diagnosis risk level intervals. The risk level analysis process is configured to perform interval matching based on preset multi-level risk thresholds.

[0140] The objective behavioral environment of the user is processed to extract environmental features and generate an environmental feature vector. The environmental feature extraction process is configured to extract contextual features from user device usage records and application logs.

[0141] The pre-diagnostic risk level intervals and environmental feature vectors are aligned across modally to establish a risk-environment association mapping. The cross-modal alignment process is configured to use a cross-modal attention mechanism to achieve feature space alignment.

[0142] Intervention strategy generation is performed based on risk-environment correlation mapping, outputting personalized intervention strategies. This intervention strategy generation process is configured to use differentiated strategy generation rules based on risk levels.

[0143] For low-risk levels, a fine-tuning strategy for the environment should be adopted;

[0144] A digital support strategy will be adopted for medium-risk areas;

[0145] For high-risk levels, a strategy of deception involving manual intervention is employed;

[0146] The generated personalized intervention strategy is subjected to feasibility verification processing to ensure its executability. The feasibility verification processing is configured to conduct strategy simulation tests based on users' historical behavior data.

[0147] Personalized intervention strategies are encoded to generate an executable sequence of intervention instructions. The strategy encoding process is configured to encapsulate strategy parameters and execution conditions in JSON format.

[0148] In this embodiment, risk level analysis refers to the process of mapping the pre-diagnosis grading results to preset risk level intervals. This process is implemented through an interval matching algorithm. The multi-level risk thresholds are set according to clinical psychology standards, discretizing continuous risk probabilities into clinically significant level classifications. Preferably, the environmental feature extraction process uses temporal feature encoding technology to capture the spatiotemporal patterns of user behavior environment. The generated environmental feature vector includes features such as device usage frequency, application switching patterns, and interaction time distribution.

[0149] Cross-modal alignment processing aligns the pre-diagnostic risk level range with the feature space of the environmental feature vector through a cross-modal attention mechanism. It establishes a risk-environment correlation mapping by automatically learning the potential correlation patterns between risk levels and environmental features. Intervention strategy generation processing employs differentiated strategy generation rules based on risk levels. Preferred strategies include: environmental fine-tuning strategy, which achieves imperceptible intervention by adjusting interface elements and notification frequency; digital companionship strategy, which introduces virtual companionship roles to provide emotional support; and human intervention camouflage strategy, which disguises professional intervention as system-recommended content. Furthermore, feasibility verification processing constructs a strategy simulation environment based on user historical behavior data and evaluates the success rate of strategy execution through a behavior prediction model.

[0150] The strategy encoding process converts personalized intervention strategies into an executable sequence of intervention instructions. It uses JSON format to encapsulate elements such as strategy type, triggering conditions, and execution parameters, ensuring the portability of the strategy across different terminal devices.

[0151] This embodiment achieves seamless integration of psychological risk pre-diagnosis and actual intervention through a progressive protocol. The cross-modal fusion technology solves the problem of traditional intervention strategies being out of touch with the environment, while the hierarchical strategy ensures a precise match between intervention intensity and risk level. While maintaining a natural user experience, it achieves a closed-loop process from risk identification to precise intervention.

[0152] In some embodiments, the method further includes:

[0153] Based on the antifragility feedback mechanism, the personalized intervention strategy is dynamically adjusted. When the user's psychological resilience is detected to be improved, negative weight decay is automatically triggered, and an adaptive execution plan is generated that includes the intensity of digital companionship, environmental adjustment gradient, and artificial intervention disguise strategy.

[0154] In this embodiment, the antifragility feedback mechanism is an adaptive system that dynamically adjusts parameters by monitoring the user's response to the intervention strategy. This mechanism quantifies the user's adaptability in real time through a psychological resilience assessment model. Preferably, the psychological resilience assessment model achieves real-time quantification of the user's adaptability through multimodal data fusion, including: collecting user interaction behavior data (such as application usage time and response speed) and physiological indicator data (such as heart rate variability); extracting features through a temporal neural network and constructing a dynamic assessment matrix by combining the pre-diagnosis grading results. The model uses a sliding window mechanism to process the real-time data stream and outputs a psychological resilience index in the 0-1 range. An increase in the index triggers a negative weight decay mechanism. The assessment process comprehensively considers features such as behavioral stability and physiological recovery speed. Negative weight decay refers to the process by which the system automatically reduces the intervention intensity parameter when an increase in the user's psychological resilience is detected. This processing is implemented through a gradient descent algorithm to ensure that the intervention intensity and the user's state remain dynamically balanced.

