A mental health intelligent assessment method and system based on machine learning

Abstract

The application discloses a kind of mental health intelligent evaluation method and system based on machine learning, it is related to the technical field of mental health assessment, the method includes: multi-modal data acquisition, cognitive bias field modeling, psychological state counterfactual simulation and mental health risk assessment;Cognitive bias field data reflecting the change of psychological state relative to historical baseline is constructed to depict the deviation intensity, continuity and change trend of psychological state;On this basis, by path constraint disturbance to key influencing factor, counterfactual input sample is constructed, and the evolution model of psychological state is combined to deduce, the psychological state change result and factor contribution degree under different hypothesis conditions are obtained;The application can realize the dynamic modeling and multidimensional analysis of mental health state, improve the accuracy and explanation ability of psychological risk identification, and provide data support for personalized intervention.

Description

Technical Field

[0001] This invention relates to the field of mental health assessment technology, specifically to a machine learning-based intelligent mental health assessment method and system. Background Technology

[0002] Machine learning-based intelligent assessment of mental health refers to a technical solution that uses multimodal data fusion technology and temporal modeling methods to dynamically analyze and determine the risk of users' mental state. It constructs a cognitive bias field to depict the changes in mental state relative to a historical baseline, and combines a counterfactual inference mechanism to analyze the degree of influence of key factors on mental state, thereby achieving a comprehensive assessment and trend prediction of mental health risk. This method can not only reflect the current risk level of mental state, but also reveal the causes of risk formation and future direction of change, and has important application value in scenarios such as mobile terminal mental health management, remote medical assisted assessment, and personalized psychological intervention.

[0003] However, in practical applications, the existing technology still has the following shortcomings:

[0004] First, most methods analyze data based on a single data source or a single time segment, making it difficult to integrate multimodal information such as text expression, speech features, behavioral patterns, and physiological rhythms, resulting in biased evaluation results.

[0005] Second, the lack of modeling of the dynamic changes in psychological state relative to an individual's historical baseline makes it difficult to distinguish between short-term fluctuations and persistent abnormalities, which can easily lead to misjudgment or missed judgment.

[0006] Third, existing methods mostly remain at the level of correlation analysis or classification prediction, lacking in-depth analysis of the mechanism of action of key influencing factors, making it difficult to explain the source of risk and predict future trends, and unable to effectively support personalized intervention decisions.

[0007] Therefore, there is an urgent need for an intelligent assessment method and system for mental health that can integrate multimodal data, characterize dynamic changes in psychological state, and possess risk interpretation and trend prediction capabilities. Summary of the Invention

[0008] In view of the above situation and to overcome the shortcomings of the existing technology, the technical solution adopted by the present invention is as follows: The present invention provides a machine learning-based intelligent assessment method for mental health, which includes the following steps:

[0009] Step S1: Multimodal data acquisition;

[0010] Step S2: Cognitive offset field modeling;

[0011] Step S3: Counterfactual simulation of mental state;

[0012] Step S4: Mental health risk assessment.

[0013] Further, in step S1, the multimodal data acquisition is used to obtain multi-source heterogeneous data related to mental health status, and to uniformly process the data to form multimodal psychological representation data for subsequent modeling; specifically, it involves acquiring multimodal information including text interaction data, voice expression data, behavioral interaction data, and physiological rhythm data, and performing time alignment processing, data format unification processing, and outlier filtering processing on the acquired data to obtain multimodal psychological representation data with consistent structure;

[0014] The multimodal psychological representation data specifically includes: text semantic feature data, speech prosody feature data, behavioral pattern feature data, and physiological rhythm feature data.

[0015] Further, in step S2, the cognitive offset field modeling is used to construct a dynamic offset structure that reflects the changes in psychological state relative to the historical baseline based on multimodal psychological representation data, so as to characterize the changing trend and abnormal fluctuation degree of psychological state; specifically, the multimodal psychological representation data is organized in the form of time series, psychological feature vectors of each time node are constructed, and the offset at the current moment is calculated based on the feature statistics results within the historical time window to obtain cognitive offset field data reflecting changes in psychological state;

[0016] The offset vector at the current moment can be obtained based on the mean characteristics of the historical window, which is used to characterize the change of the current psychological state relative to the historical baseline. On this basis, by calculating the amplitude and analyzing the trend of change of the offset vector, cognitive offset field data containing offset intensity information and trend information is constructed.

[0017] The cognitive offset field modeling employs a cognitive offset field construction method based on temporal baseline constraints and multi-scale offset evolution. This method is used to construct a dynamic offset structure reflecting changes in psychological states relative to historical baselines based on multimodal psychological representation data, thereby characterizing the changing trends and abnormal fluctuations of psychological states. Specifically, it includes the following steps:

[0018] Step S21: Construction of temporal psychological feature vectors, specifically, organizing the multimodal psychological representation data in chronological order, constructing a sequence of psychological feature vectors of a unified dimension at each time node to form temporal psychological feature data that can be used for subsequent offset calculation, and obtaining temporal psychological feature vector data;

[0019] Step S22: Baseline constraint offset vector calculation, specifically, statistical processing of the time-series psychological feature vector data based on a preset historical time window, calculation of the historical baseline features of each time node, and difference calculation between the current psychological feature vector and the corresponding historical baseline features to obtain offset vector data reflecting the change of psychological state relative to the historical level.

[0020] Step S23: Quantitative calculation of offset intensity, specifically, the amplitude of the offset vector data is calculated, and the offsets of different dimensions are weighted based on preset feature weights to obtain offset intensity data used to characterize the degree of deviation of the psychological state from the historical baseline.

[0021] Step S24: Modeling the offset continuity feature, specifically, based on the offset intensity data within a continuous time window, statistically analyzing the offset state at each time node, and determining the continuous offset behavior based on whether the offset intensity exceeds a preset threshold, to obtain offset continuity data reflecting the continuity of offset changes.

[0022] Step S25: Multi-scale migration trend analysis, specifically, performing differential calculation or rate of change analysis on the migration intensity data based on historical windows of different time scales to obtain the migration change trend at each time scale, and performing weighted fusion processing on the multi-scale trends to obtain migration trend data reflecting the direction and speed of psychological state evolution.

[0023] Step S26: Construction of cognitive offset field, specifically, the structured integration of the offset vector data, offset intensity data, offset continuity data and offset trend data to construct cognitive offset field data, which is used to characterize the overall change characteristics of psychological state in the time dimension.

