Mental state data screening management method and system fused with machine learning
By employing multimodal data acquisition and machine learning methods, we simultaneously collect mental state data and configure dynamic gradients, which solves the problem of insufficient authenticity of mental test data and enables high-quality data screening and analysis.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHANGHAI RUNJIA BIOMEDICAL TECH CO LTD
- Filing Date
- 2025-10-30
- Publication Date
- 2026-06-19
AI Technical Summary
Existing methods for collecting mental health test data are easily affected by users' immediate physiological or psychological state, resulting in insufficient data authenticity and credibility, which affects the accuracy of subsequent analysis results.
By simultaneously acquiring multimodal data (report data, behavioral data, and physiological data) and EEG parameters, and combining machine learning methods, human brain feature sequences are segmented, mental classification gradients are configured and corrected, and abnormalities are classified and screened for management.
This improved the authenticity and reliability of mental health test data, ensured data quality, reduced data distortion caused by temporary interference, and enhanced the accuracy of analysis results.
Smart Images
Figure CN121439256B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of data processing technology, specifically to a method and system for filtering and managing mental state data that integrates machine learning. Background Technology
[0002] In the current field of mental health assessment, the methods for collecting mental health test data have become diversified, enabling the capture of users' mental health-related indicators from multiple dimensions. This has become a routine means of conducting state diagnosis, ability assessment, and efficacy tracking.
[0003] However, under the existing data collection model for mental health testing, the authenticity of the data is easily affected by the user's immediate physiological or psychological state. During the test, the user may have abnormal reactions due to unfamiliar environment, tense process, or other factors, which may lead to a deviation between the collected data and the user's true mental state, thereby reducing the credibility of the data and affecting the accuracy of subsequent analysis results and the reference value for practical application. Summary of the Invention
[0004] This application provides a method and system for screening and managing mental state data that integrates machine learning, with the aim of solving the technical problem of insufficient authenticity and credibility of mental test data in the prior art.
[0005] In view of the above problems, this application provides a method and system for filtering and managing mental state data that integrates machine learning.
[0006] Firstly, this application provides a method for filtering and managing mental state data that integrates machine learning, including:
[0007] Multiple mental test data of target users were collected through multiple testing methods, and human brain feature parameters of target users were collected during the testing process to obtain human brain feature sequences;
[0008] Based on the test time of the multiple mental test data, the human brain feature sequence is divided to obtain multiple human brain feature sub-sequences. Multiple sub-fluctuation coefficients are analyzed respectively, and multiple mental segmentation gradients are configured.
[0009] Analyze the total fluctuation coefficient of the human brain feature sequence, correct the multiple mental segmentation gradients, and obtain multiple corrected mental segmentation gradients;
[0010] The multiple modified mental gradients are used to classify multiple mental test data into anomalies, obtain a first anomaly degree, and obtain a second anomaly degree based on the total fluctuation coefficient. The anomaly degree is calculated, and the multiple mental test data are then filtered and managed.
[0011] Secondly, this application provides a mental state data screening and management system that integrates machine learning, including:
[0012] The feature acquisition module is used to collect multiple mental test data of the target user through multiple test methods, and to collect the target user's human brain feature parameters during the test to obtain the human brain feature sequence;
[0013] The gradient partitioning module is used to partition the human brain feature sequence according to the test time of the multiple mental test data, obtain multiple human brain feature sub-sequences, analyze multiple sub-fluctuation coefficients respectively, and configure multiple mental partitioning gradients.
[0014] The gradient correction module is used to analyze the total fluctuation coefficient of the human brain feature sequence, correct the multiple mental segmentation gradients, and obtain multiple corrected mental segmentation gradients.
[0015] The anomaly assessment module is used to classify multiple mental test data into anomalies using the multiple modified mental division gradients, obtain a first anomaly degree, and obtain a second anomaly degree based on the total fluctuation coefficient analysis, calculate the anomaly degree, and filter and manage the multiple mental test data.
[0016] One or more technical solutions provided in this application have at least the following technical effects or advantages:
[0017] This application provides a method and system for screening and managing mental state data that integrates machine learning. It simultaneously collects behavioral data and brain feature parameters, laying the foundation for multimodal data; it divides brain feature sequences according to test time and configures preliminary evaluation gradients to sensitively identify transient state fluctuations caused by factors such as stress; it corrects the evaluation gradient based on the overall fluctuation coefficient, making the anomaly judgment criteria more scientific and consistent; it integrates local and overall anomaly indicators to achieve precise quantitative screening of test data; and it can identify and filter out distorted data affected by temporary states, effectively improving the authenticity and credibility of mental state test data. Attached Figure Description
[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0019] Figure 1 A flowchart illustrating the mental state data screening and management method incorporating machine learning, provided in an embodiment of this application;
[0020] Figure 2 A schematic diagram of the structure of the mental state data filtering and management system that integrates machine learning provided in the embodiments of this application;
[0021] The components represented by each number in the attached diagram are explained below:
[0022] Feature acquisition module 11, gradient partitioning module 12, gradient correction module 13, anomaly assessment module 14. Detailed Implementation
[0023] This application provides a method and system for screening and managing mental state data that integrates machine learning, in order to address the technical problem of insufficient authenticity and credibility of mental test data in the prior art.