[0155] Digital companionship intensity characterizes the interaction frequency and emotional support level of virtual companionship roles; environmental adjustment gradient reflects the gradual extent of interface adjustments and environmental optimization; and artificial intervention camouflage strategy refers to the degree of natural presentation of professional intervention content. Adaptive implementation plans are generated through multi-parameter collaborative optimization, comprehensively considering current intervention effects and user acceptance. The detection of improved psychological resilience is based on a comprehensive judgment of changes in user behavior patterns and improvements in physiological indicators, employing a sliding window statistical method to identify sustained positive trends.

[0156] This embodiment introduces an antifragile feedback mechanism, enabling the intervention strategy to self-evolve. The system not only responds to the user's immediate state but also identifies long-term changes in adaptability, achieving a paradigm shift from passive intervention to proactive growth. The dynamic parameter adjustment process maximizes user autonomy while ensuring the intervention's effectiveness, forming a virtuous cycle of psychological development.

[0157] In some embodiments, dynamically tuning personalized intervention strategies based on an antifragility feedback mechanism includes:

[0158] Collect user response behavior characteristics, and perform psychological resilience assessment on the response behavior characteristics to generate resilience improvement indicators. The psychological resilience assessment includes multi-dimensional analysis based on behavioral pattern change characteristics and physiological parameter stability characteristics.

[0159] The current personalized intervention strategy is obtained and its utility is evaluated to generate a strategy utility score. The utility evaluation process is configured to combine short-term effect indicators and long-term trend indicators for comprehensive calculation.

[0160] Based on the resilience enhancement index and strategy utility score, the strategy parameters are optimized, and tuning instructions are generated. The strategy parameter optimization is configured to be implemented using a dynamic weight adjustment algorithm.

[0161] The parameter tuning instructions are updated to personalize the intervention strategy, generating an adaptive execution plan.

[0162] In this embodiment, response behavior characteristics are a set of user interaction behavior data generated in response to the current intervention strategy, including dimensional features such as operation response time, function usage frequency, and emotional feedback intensity, collected through smart terminal sensors and interaction logs. Psychological resilience assessment processing refers to the process of feature extraction and quantitative analysis of the raw behavioral data. Specifically, behavioral pattern change characteristics are identified using a Markov chain model to determine the stable transitions in behavioral sequences, while physiological parameter stability characteristics are calculated based on biosignals such as skin conductance response and heart rate variability to assess autonomic nervous system regulation.

[0163] The utility assessment module is an analytical module that quantifies the effects of ongoing personalized intervention strategies. Short-term effect indicators reflect the immediate magnitude of behavioral change and can be calculated using a before-and-after test comparison method. Long-term trend indicators assess the sustained intervention effect, employing a time series prediction model to analyze the cumulative effect of behavioral improvement. The strategy utility score integrates these indicators using a weighted fusion algorithm, with weighting coefficients dynamically adjusted according to the intervention stage.

[0164] The strategy parameter optimization process is a decision-making process that generates control instructions based on resilience enhancement indicators and strategy utility scores. The dynamic weight adjustment algorithm adopts a bi-objective optimization framework to minimize the user's psychological burden while ensuring the intervention effect. The parameter tuning instructions include the frequency correction value of digital companionship intensity, the step size parameter of environmental adjustment gradient, and the trigger threshold of artificial intervention camouflage strategy. These parameters are obtained by constrained optimization algorithm within a preset feasible region.

[0165] The process of generating adaptive implementation plans embodies the core characteristics of antifragile systems. Specifically, when resilience enhancement indicators show increased user adaptability, the system automatically reduces the intensity of intervention and shifts decision-making weight to the user's autonomy dimension. This mechanism maintains a dynamic balance between intervention pressure and psychological resilience through the principle of negative feedback regulation, avoiding over-reliance on manual intervention while preventing maladaptive effects caused by premature withdrawal of support, and ensuring that the intervention strategy always maintains an optimal match with the user's psychological state.