[0024] The cognitive offset field data specifically includes: offset vector data, offset intensity data, offset continuity data, and offset trend data;

[0025] The offset vector data is used to characterize the directional features of changes in psychological state;

[0026] The offset intensity data is used to characterize the degree to which the psychological state deviates from the historical baseline;

[0027] The offset continuity data is used to characterize the persistence of offset changes;

[0028] The offset trend data is used to characterize the evolutionary trend of changes in psychological state.

[0029] Further, in step S3, the counterfactual simulation of psychological state is used to construct the results of changes in psychological state under different hypothetical conditions based on cognitive offset field data, and to analyze the degree of influence of key influencing factors on mental health risks. Specifically, it involves constructing a psychological state evolution model based on the cognitive offset field data, selecting key feature variables that have an impact on psychological state, and performing conditional adjustment processing on the key feature variables to obtain input data under different hypothetical conditions. By constructing corresponding counterfactual input samples and inputting the counterfactual input samples into the psychological state evolution model for deduction, the results of changes in psychological state under different conditions are obtained. By comparing and analyzing the original state results and the counterfactual state results, the degree of influence of each key feature variable on the changes in psychological state can be obtained, thereby obtaining the counterfactual simulation result data of psychological state.

[0030] The counterfactual simulation of psychological states employs a counterfactual inference method based on path constraint perturbation and factor contribution decomposition. It is used to construct the changes in psychological states under different hypothetical conditions based on cognitive bias field data, and to analyze the extent to which key influencing factors affect mental health risks. Specifically, it includes the following steps:

[0031] Step S31: Key Influencing Factor Identification, used to determine the key feature variables participating in the counterfactual simulation. Specifically, based on the offset intensity information, offset continuity information, and offset trend information in the cognitive offset field data, each dimension of the psychological feature vector is screened, and feature dimensions with offset amplitude exceeding a preset threshold, continuous offset degree exceeding a preset threshold, or significant impact on the overall offset trend are selected as key influencing factors to obtain key influencing factor set data.

[0032] Step S32: Constructing path-constrained counterfactual inputs to generate counterfactual samples that conform to the laws of psychological state change. Specifically, the key influencing factors are conditionally adjusted, and the disturbance amplitude or replacement range of each key influencing factor is restricted according to the historical sample distribution range, feature upper and lower limit constraints, and multi-feature synergistic change relationship. Counterfactual input samples under different assumptions are constructed to obtain counterfactual input sample data.

[0033] Step S33: Psychological state evolution deduction, used to simulate the psychological state change results under different assumptions, specifically, inputting the counterfactual input sample data into the pre-constructed psychological state evolution model, and combining it with the cognitive offset field data corresponding to the current moment, deducing and calculating the psychological state of each counterfactual sample under the corresponding assumptions to obtain counterfactual state change result data;

[0034] Step S34: Factor effect quantitative analysis, used to calculate the degree of influence of key influencing factors on changes in psychological state. Specifically, by performing a difference analysis between the original state results and the counterfactual state change results, and combining the disturbance or replacement amplitude of each key influencing factor, the contribution and response of each key influencing factor to changes in psychological state are calculated, and the contribution data of key influencing factors and the sensitivity data of changes in psychological state are obtained.

[0035] Step S35: Risk change trend prediction, used to extrapolate the subsequent change direction of mental health risk based on multiple sets of counterfactual simulation results. Specifically, it involves statistically analyzing the counterfactual state change results under different assumptions, and combining the contribution of key influencing factors and the sensitivity of mental state changes to extrapolate the changes in mental health risk in the future time range, thereby obtaining risk change trend prediction data.

[0036] The counterfactual simulation results of the psychological state specifically include: contribution data of key influencing factors, counterfactual state change results data, psychological state change sensitivity data, and risk change trend prediction data.

[0037] Further, in step S4, the mental health risk assessment is used to comprehensively determine mental health risks based on cognitive bias field data and counterfactual simulation results of mental states; specifically, it involves comprehensively analyzing mental health status based on the bias intensity and trend information in the cognitive bias field data, and the contribution of key influencing factors and risk change trends in the counterfactual simulation results of mental states, to obtain mental health risk assessment results; the mental health risk assessment results specifically include: mental health risk level data, risk change trend data, and risk source analysis data.

[0038] The present invention provides a machine learning-based intelligent assessment system for mental health, comprising a multimodal data acquisition module, a cognitive bias field modeling module, a counterfactual simulation module for mental states, and a mental health risk assessment module;

[0039] The multimodal data acquisition module is used for multimodal data acquisition. Through multimodal data acquisition, multimodal mental representation data is obtained, and the multimodal mental representation data is sent to the cognitive offset field modeling module.

[0040] The cognitive offset field modeling module is used for cognitive offset field modeling. Through cognitive offset field modeling, cognitive offset field data is obtained, and the cognitive offset field data is sent to the psychological state counterfactual simulation module and the mental health risk assessment module.

[0041] The counterfactual simulation module is used for counterfactual simulation of psychological states. Through counterfactual simulation, it obtains counterfactual simulation result data and sends the counterfactual simulation result data to the mental health risk assessment module.

[0042] The mental health risk assessment module is used for mental health risk assessment, and through the mental health risk assessment, the results of the mental health risk assessment are obtained.

[0043] The beneficial effects achieved by adopting the above solution are as follows:

[0044] (1) In view of the fact that existing mental health assessment methods rely solely on a single scale, single question and answer results or isolated behavioral records to make state judgments, it is difficult to dynamically depict the continuous changes in the user's mental state and it is difficult to simultaneously take into account multiple sources of information such as text expression, voice fluctuations, behavioral abnormalities and physiological rhythm changes. As a result, the assessment results are prone to remain at a static and fragmented level. In mobile terminal mental health application scenarios, it is particularly prone to technical problems such as delayed risk identification, unclear sources of abnormalities and insufficient stability of comprehensive judgment. This solution creatively adopts an overall technical architecture that combines multimodal data collection, cognitive bias field modeling, psychological state counterfactual simulation and mental health risk assessment. It integrates text interaction data, voice expression data, behavioral interaction data and physiological rhythm data into the same assessment process and realizes comprehensive analysis of mental health risks based on historical baselines and dynamic change results.