[0024] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.
[0025] It should be noted that the terms "comprising" and "having" are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or modules that are not explicitly listed or that are inherent to these processes, methods, products, or devices.
[0026] Example 1, as Figure 1 As shown, this application provides a method for filtering and managing mental state data that integrates machine learning, the method comprising:
[0027] S100: Collects multiple mental test data of target users through multiple testing methods, and collects human brain characteristic parameters of target users during the testing process to obtain human brain characteristic sequences.
[0028] In this embodiment, multiple mental health test data of the target user are collected through various testing methods, and brain feature parameters of the target user are collected during the testing process to obtain brain feature sequences. A comprehensive dataset including subjective reports, objective behavioral performance, and physiological brain activity signals is constructed through a multimodal data acquisition strategy. Single-type mental health test data is easily affected by the user's subjective deliberate control or temporary emotional fluctuations, which may lead to data distortion. By simultaneously collecting behavioral test data and physiological brain feature parameters of the target user, an objective verification benchmark can be provided for subsequent data credibility analysis, thus laying the foundation for verifying the authenticity of the data from the source.
[0029] Step S100 in the method provided in this application embodiment includes:
[0030] Multiple psychological test data were collected from the target users through various testing methods, including report data, behavioral data, and physiological data.
[0031] During the testing process, human brain feature parameters of the target user are collected to obtain human brain feature sequences, including brainwave parameters.
[0032] First, multiple mental health test data were collected from the target users through various testing methods, including report data, behavioral data, and physiological data. Corresponding report data collection scales, standardized behavioral test tasks, and physiological detection indicators were designed. All three types of data were collected simultaneously through a unified testing process to ensure that the data corresponded to the mental state under the same testing scenario.
[0033] First, collect report data from target users. Report data refers to subjective feedback data from target users regarding pre-defined mental health-related issues, reflecting users' perceptions and feelings about their own mental state. For example, develop an attention self-assessment scale, including 10 questions such as the duration of sustained attention to the task during the test, the frequency of mind wandering, and the focus rating when completing the task. Each question has a rating scale of 1 to 5 points, and collect the scale rating results filled in by users according to their actual feelings.
[0034] Secondly, collect behavioral data from the target users. Behavioral data refers to observable actions, reactions, and task completion data exhibited by the target users during the test, and has intuitive and quantifiable characteristics. For example, design a number elimination test task, presenting a matrix of numbers randomly arranged from 0 to 9 on the screen, requiring users to find and mark a specific target number 6 within a specified time. Through image acquisition equipment and task execution system, record data such as the user's marking accuracy, elimination reaction time, number of missed marks, and number of repeated marks.
[0035] In addition, physiological data of the target users is collected. Physiological data refers to objective detection data generated by the physiological functions of the target users during the testing period. It is not directly affected by subjective consciousness and has strong objectivity. For example, heart rate variability and skin conductivity data are collected in real time during the user's testing period through wearable physiological sensing devices. Heart rate variability refers to the degree of variation in the time interval between two consecutive heartbeats, which can reflect the state of autonomic nervous system function; skin conductivity is the conductivity of the skin surface, and its changes are related to the degree of emotional arousal.
[0036] Simultaneously, during the testing process, brain characteristic parameters of the target user are collected to obtain a brain characteristic sequence. These parameters include electroencephalogram (EEG) parameters. EEG parameters are quantified parameters obtained after the electrical signals generated by brain neurons during activity are collected and processed by detection equipment. Different frequencies of EEG correspond to different brain activity states. When the target user is stressed, their EEG characteristics may fluctuate, potentially introducing errors into the collected mental health test data and affecting its validity. Therefore, while collecting mental health test data using multiple testing methods, professional EEG acquisition equipment is used to continuously collect the target user's EEG parameters. The collected parameters are then arranged in an orderly manner according to the test timestamps to ensure a one-to-one correspondence between the EEG parameters and the mental health test data in the time dimension.