[0166] Please see Figure 3 In a second aspect, this embodiment also provides a mental illness pre-diagnosis information processing system 1, applicable to the mental illness pre-diagnosis information processing method described in the first aspect. The system includes: a signal acquisition unit 11, a signal processing unit 12, and a pre-diagnosis analysis unit 13. The signal acquisition unit 11 is used to acquire multimodal behavioral signals and physiological parameter signals of users. The multimodal behavioral signals include keyboard interaction dynamics features, screen touch trajectory features, and environmental acoustic marker features. The signal processing unit 12 is used to perform a first preprocessing on the multimodal behavioral signals to obtain a standardized behavioral feature sequence. The first preprocessing is configured as differential privacy protection processing based on edge computing. The system also performs a second preprocessing on the physiological parameter signals to obtain a temporal physiological feature vector. The second preprocessing is configured as noise separation processing based on wavelet transform. The pre-diagnosis analysis unit 13... Unit 13 is used to extract behavioral node features and psychological state markers from standardized behavioral feature sequences. It then constructs a three-dimensional association matrix using a dynamic knowledge graph, integrating the behavioral node features, psychological state markers, and temporal physiological feature vectors. The dimensions of the three-dimensional association matrix include behavioral pattern dimension, time dimension, and physiological response dimension. Based on a graph neural network model, it calculates the potential risk probability of the three-dimensional association matrix to obtain a pre-diagnosis and grading result. A progressive protocol is used to cross-modally fuse the pre-diagnosis and grading result with the user's objective behavioral environment to generate a personalized intervention strategy. The pre-diagnosis and grading result is configured as a multi-dimensional output including risk level, development trajectory, and vulnerability factors. Finally, based on the behavioral node features, the pre-diagnosis and grading result, and the personalized intervention strategy, it generates a comprehensive pre-diagnosis report including an individual psychological profile, environmental adaptability assessment, and intervention effect prediction.

[0167] In this embodiment, the signal acquisition unit 11 synchronously acquires the user's multimodal behavioral signals (including keyboard interaction dynamics, screen touch trajectory features and environmental acoustic marker features) and physiological parameter signals (such as heart rate variability and skin conductance response) through a high-precision sensor module, and ensures the temporal synchronization of data acquisition.

[0168] The signal processing unit 12 performs edge computing-based differential privacy protection processing on the multimodal behavioral signals at the edge computing node to generate a standardized behavioral feature sequence, and performs wavelet transform-based noise separation processing on the physiological parameter signals to output a time-series physiological feature vector.

[0169] The pre-diagnosis analysis unit 13 first extracts behavioral node features and psychological state markers from the standardized behavioral feature sequence. Then, it constructs a three-dimensional association matrix (behavioral pattern dimension × time dimension × physiological response dimension) with the time-series physiological feature vector through a dynamic knowledge graph. It then uses a graph neural network model to calculate the probability of potential risks to generate a pre-diagnosis classification result that includes risk level, development trajectory, and vulnerability factors. This unit further adopts a progressive protocol to cross-modally integrate the pre-diagnosis classification result with the user's objective behavioral environment to generate personalized intervention strategies. Finally, it integrates behavioral node features, pre-diagnosis classification results, and intervention strategies to form a comprehensive pre-diagnosis report that includes individual psychological profiles, environmental adaptability assessment, and intervention effect prediction.

[0170] The mental illness pre-diagnosis information processing system 1 described in this embodiment executes the mental illness pre-diagnosis information processing method described in the first aspect. For details, please refer to the foregoing embodiment, which will not be repeated here.