[0045] (2) In view of the technical problems in the existing methods for analyzing changes in psychological state, such as the lack of a stable reference baseline for the temporal changes of psychological characteristics, the difficulty in distinguishing between short-term random fluctuations and continuous abnormal shifts, and the lack of a unified modeling mechanism for the direction and speed of change at different time scales, which makes the identification of abnormal psychological state easily affected by instantaneous noise interference and the trend judgment inaccurate, this solution creatively adopts a cognitive shift field construction method based on temporal baseline constraints and multi-scale shift evolution. Through the construction of temporal psychological feature vectors, calculation of baseline constraint shift vectors, quantitative calculation of shift intensity, modeling of shift continuity features, multi-scale shift trend analysis and cognitive shift field construction, the degree of deviation, continuous features and evolution trend of psychological state relative to historical level are expressed in a structured way.

[0046] (3) In view of the technical problem that existing methods for predicting mental health risks can only provide the current risk level or correlation analysis results, and it is difficult to further explain the specific path of key influencing factors on changes in mental state under different assumptions, thus failing to effectively support the tracing of risk sources, the judgment of intervention priorities, and the prediction of future risk trends, this solution creatively adopts a counterfactual inference method for mental state based on path constraint perturbation and factor contribution decomposition. Through key influencing factor identification, path constraint counterfactual input construction, mental state evolution inference, factor effect quantitative analysis, and risk change trend prediction, under the premise of conforming to the historical sample distribution pattern and multi-feature synergistic change constraints, the solution adjusts the conditions of key variables such as sleep duration, speech rate, text emotional tendency, and interaction frequency, and then analyzes the degree of contribution and sensitivity of different factors to mental health risks. Attached Figure Description

[0047] Figure 1 A flowchart illustrating a machine learning-based intelligent assessment method for mental health provided by this invention;

[0048] Figure 2 A schematic diagram of a machine learning-based intelligent mental health assessment system provided by the present invention;

[0049] Figure 3 A flowchart illustrating the cognitive offset field modeling process for step S2;

[0050] Figure 4 This is a flowchart illustrating the counterfactual simulation of the mental state in step S3.

[0051] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation

[0052] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0053] In the description of this invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0054] Example 1, see Figure 1 This invention provides a machine learning-based intelligent assessment method for mental health, which includes the following steps:

[0055] Step S1: Multimodal data acquisition;

[0056] Step S2: Cognitive offset field modeling;

[0057] Step S3: Counterfactual simulation of mental state;

[0058] Step S4: Mental health risk assessment.

[0059] By performing the above operations, this solution addresses the shortcomings of existing mental health assessment methods. These methods rely solely on a single scale, a single question-and-answer result, or isolated behavioral records for state judgment, making it difficult to dynamically depict the continuous changes in a user's mental state. Furthermore, they struggle to simultaneously consider multi-source information such as text expression, voice fluctuations, behavioral abnormalities, and physiological rhythm changes, resulting in assessment results that tend to remain static and fragmented. In mobile terminal mental health application scenarios, this approach is particularly prone to technical problems such as delayed risk identification, unclear sources of abnormalities, and insufficient stability in comprehensive judgment. This solution creatively adopts an overall technical architecture that combines multimodal data acquisition, cognitive bias field modeling, counterfactual simulation of mental state, and mental health risk assessment. It integrates text interaction data, voice expression data, behavioral interaction data, and physiological rhythm data into the same assessment process and achieves comprehensive analysis of mental health risks based on historical baselines and dynamic changes.

[0060] For example, if a user exhibits increased negative semantics in text, decreased speech rate, increased nighttime usage time, and disrupted sleep cycles in a short period, traditional methods may only provide localized abnormality alerts based on a single indicator. However, this solution can identify the risk accumulation process from the perspective of multimodal joint changes and continuous temporal evolution, thereby improving the comprehensiveness, timeliness, and reliability of mental health assessments.

[0061] Example 2, see Figure 1 and Figure 2This embodiment is based on the above embodiment. In step S1, the multimodal data acquisition is used to acquire multi-source heterogeneous data related to mental health status, and to perform unified processing on the data to form multimodal psychological representation data for subsequent modeling. Specifically, it involves acquiring multimodal information including text interaction data, voice expression data, behavioral interaction data, and physiological rhythm data, and performing time alignment processing, data format unification processing, and outlier filtering processing on the acquired data to obtain multimodal psychological representation data with consistent structure.

[0062] In this embodiment, a mental health assessment application based on a mobile terminal is used as the application scenario. The text interaction data can come from the user's daily text input content in the application. The voice expression data can be collected by the user's voice input signal through the microphone module of the terminal device. The behavioral interaction data can be obtained by recording behavioral information such as user operation frequency, usage duration and interaction interval. The physiological rhythm data can be obtained through wearable devices, including heart rate change data and sleep cycle data.

[0063] During data processing, when performing time alignment on the multimodal data, data from different sources can be synchronously matched based on a unified timestamp, and the data can be segmented within a preset time window (e.g., 5 minutes, 15 minutes, or 1 hour) to ensure the consistency of the multimodal data in the time dimension.

[0064] During the data format unification process, data from different modalities can be converted into a unified vector representation. For text data, a preset word vector model can be used for encoding. For speech data, prosodic features such as speech rate, pitch, and energy distribution can be extracted. For behavioral data, frequency and interval features can be obtained through statistical methods. For circadian rhythm data, statistical indicators such as mean heart rate, fluctuation amplitude, and sleep duration can be extracted.

[0065] During the outlier filtering process, abnormal data can be removed or corrected based on a preset threshold range or sliding window statistical results. For example, when heart rate data is detected to be outside the normal physiological range or behavioral data shows abnormal mutations, the corresponding data can be marked or filtered to improve the stability of subsequent modeling data.

[0066] The multimodal psychological representation data specifically includes: text semantic feature data, speech prosody feature data, behavioral pattern feature data, and physiological rhythm feature data.

[0067] Example 3, see Figure 1 , Figure 2 and Figure 3This embodiment is based on the above embodiment. In step S2, the cognitive offset field modeling is used to construct a dynamic offset structure that reflects the change of psychological state relative to the historical baseline based on multimodal psychological representation data, so as to characterize the changing trend and abnormal fluctuation degree of psychological state; specifically, the multimodal psychological representation data is organized in the form of time series, psychological feature vectors of each time node are constructed, and the offset at the current moment is calculated according to the feature statistics results within the historical time window to obtain cognitive offset field data reflecting the change of psychological state;

[0068] In this embodiment, the offset vector at the current moment can be obtained based on the mean characteristics of the historical window, which is used to characterize the change of the current psychological state relative to the historical baseline. On this basis, by performing amplitude calculation and trend analysis on the offset vector, cognitive offset field data containing offset intensity information and trend information is constructed.

[0069] The cognitive offset field modeling employs a cognitive offset field construction method based on temporal baseline constraints and multi-scale offset evolution. This method is used to construct a dynamic offset structure reflecting changes in psychological states relative to historical baselines based on multimodal psychological representation data, thereby characterizing the changing trends and abnormal fluctuations of psychological states. Specifically, it includes the following steps:

[0070] Step S21: Construction of temporal psychological feature vectors, specifically, organizing the multimodal psychological representation data in chronological order, constructing a sequence of psychological feature vectors of a unified dimension at each time node to form temporal psychological feature data that can be used for subsequent offset calculation, and obtaining temporal psychological feature vector data;

[0071] In this embodiment, the psychological feature vector at each time point can be composed of a combination of text semantic features, speech prosody features, behavioral pattern features and physiological rhythm features, and is represented by a unified feature dimension.

[0072] Step S22: Baseline constraint offset vector calculation, specifically, statistical processing of the time-series psychological feature vector data based on a preset historical time window, calculation of the historical baseline features of each time node, and difference calculation between the current psychological feature vector and the corresponding historical baseline features to obtain offset vector data reflecting the change of psychological state relative to the historical level.

[0073] In this embodiment, the historical baseline features can be obtained by calculating the mean of the feature vectors within the historical window to ensure that the offset calculation results have a stable reference benchmark.

[0074] Furthermore, to avoid distortion in offset determination due to differences in historical fluctuation amplitudes across different psychological feature dimensions, this embodiment introduces an adaptive constraint mechanism based on historical stability into the offset vector data. The offsets of each dimension are normalized to obtain stable constraint offset vector data, calculated using the following formula:

[0075] ;

[0076] In the formula, It is the offset vector data of the stability constraint at time t corresponding to the k-th dimension feature. It is temporal psychological feature vector data. It is a historical baseline characteristic. It is the fluctuation scale of the k-th dimension feature within the historical time window. It is a tiny constant that prevents the denominator from being zero;

[0077] Step S23: Quantitative calculation of offset intensity, specifically, the amplitude of the offset vector data is calculated, and the offsets of different dimensions are weighted based on preset feature weights to obtain offset intensity data used to characterize the degree of deviation of the psychological state from the historical baseline.

[0078] In this embodiment, different weights are assigned to each feature dimension to reflect the degree of influence of different modal features on changes in psychological state, thereby improving the accuracy of the offset intensity calculation. The offset intensity is based on the offset vector data of the stability constraint, constructing an offset response term, which serves as the basic input for subsequent dynamic potential field modeling. The calculation formula is as follows:

[0079] ;

[0080] In the formula, This is offset intensity data, used to represent the instantaneous offset response intensity after stabilization constraints. The feature weights corresponding to the k-th dimension feature satisfy the following conditions: , K is the total number of feature dimensions of the psychological feature vector;

[0081] Step S24: Modeling the offset continuity feature, specifically, based on the offset intensity data within a continuous time window, statistically analyzing the offset state at each time node, and determining the continuous offset behavior based on whether the offset intensity exceeds a preset threshold, to obtain offset continuity data reflecting the continuity of offset changes.

[0082] In this embodiment, by accumulating and statistically analyzing the offset states at multiple consecutive time points, short-term fluctuations and persistent abnormal changes are distinguished, thereby improving the ability to identify abnormal changes in psychological states. The calculation formula for the offset continuity data is as follows:

[0083] ;

[0084] In the formula, It is the offset memory state quantity at time t, used as the offset continuity data. It is the memory retention coefficient, with a preferred value of 0.70 to 0.90;

[0085] Step S25: Multi-scale migration trend analysis, specifically, performing differential calculation or rate of change analysis on the migration intensity data based on historical windows of different time scales to obtain the migration change trend at each time scale, and performing weighted fusion processing on the multi-scale trends to obtain migration trend data reflecting the direction and speed of psychological state evolution.

[0086] In this embodiment, multiple time windows (short-term, medium-term, and long-term) can be set to analyze the offset changes, thereby improving the stability and robustness of trend determination. The calculation formula for setting multiple time windows (short-term, medium-term, and long-term) to analyze the offset changes is as follows:

[0087] ;

[0088] In the formula, It is a comprehensive trend traction quantity, used as offset trend data. This is the short-term weighting coefficient, with a preferred value of 0.50. It is the short-term offset rate of change. This is the mid-term weighting coefficient, with a preferred value of 0.30. It is the rate of change of mid-term offset. This is the long-term weighting coefficient, with a preferred value of 0.20. It is the long-term offset change rate;

[0089] Based on this, the instantaneous offset response term, offset memory state quantity, and trend traction quantity are uniformly coupled to construct the cognitive offset potential field strength, and the offset potential field strength is obtained. The calculation formula is as follows:

[0090] ;

[0091] In the formula, It is the intensity of the offset potential field. This is the instantaneous response weight, with a preferred value of 0.45. This is the memory coupling weight, with a preferred value of 0.35. This is the trend-driven weight, with a preferred value of 0.20;

[0092] Step S26: Construction of cognitive offset field, specifically, the structured integration of the offset vector data, offset intensity data, offset continuity data and offset trend data to construct cognitive offset field data, which is used to characterize the overall change characteristics of psychological state in the time dimension.

[0093] The cognitive offset field data specifically includes: offset vector data, offset intensity data, offset continuity data, and offset trend data;

[0094] The offset vector data is used to characterize the directional features of changes in psychological state;

[0095] The offset intensity data is used to characterize the degree to which the psychological state deviates from the historical baseline;

[0096] The offset continuity data is used to characterize the persistence of offset changes;

[0097] The offset trend data is used to characterize the evolutionary trend of changes in psychological state;

[0098] Furthermore, in some implementations, the offset intensity can be dynamically updated based on the time series rate of change to obtain the cognitive offset distribution results that evolve over time, forming the basic input data for subsequent counterfactual simulations.