[0171] By adopting the above technical solutions, this invention differs from existing technologies and possesses the following beneficial effects: It resolves the contradiction between privacy protection and data availability in the pre-diagnosis of mental illness through the collaborative processing of multimodal behavioral signals and physiological parameter signals. Edge computing-based differential privacy protection processing locally adds noise and extracts features from keyboard interaction dynamics, screen touch trajectory features, and environmental acoustic markers on the user terminal. A dynamically adjusted privacy budget ensures that the original behavioral data remains within the device while the generated standardized behavioral feature sequence retains the continuity of behavioral patterns. Wavelet transform noise separation processing extracts temporal physiological feature vectors reflecting the activity characteristics of the autonomic nervous system from the physiological parameter signals. A dynamic knowledge graph constructs a three-dimensional association matrix from behavioral node features, mental state markers, and temporal physiological feature vectors, and models cross-dimensional feature interaction relationships through a graph attention network. A progressive protocol employs a three-level processing approach—time alignment, spatial association, and environmental fusion—to perform cross-modal matching between the pre-diagnosis grading results and the objective behavioral environment, generating personalized intervention plans that include environmental fine-tuning strategies, digital companionship strategies, and artificial intervention camouflage strategies. The antifragility feedback mechanism dynamically adjusts strategy parameters based on users' psychological resilience indicators, forming an adaptive intervention loop. This technical solution, while meeting strict privacy requirements, achieves dynamic correlation analysis between behavioral characteristics and physiological responses, providing a multimodal fusion solution for the early identification of mental illnesses.

[0172] Finally, it should be noted that although the above embodiments have been described in the text and drawings of this application, this should not limit the scope of patent protection of this application. Any technical solutions that are based on the essential concept of this application and utilize the content described in the text and drawings of this application, resulting in equivalent structural or procedural substitutions or modifications, as well as the direct or indirect application of the technical solutions of the above embodiments to other related technical fields, are all included within the scope of patent protection of this application.

Claims

1. A method for processing information for the pre-diagnosis of mental illness, characterized in that, include: The system collects multimodal behavioral signals and physiological parameter signals from users. The multimodal behavioral signals include keyboard interaction dynamics, screen touch trajectory features, and environmental acoustic marker features. The physiological parameter signals are indirectly acquired through built-in sensors in the device. The multimodal behavioral signal is subjected to a first preprocessing to obtain a standardized behavioral feature sequence. The first preprocessing is configured as differential privacy protection processing based on edge computing. The physiological parameter signal is subjected to a second preprocessing to obtain a temporal physiological feature vector. The second preprocessing is configured as noise separation processing based on wavelet transform. Extracting behavioral node features and psychological state markers from the standardized behavioral feature sequence includes: The standardized behavioral feature sequence is subjected to high-information node detection processing to identify behavioral node features. The high-information node detection processing is configured to perform detection based on the significance level of behavioral feature changes. The identified behavioral node features are subjected to psychological state association analysis to extract psychological state markers. The psychological state association analysis is configured to perform feature matching based on a preset psychological-behavior mapping relationship model. The behavioral node features and psychological state markers are subjected to temporal continuity verification processing to ensure the temporal consistency of feature extraction. The temporal continuity verification processing is configured to perform state transition probability analysis based on a Markov chain model. The verified behavioral node features and psychological state markers are subjected to feature encoding to generate structured features; The behavioral node features, psychological state markers, and temporal physiological feature vectors are used to construct a three-dimensional association matrix through a dynamic knowledge graph. The dimensions of this three-dimensional association matrix include behavioral pattern dimension, time dimension, and physiological response dimension, including: The structured features are encoded using behavioral pattern dimensions to generate behavioral pattern feature vectors; The time-series physiological feature vectors are subjected to physiological response dimension encoding to generate physiological response feature vectors; A dynamic correlation is established between the behavioral pattern dimension, the time dimension, and the physiological response dimension. The dynamic correlation is configured to perform multi-dimensional feature interaction learning based on a graph attention network. The behavioral pattern feature vector, physiological response feature vector and their dynamic correlation relationship are matrix-represented to generate the three-dimensional correlation matrix. The potential risk probability is calculated based on the graph neural network model to obtain the pre-diagnosis and classification results. The pre-diagnosis and classification results are then fused with the user's objective behavioral environment across modalities using a progressive protocol to generate personalized intervention strategies. The pre-diagnosis and classification results are configured as a multi-dimensional output including risk level, development trajectory, and vulnerability factors. Based on the behavioral node characteristics, pre-diagnosis grading results, and personalized intervention strategies, a comprehensive pre-diagnosis report is generated, which includes an individual psychological profile, environmental adaptability assessment, and intervention effect prediction.