[0099] By performing the above operations, this solution addresses the technical problems in existing methods for analyzing changes in psychological states. These problems include the lack of a stable reference baseline for the temporal changes of psychological characteristics, difficulty in distinguishing between short-term random fluctuations and persistent abnormal shifts, and the lack of a unified modeling mechanism for the direction and speed of change at different time scales. This leads to the identification of abnormal psychological states being easily affected by instantaneous noise interference and inaccurate trend judgment. This solution creatively adopts a cognitive offset field construction method based on temporal baseline constraints and multi-scale offset evolution. Through the construction of temporal psychological feature vectors, calculation of baseline constraint offset vectors, quantitative calculation of offset intensity, modeling of offset continuity features, multi-scale offset trend analysis, and construction of cognitive offset fields, the solution can structurally express the degree of deviation, persistence characteristics, and evolutionary trend of psychological states relative to historical levels.

[0100] For example, in mobile terminal applications, when a user exhibits a decrease in interaction frequency, weakened voice energy, and abnormally shortened sleep duration over multiple consecutive time windows, traditional methods may treat these as several discrete anomalies. However, this solution can combine the shift changes under short-term, medium-term, and long-term windows to identify that the user has changed from occasional fluctuations to a continuous shift in psychological state, thereby improving the accuracy and robustness of anomaly identification.

[0101] Example 4, see Figure 1 , Figure 2 and Figure 4This embodiment is based on the above embodiment. In step S3, the counterfactual simulation of the psychological state is used to construct the results of the change of the psychological state under different assumptions based on the cognitive offset field data, and to analyze the degree of influence of key influencing factors on mental health risks. Specifically, it constructs a psychological state evolution model based on the cognitive offset field data, selects key feature variables that have an impact on the psychological state, and performs condition adjustment processing on the key feature variables to obtain input data under different assumptions.

[0102] In this embodiment, by perturbing or replacing key feature variables, corresponding counterfactual input samples are constructed, and these counterfactual input samples are input into a psychological state evolution model for deduction to obtain psychological state change results under different conditions. By comparing and analyzing the original state results with the counterfactual state results, the degree of influence of each key feature variable on psychological state changes can be obtained, thereby obtaining psychological state counterfactual simulation result data.

[0103] Preferably, the psychological state evolution model is a time-series prediction model trained through supervised learning based on historical multimodal psychological representation data and corresponding psychological state labels, used to characterize the mapping relationship between multimodal features and psychological state changes; in one embodiment, the psychological state evolution model is used to jointly model the multimodal psychological feature sequence and the cognitive bias potential field, thereby characterizing the dynamic relationship of psychological state evolution over time; the model structure of the psychological state evolution model is not limited to a specific type, and can be implemented by selecting long short-term memory networks, gated recurrent units, or other models with time-series modeling capabilities according to actual application needs;

[0104] The counterfactual simulation of psychological states employs a counterfactual inference method based on path constraint perturbation and factor contribution decomposition. It is used to construct the changes in psychological states under different hypothetical conditions based on cognitive bias field data, and to analyze the extent to which key influencing factors affect mental health risks. Specifically, it includes the following steps:

[0105] Step S31: Key Influencing Factor Identification, used to determine the key feature variables participating in the counterfactual simulation. Specifically, based on the offset intensity information, offset continuity information, and offset trend information in the cognitive offset field data, each dimension of the psychological feature vector is screened, and feature dimensions with offset amplitude exceeding a preset threshold, continuous offset degree exceeding a preset threshold, or significant impact on the overall offset trend are selected as key influencing factors to obtain key influencing factor set data.

[0106] In this embodiment, the abnormal change dimensions in text semantic features, speech prosody features, behavioral pattern features and physiological rhythm features can be prioritized by combining the offset intensity ranking results and the trend change rate within the time window, so as to improve the pertinence of subsequent counterfactual simulations.

[0107] More preferably, a state-dependent activation mechanism based on a cognitive offset potential field is introduced for key influencing factors. These factors only participate in subsequent perturbation propagation when the corresponding feature has a significant effect in the current offset field. Specifically, let the offset potential field strength obtained in step S2 be... The corresponding stable constraint offsets for each feature dimension are: Then, the factor activation decision function is constructed, and the calculation formula is:

[0108] ;

[0109] In the formula, It is the indicator function value for screening key impact factors, if and only if At that time, the k-th dimension feature is included in the set of key influencing factors. It is a preset threshold for feature selection, used to characterize feature dimensions that have a significant impact on the overall offset state;

[0110] Step S32: Constructing path-constrained counterfactual inputs to generate counterfactual samples that conform to the laws of psychological state change. Specifically, the key influencing factors are conditionally adjusted, and the disturbance amplitude or replacement range of each key influencing factor is restricted according to the historical sample distribution range, feature upper and lower limit constraints, and multi-feature synergistic change relationship. Counterfactual input samples under different assumptions are constructed to obtain counterfactual input sample data.

[0111] In this embodiment, the condition adjustment process may include adding or subtracting perturbations to a single key influencing factor, or combining and replacing multiple key influencing factors; for features such as sleep duration, speech rate and interaction frequency, counterfactual input samples may be generated based on a preset ratio of fluctuation above and below the historical average, so as to avoid generating abnormal samples that significantly deviate from the actual psychological state change pattern.

[0112] In this embodiment, the counterfactual perturbation is propagated in a restricted manner along the historical evolution path of the psychological state to ensure that the counterfactual samples conform to the dynamic constraints in the real psychological change process. Specifically, let the perturbation applied to the key influencing factor corresponding to the k-th dimension feature at time t be... Then, the formula for calculating the propagation state representation of the disturbance in subsequent time steps is:

[0113] ;

[0114] In the formula, It is a propagation state representation of counterfactual perturbations, used as counterfactual input samples under different assumptions. This is the disturbance attenuation coefficient, preferably with a value of 0.20~0.50. This is a modulation function based on the offset potential field strength, used to describe the amplification or suppression effect of the current mental state on the propagation of disturbances. i is the time step of the disturbance propagation. In a typical implementation, when the mental state is in a high offset potential field (i.e., ... When the potential field is large, the rate of decay of the disturbance slows down in subsequent time steps, thus simulating the process of the continuous amplification of negative effects under abnormal psychological conditions; while under low offset potential field conditions, the disturbance decays rapidly to avoid generating counterfactual trajectories that do not conform to the normal psychological fluctuation pattern.

[0115] Step S33: Psychological state evolution deduction, used to simulate the psychological state change results under different assumptions, specifically, inputting the counterfactual input sample data into the pre-constructed psychological state evolution model, and combining it with the cognitive offset field data corresponding to the current moment, deducing and calculating the psychological state of each counterfactual sample under the corresponding assumptions to obtain counterfactual state change result data;

[0116] In this embodiment, the psychological state evolution model can be trained based on historical time series data to characterize the evolutionary relationship of psychological states as they change with multimodal features.

[0117] More preferably, when performing counterfactual inference, the psychological state evolution model not only uses multimodal features as input, but also incorporates the offset potential field intensity. As a state modulation variable introduced into the model calculation process, the calculation formula for the counterfactual state change result data is as follows:

[0118] ;

[0119] In the formula, It is a psychological state risk score output by the psychological state evolution model. It is an evolutionary model mapping function, which can be constructed using a neural network model based on a temporal recursive structure in its specific implementation.

[0120] Step S34: Factor effect quantitative analysis, used to calculate the degree of influence of key influencing factors on changes in psychological state. Specifically, by performing a difference analysis between the original state results and the counterfactual state change results, and combining the disturbance or replacement amplitude of each key influencing factor, the contribution and response of each key influencing factor to changes in psychological state are calculated, and the contribution data of key influencing factors and the sensitivity data of changes in psychological state are obtained.

[0121] In this embodiment, the degree of influence of different characteristic variables can be quantified by adjusting each key influencing factor one by one and recording the output changes, thereby achieving a detailed analysis of the sources of psychological risk.

[0122] Specifically, let the counterfactual perturbation of the key influencing factor corresponding to the k-th dimension feature produce the output change at each time step as follows: The formula for calculating the contribution data of the key influencing factors is as follows:

[0123] ;

[0124] In the formula, This is data on the contribution of key impact factors. It represents the output change at each time step caused by the counterfactual perturbation of the key influencing factor corresponding to the k-th feature. In other words, it represents the change in psychological state relative to the original output after applying a perturbation to the key influencing factor corresponding to the k-th feature. It is the duration of the counterfactual perturbation propagation, where i is the perturbation propagation time step. It is the time weighting coefficient;

[0125] Let the perturbation amplitude of the key influencing factor corresponding to the k-th dimension feature be... The formula for calculating the sensitivity data to changes in psychological state is as follows:

[0126] ;

[0127] In the formula, It is data on sensitivity to changes in psychological state. It is the perturbation amplitude of the key influencing factor corresponding to the k-th dimension feature;

[0128] Step S35: Risk change trend prediction, used to extrapolate the subsequent change direction of mental health risk based on multiple sets of counterfactual simulation results. Specifically, it involves statistically analyzing the counterfactual state change results under different assumptions, and combining the contribution of key influencing factors and the sensitivity of mental state changes to extrapolate the changes in mental health risk in the future time range, thereby obtaining risk change trend prediction data.

[0129] In this embodiment, the counterfactual simulation constructs multiple counterfactual evolution paths for different perturbation conditions. Each path corresponds to a complete set of psychological state change processes. Specifically, let the input sequence corresponding to the j-th counterfactual path be... Then, by using a psychological state evolution model, the output sequence of this path at each time step can be obtained. Based on this, the overall output of the path is summarized to obtain the corresponding counterfactual path output value. The specific calculation formula is as follows:

[0130] ;

[0131] In the formula, It is the predicted psychological state result corresponding to the j-th counterfactual path, where i is the perturbation propagation time step. It is a time-weighted coefficient. It is the predicted output of the mental state of the j-th counterfactual path;

[0132] In this embodiment, different weights can be assigned to multiple sets of counterfactual simulation results according to their temporal proximity or the reasonableness of the perturbation, so as to improve the stability of the risk change trend prediction results. The calculation formula for the risk change trend prediction data is as follows:

[0133] ;

[0134] In the formula, This is risk change trend prediction data, and j is the counterfactual path index. It is a counterfactual path weight. This is the predicted psychological state result corresponding to the j-th counterfactual path;

[0135] The counterfactual simulation results of the psychological state specifically include: contribution data of key influencing factors, counterfactual state change results data, psychological state change sensitivity data, and risk change trend prediction data;

[0136] Furthermore, in some implementations, statistical analysis can be performed on the psychological state change path based on multiple sets of counterfactual simulation results to identify the combination of key factors that have a significant impact on mental health risks.

[0137] By performing the above operations, this solution addresses the technical problem that existing mental health risk prediction methods can only provide the current risk level or correlation analysis results, making it difficult to further explain the specific path of key influencing factors on changes in mental state under different assumptions. As a result, these methods cannot effectively support risk source tracing, intervention priority judgment, and future risk trend prediction. This solution creatively adopts a counterfactual inference method for mental state based on path constraint perturbation and factor contribution decomposition. Through key influencing factor identification, path constraint counterfactual input construction, mental state evolution inference, factor effect quantitative analysis, and risk change trend prediction, under the premise of conforming to the historical sample distribution pattern and multi-feature synergistic change constraints, it adjusts the conditions of key variables such as sleep duration, speech rate, text emotional tendency, and interaction frequency, and then analyzes the degree of contribution and sensitivity of different factors to mental health risk.

[0138] For example, when the system identifies that a user's recent risk has increased mainly due to sleep cycle disorder and a sharp drop in interaction activity, this solution can construct counterfactual samples such as "sleep duration recovers to near the historical average" or "interaction frequency returns to the normal range" to deduce the changes in psychological state under the corresponding conditions, and quantify the driving role of different factors on risk changes, providing a more targeted basis for subsequent psychological risk intervention.

[0139] Example 5, see Figure 1 , Figure 2 This embodiment is based on the above embodiment. In step S4, the mental health risk assessment is used to comprehensively determine mental health risk based on cognitive bias field data and counterfactual simulation results of mental state. Specifically, it involves comprehensively analyzing the mental health status based on the bias intensity and trend information in the cognitive bias field data, and the contribution of key influencing factors and risk change trends in the counterfactual simulation results of mental state, to obtain a mental health risk assessment result. The mental health risk assessment result specifically includes: mental risk level data, risk change trend data, and risk source analysis data.

[0140] The change trend information is determined based on the offset intensity data, offset trend data, and offset potential field intensity.