2. The method for processing information for the pre-diagnosis of mental illness according to claim 1, characterized in that, The multimodal behavioral signal undergoes a first preprocessing step to obtain a standardized behavioral feature sequence. This first preprocessing step is configured as edge-computation-based differential privacy protection processing, including: The keystroke interval analysis and pressure intensity analysis are performed on the keyboard interaction dynamics characteristics to generate a first row of feature sub-vectors. The keystroke interval analysis includes calculating the standard deviation of the time difference between adjacent keystrokes, and the pressure intensity analysis includes statistically analyzing the kurtosis characteristics of the pressure values. The screen touch trajectory features are subjected to sliding trajectory curvature analysis and touch point distribution analysis to generate a second row of feature sub-vectors. The sliding trajectory curvature analysis uses cubic spline interpolation to calculate the average curvature, and the touch point distribution analysis includes calculating the spatial autocorrelation index of touch point density. The environmental acoustic marker features are subjected to voiceprint entropy analysis and silence duration analysis to generate a third row of feature sub-vectors. The voiceprint entropy analysis uses Mel frequency cepstral coefficients to calculate spectral entropy, and the silence duration analysis includes statistically analyzing the duration distribution of silence segments per unit time. The first behavior feature sub-vector, the second behavior feature sub-vector, and the third behavior feature sub-vector are subjected to feature fusion processing based on attention weights to obtain the intermediate behavior feature vector. The intermediate vector of the behavioral features is subjected to noise addition processing based on the Laplacian mechanism. The noise scale parameter of the noise addition processing is dynamically adjusted according to the feature sensitivity to generate a noisy feature vector that meets the preset differential privacy requirements. The noisy feature vector is subjected to Z-score normalization transformation to obtain the normalized behavioral feature sequence.

3. The method for processing information for the pre-diagnosis of mental illness according to claim 1, characterized in that, The physiological parameter signal undergoes a second preprocessing step to obtain a time-series physiological feature vector. This second preprocessing step is configured as a noise separation process based on wavelet transform, and includes: The physiological parameter signal is subjected to multi-scale wavelet decomposition to generate a set of wavelet coefficients containing approximation coefficients and detail coefficients. The multi-scale wavelet decomposition is configured to be performed using orthogonal wavelet basis functions with compact support characteristics. Adaptive threshold denoising is performed on the wavelet coefficient set to obtain denoised wavelet coefficients. The adaptive threshold denoising process includes threshold functions at various scales, and the threshold functions at various scales are configured to be determined based on the statistical distribution characteristics of the wavelet coefficients. The denoised wavelet coefficients are subjected to signal reconstruction processing to obtain a denoised parameter signal. The signal reconstruction processing is configured to filter and retain scale coefficients that reflect physiological rhythm characteristics. The denoised parameter signal is subjected to time-domain feature extraction processing to generate the time-series physiological feature vector. The time-domain feature extraction processing includes calculating time-varying feature parameters that reflect the activity characteristics of the autonomic nervous system.

4. The method for processing information for the pre-diagnosis of mental illness according to claim 1, characterized in that, The pre-diagnosis and classification results are obtained by calculating the potential risk probability of the three-dimensional correlation matrix based on a graph neural network model, including: The three-dimensional correlation matrix is ​​subjected to graph structure modeling to construct a dynamic heterogeneous graph network. The graph structure modeling is configured to model the behavioral pattern dimension, time dimension and physiological response dimension as different types of nodes, respectively. The dynamic heterogeneous graph network is subjected to multi-hop neighbor feature aggregation processing to generate node enhanced feature representations. The multi-hop neighbor feature aggregation processing is configured to use a gated attention mechanism to achieve cross-dimensional feature propagation. The node enhanced feature representation is subjected to risk probability prediction processing to calculate the potential risk probability of each mental illness category. The risk probability prediction processing is configured to be implemented by a multi-label classifier based on the sigmoid activation function. The potential risk probability is subjected to a grading threshold judgment process to generate a pre-diagnosis grading result. The grading threshold judgment process is configured to set multiple risk thresholds based on clinical diagnostic criteria.