[0141] In one implementation, a weighted risk scoring function can be constructed to weight and fuse the offset intensity, trend traction, and contribution of key influencing factors, and psychological risk levels can be divided according to preset thresholds.

[0142] Let the offset intensity data obtained in step S2 be denoted as Offset continuity data is The offset trend data is The offset potential field strength is Let the contribution data of the key influencing factors corresponding to the k-th dimension feature obtained in step S3 be denoted as . The data on sensitivity to changes in psychological state are Risk change trend prediction data is ;

[0143] To avoid the influence of different data units on the overall evaluation results, we first... , R , Normalization was performed separately to obtain the normalized results. , , , , ;

[0144] Furthermore, to highlight the impact of the upward trend of psychological risk on the overall judgment result, a risk upward trend enhancement term is constructed, and the calculation formula is as follows:

[0145] ;

[0146] In the formula, It is a risk-increasing trend enhancement term, used to characterize the portion of the offset trend data pointing in the direction of increasing risk; when the normalized offset trend data When the current offset trend does not show an increasing risk direction, the risk increase trend enhancement term is set to 0.

[0147] Furthermore, based on the contribution data of key influencing factors and the sensitivity data to changes in psychological state, the source fusion weights of the key influencing factors are constructed, and the calculation formula is as follows:

[0148] ;

[0149] In the formula, It is the source fusion weight of the key influencing factor corresponding to the k-th dimension feature. It is the set of key influencing factors selected at time t. These are the contribution data of key influencing factors corresponding to the k-th dimension feature after normalization. These are the sensitivity data of key influencing factors corresponding to the k-th dimension feature after normalization. It is a tiny constant that prevents the denominator from being zero;

[0150] The above settings give higher weight to key influencing factors that contribute more and are more sensitive to disturbances in the risk source analysis.

[0151] Furthermore, a comprehensive mental health risk scoring function is constructed, and the calculation formula is as follows:

[0152] ;

[0153] In the formula, It is the comprehensive mental health risk score at time t; It is the Sigmoid normalization function, used to map the comprehensive risk score to a range of 0 to 1; It is the normalized offset potential field strength, used to characterize the overall degree of deviation of the current psychological state relative to the historical baseline; It is an amplifying factor in the upward trend of risk; It is normalized risk change trend prediction data, used to characterize the direction of subsequent risk changes obtained from counterfactual simulation; It is the fusion contribution of key influencing factors, used to characterize the comprehensive effect of key influencing factors on changes in psychological state; , , , For risk assessment fusion weights, and satisfying ;

[0154] In a preferred embodiment, , , , The values ​​can be 0.40, 0.20, 0.25, and 0.15 respectively; among which, This is used to enhance the fundamental role of the current cognitive bias field in risk level determination. Used to enhance the impact of an upward trend in risk assessment. Used to incorporate the results of future risk changes obtained from counterfactual simulations. Used to introduce the degree of contribution of key influencing factors to changes in psychological state;

[0155] In practical applications, the above weights can be adjusted according to the specific mental health management scenario; for example, when the application scenario focuses more on short-term risk warnings, the weights can be increased. and The value of can be improved when the application scenario focuses more on explaining the causes of risk. The value of ; further, based on the aforementioned comprehensive mental health risk score Generate psychological risk level data;

[0156] In one implementation, when At that time, the psychological risk level was determined to be low risk;

[0157] when At that time, the psychological risk level was determined to be medium risk;

[0158] when At that time, the psychological risk level was determined to be high risk;

[0159] The risk change trend data is based on the comprehensive mental health risk score within a continuous time window. The rate of change and the predicted risk trend data R are determined; when multiple consecutive time windows... When the risk change trend prediction data R shows an upward trend and exceeds a preset trend threshold, it is determined that the mental health risk is on an upward trend.

[0160] The risk source analysis data is then weighted according to the sources of each key influencing factor. and key impact factor contribution data It is determined that the text semantic features, phonetic prosodic features, behavioral pattern features, or physiological rhythm features that have a significant impact on the comprehensive risk score of mental health are used to output the text semantic features, phonetic prosodic features, behavioral pattern features, or physiological rhythm features.

[0161] Example 6, see Figure 1 and Figure 2 Based on the above embodiments, this embodiment provides a machine learning-based intelligent assessment system for mental health, including a multimodal data acquisition module, a cognitive bias field modeling module, a counterfactual simulation module for mental states, and a mental health risk assessment module.

[0162] The multimodal data acquisition module is used for multimodal data acquisition. Through multimodal data acquisition, multimodal mental representation data is obtained, and the multimodal mental representation data is sent to the cognitive offset field modeling module.

[0163] The cognitive offset field modeling module is used for cognitive offset field modeling. Through cognitive offset field modeling, cognitive offset field data is obtained, and the cognitive offset field data is sent to the psychological state counterfactual simulation module and the mental health risk assessment module.

[0164] The counterfactual simulation module is used for counterfactual simulation of psychological states. Through counterfactual simulation, it obtains counterfactual simulation result data and sends the counterfactual simulation result data to the mental health risk assessment module.

[0165] The mental health risk assessment module is used for mental health risk assessment, and through the mental health risk assessment, the results of the mental health risk assessment are obtained.

[0166] It should be noted that, in this document, relational 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 such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0167] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.

[0168] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A machine learning-based intelligent assessment method for mental health, characterized in that: The method includes the following steps: Step S1: Multimodal data acquisition to obtain multimodal psychological representation data; Step S2: Cognitive offset field modeling, organizing multimodal mental representation data in a time-series format, constructing mental feature vectors for each time node, and calculating the offset at the current moment based on the feature statistics within the historical time window to obtain cognitive offset field data reflecting changes in mental state; Step S3: Counterfactual simulation of psychological state. Based on the cognitive bias field data, construct a psychological state evolution model, select key feature variables that have an impact on psychological state, and perform conditional adjustment processing on the key feature variables to obtain the counterfactual simulation result data of psychological state. Step S4: Mental health risk assessment. Based on the offset intensity and trend information in the cognitive offset field data, as well as the contribution of key influencing factors and risk change trends in the counterfactual simulation results of the mental state, a comprehensive analysis of the mental health status is conducted to obtain the mental health risk assessment results.

2. The intelligent mental health assessment method based on machine learning according to claim 1, characterized in that: In step S1, the multimodal mental representation data specifically includes: text semantic feature data, speech prosody feature data, behavioral pattern feature data, and physiological rhythm feature data.

3. The intelligent mental health assessment method based on machine learning according to claim 2, characterized in that: In step S2, the cognitive offset field modeling adopts a cognitive offset field construction method based on temporal baseline constraints and multi-scale offset evolution. This method is used to construct a dynamic offset structure that reflects the changes in psychological state relative to the historical baseline based on multimodal psychological representation data, so as to characterize the changing trend and abnormal fluctuation degree of psychological state. Specifically, it includes the following steps: construction of temporal psychological feature vector, calculation of baseline constraint offset vector, quantitative calculation of offset intensity, modeling of offset continuity features, multi-scale offset trend analysis, and construction of cognitive offset field. The construction of the temporal psychological feature vector involves organizing the multimodal psychological representation data in chronological order and constructing a sequence of psychological feature vectors of a unified dimension at each time node to obtain temporal psychological feature vector data. The baseline constraint offset vector calculation is based on the statistical processing of the time-series psychological feature vector data through a preset historical time window, calculating the historical baseline features of each time node, and performing difference calculation between the current psychological feature vector and the corresponding historical baseline features to obtain offset vector data reflecting the change of psychological state relative to the historical level.

4. The intelligent mental health assessment method based on machine learning according to claim 3, characterized in that: In step S2, the offset intensity quantization calculation involves calculating the amplitude of the offset vector data and weighting the offsets of different dimensions based on preset feature weights to obtain offset intensity data that characterizes the degree to which the psychological state deviates from the historical baseline. The offset continuity feature modeling is based on offset intensity data within a continuous time window. The offset status at each time node is statistically analyzed, and the continuous offset behavior is determined according to whether the offset intensity exceeds a preset threshold, so as to obtain offset continuity data that reflects the continuity of offset change. The multi-scale migration trend analysis is based on the historical windows of different time scales to perform differential calculation or rate of change analysis on the migration intensity data to obtain the migration change trend at each time scale. The multi-scale trend is then weighted and fused to obtain migration trend data that reflects the direction and speed of psychological state evolution. The cognitive offset field construction involves structurally integrating the offset vector data, offset intensity data, offset continuity data, and offset trend data to construct cognitive offset field data, which is used to characterize the overall change characteristics of psychological states in the time dimension.

5. The intelligent mental health assessment method based on machine learning according to claim 4, characterized in that: In step S2, the cognitive offset field data specifically includes: offset vector data, offset intensity data, offset continuity data, and offset trend data; The offset vector data is used to characterize the directional features of changes in psychological state; The offset intensity data is used to characterize the degree to which the psychological state deviates from the historical baseline; The offset continuity data is used to characterize the persistence of offset changes; The offset trend data is used to characterize the evolutionary trend of changes in psychological state.

6. The intelligent mental health assessment method based on machine learning according to claim 5, characterized in that: In step S3, the counterfactual simulation of the psychological state adopts a counterfactual inference method based on path constraint perturbation and factor contribution decomposition. It is used to construct the change results of the psychological state under different assumptions based on cognitive offset field data, and analyze the degree of influence of key influencing factors on mental health risks. Specifically, it includes the following steps: identification of key influencing factors, construction of path constraint counterfactual input, psychological state evolution inference, quantitative analysis of factor effects, and prediction of risk change trends. The identification of key influencing factors specifically involves filtering each dimension of the psychological feature vector based on the offset intensity information, offset continuity information, and offset trend information in the cognitive offset field data. Feature dimensions with offset amplitude exceeding a preset threshold, continuous offset degree exceeding a preset threshold, or significant impact on the overall offset trend are selected as key influencing factors, thereby obtaining a set of key influencing factor data. The path-constrained counterfactual input construction involves adjusting the conditions of the key influencing factors and limiting the perturbation amplitude or replacement range of each key influencing factor based on the historical sample distribution range, feature upper and lower limit constraints, and the synergistic change relationship of multiple features. This process constructs counterfactual input samples under different assumptions, resulting in counterfactual input sample data. The psychological state evolution deduction involves inputting the counterfactual input sample data into a pre-constructed psychological state evolution model, and combining it with the cognitive offset field data corresponding to the current moment to deduce and calculate the psychological state of each counterfactual sample under the corresponding assumption conditions, thereby obtaining the counterfactual state change result data. The quantitative analysis of the factor effect involves analyzing the difference between the original state results and the counterfactual state change results, and combining the disturbance or replacement amplitude of each key influencing factor to calculate the contribution and response of each key influencing factor to the change in psychological state, thereby obtaining the contribution data of key influencing factors and the sensitivity data of the change in psychological state. The risk change trend prediction involves statistically analyzing the counterfactual state changes under different assumptions, and combining the contribution of key influencing factors and the sensitivity to changes in psychological state to extrapolate the changes in mental health risks over a future time frame, thereby obtaining risk change trend prediction data. The counterfactual simulation results of the psychological state specifically include: contribution data of key influencing factors, counterfactual state change results data, psychological state change sensitivity data, and risk change trend prediction data.

7. The intelligent mental health assessment method based on machine learning according to claim 6, characterized in that: In step S4, the mental health risk assessment results specifically include: mental health risk level data, risk change trend data, and risk source analysis data.

8. A machine learning-based intelligent mental health assessment system, used to implement the machine learning-based intelligent mental health assessment method as described in any one of claims 1-7, characterized in that: It includes a multimodal data acquisition module, a cognitive offset field modeling module, a psychological state counterfactual simulation module, and a mental health risk assessment module.

9. The intelligent mental health assessment system based on machine learning according to claim 8, characterized in that: The multimodal data acquisition module is used for multimodal data acquisition. Through multimodal data acquisition, multimodal mental representation data is obtained, and the multimodal mental representation data is sent to the cognitive offset field modeling module. The cognitive offset field modeling module is used for cognitive offset field modeling. Through cognitive offset field modeling, cognitive offset field data is obtained, and the cognitive offset field data is sent to the psychological state counterfactual simulation module and the mental health risk assessment module. The counterfactual simulation module is used for counterfactual simulation of psychological states. Through counterfactual simulation, it obtains counterfactual simulation result data and sends the counterfactual simulation result data to the mental health risk assessment module. The mental health risk assessment module is used for mental health risk assessment, and through the mental health risk assessment, the results of the mental health risk assessment are obtained.

Images