5. The method for processing information for the pre-diagnosis of mental illness according to claim 4, characterized in that, A progressive protocol is used to perform cross-modal fusion of the pre-diagnosis and grading results with the user's objective behavioral environment to generate personalized intervention strategies, including: The pre-diagnosis grading results are subjected to risk level analysis processing to generate pre-diagnosis risk level intervals. The risk level analysis processing is configured to perform interval matching based on preset multi-level risk thresholds. The objective behavioral environment of the user is subjected to environmental feature extraction processing to generate an environmental feature vector. The environmental feature extraction processing is configured to extract contextual features from user device usage records and application logs. The pre-diagnosed risk level intervals are aligned with environmental feature vectors across modalities to establish a risk-environment association mapping. The cross-modal alignment process is configured to use a cross-modal attention mechanism to achieve feature space alignment. Based on the risk-environment correlation mapping, intervention strategies are generated and personalized intervention strategies are output. Personalized intervention strategies are encoded to generate an executable sequence of intervention instructions. The strategy encoding process is configured to encapsulate strategy parameters and execution conditions in JSON format.

6. The method for processing information for the pre-diagnosis of mental illness according to claim 1, characterized in that, The method further includes: The personalized intervention strategy is dynamically adjusted based on the antifragility feedback mechanism. When an increase in the user's psychological resilience is detected, a negative weight decay is automatically triggered, and an adaptive execution plan is generated that includes the intensity of digital companionship, the gradient of environmental adjustment, and the artificial intervention disguise strategy.

7. The method for processing information for the pre-diagnosis of mental illness according to claim 6, characterized in that, The personalized intervention strategy is dynamically tuned based on the antifragility feedback mechanism, including: Collect user response behavior characteristics, and perform psychological resilience assessment processing on the response behavior characteristics to generate resilience improvement indicators. The psychological resilience assessment processing includes multi-dimensional analysis based on behavioral pattern change characteristics and physiological parameter stability characteristics. The current personalized intervention strategy is obtained and its utility is evaluated to generate a strategy utility score. The utility evaluation is configured to combine short-term effect indicators and long-term trend indicators for comprehensive calculation. Based on the resilience enhancement index and strategy utility score, the strategy parameters are optimized to generate tuning instructions. The strategy parameter optimization is configured to be implemented using a dynamic weight adjustment algorithm. The parameter tuning instructions are used to update the personalized intervention strategy, thereby generating the adaptive execution plan.

8. A mental illness pre-diagnosis information processing system, characterized in that, The mental illness pre-diagnosis information processing method according to any one of claims 1 to 7, the system comprising: The signal acquisition unit is used to acquire the user's multimodal behavioral signals and physiological parameter signals, wherein the multimodal behavioral signals include keyboard interaction dynamics features, screen touch trajectory features and environmental acoustic marker features; The signal processing unit is configured to perform a first preprocessing on the multimodal behavioral signal to obtain a standardized behavioral feature sequence, wherein the first preprocessing is configured as differential privacy protection processing based on edge computing, and to perform a second preprocessing on the physiological parameter signal to obtain a temporal physiological feature vector, wherein the second preprocessing is configured as noise separation processing based on wavelet transform. The pre-diagnosis analysis unit is used to extract behavioral node features and psychological state markers from the standardized behavioral feature sequence. It then constructs a three-dimensional association matrix using a dynamic knowledge graph, combining the behavioral node features, psychological state markers, and temporal physiological feature vectors. The dimensions of this three-dimensional association matrix include behavioral pattern dimensions, time dimensions, and physiological response dimensions. Based on a graph neural network model, it calculates the potential risk probability of the three-dimensional association matrix to obtain a pre-diagnosis grading result. A progressive protocol is used to cross-modally fuse the pre-diagnosis grading result with the user's objective behavioral environment to generate a personalized intervention strategy. The pre-diagnosis grading result is configured as a multi-dimensional output including risk level, development trajectory, and vulnerability factors. Finally, based on the behavioral node features, the pre-diagnosis grading result, and the personalized intervention strategy, it generates a comprehensive pre-diagnosis report including an individual psychological profile, environmental adaptability assessment, and intervention effect prediction.

Images