[0037] For example, during the aforementioned test task, the target user wears a portable EEG acquisition device. The device's electrodes are attached to areas such as the prefrontal and parietal lobes of the user's scalp to continuously detect the electrical signals generated by the activity of neurons in the brain. The device's built-in signal processing module filters, amplifies, and extracts features from the acquired raw electrical signals, obtaining EEG parameters such as alpha wave frequencies (8-13 Hz), beta wave frequencies (14-30 Hz), and theta wave frequencies (4-7 Hz). Based on the timestamps of the EEG parameter acquisition, a set of EEG parameters acquired every 100 milliseconds during the test is sequentially arranged to form a human brain feature sequence containing dynamic information about the entire test process. This human brain feature sequence is synchronized with the concurrently acquired report data, behavioral data, and physiological data in the time dimension.
[0038] In this embodiment, report data, behavioral data, and physiological data are collected simultaneously through multiple testing methods, achieving multi-dimensional coverage of mental health test data and ensuring the comprehensiveness and richness of the data. Simultaneously, electroencephalogram (EEG) parameters are collected and human brain characteristic sequences are formed, establishing a temporal correlation between mental health test data and brain neural activity data. This provides physiological support for subsequent verification of data credibility based on neural activity fluctuation analysis, laying the data foundation for the entire method and ensuring the accuracy and reliability of subsequent data processing and screening.
[0039] S200: According to the test time of the multiple mental test data, divide the human brain feature sequence to obtain multiple human brain feature sub-sequences, analyze multiple sub-fluctuation coefficients respectively, and configure multiple mental segmentation gradients.
[0040] Step S200 in the method provided in this application embodiment includes:
[0041] Based on the testing time of the multiple mental test data, the human brain feature sequence is divided to obtain human brain feature sub-sequences during the testing process of the multiple mental test data;
[0042] The fluctuation amplitude of the multiple human brain feature sub-sequences is analyzed and calculated as multiple sub-fluctuation coefficients, wherein the fluctuation amplitude is the maximum deviation between any human brain feature within the human brain feature sub-sequence and the mean of human brain features.
[0043] Multiple mental partitioning gradients are configured based on multiple sub-fluctuation coefficients.
[0044] First, based on the testing time of the multiple mental health test data, the human brain feature sequence is divided to obtain human brain feature sub-sequences during the testing process of multiple mental health test data. The start and end times corresponding to each mental health test data are extracted. Using this time interval as a basis, all data within the corresponding time period are extracted from the complete human brain feature sequence to form human brain feature sub-sequences that correspond one-to-one with the human brain feature sub-sequences, ensuring accurate matching in the time dimension. For example, the three types of mental health test data collected in S100 correspond to different testing stages. The report data test time is from minutes 1 to 3, the behavioral data test time is from minutes 4 to 8, and the physiological data collection covers the entire process but is divided into sub-data corresponding to the above two stages according to the test task. The time intervals of these two testing stages are extracted. From the complete human brain feature sequence, the EEG parameter data from minutes 1 to 3 are extracted as the human brain feature sub-sequence for the report data stage, and the EEG parameter data from minutes 4 to 8 are extracted as the human brain feature sub-sequence for the behavioral data stage.
[0045] Secondly, the fluctuation amplitudes of the multiple human brain feature subsequences are analyzed and calculated, serving as multiple sub-fluctuation coefficients. The fluctuation amplitude represents the maximum deviation between any human brain feature within the subsequence and the mean of human brain features. The fluctuation amplitude quantifies the dispersion of human brain feature parameters within the subsequence. The sub-fluctuation coefficient, using fluctuation amplitude as a quantitative indicator, reflects the intensity of fluctuation in human brain feature parameters within a single testing phase and is a key parameter for assessing the stability of neural activity during that phase. For each human brain feature subsequence, the mean of all human brain feature parameters within the subsequence is first calculated. Then, the absolute value of the difference between each parameter and the mean is calculated one by one, and the maximum value is selected as the fluctuation amplitude of that human brain feature subsequence; this fluctuation amplitude is the sub-fluctuation coefficient.
[0046] For example, regarding the human brain feature subsequence in the reporting data phase, taking the alpha wave frequency as an example, one set of EEG parameters was collected every 100 milliseconds, resulting in 1800 sets of alpha wave frequency data for this subsequence. The mean of these 1800 sets of data was calculated to be 10.5 Hz. Then, the absolute value of the difference between each set of data and 10.5 Hz was calculated, with the largest absolute difference being 2.3 Hz. Therefore, the sub-fluctuation coefficient of the human brain feature subsequence in the reporting data phase is 2.3 Hz, indicating large fluctuations and low data quality. Similarly, the mean alpha wave frequency of the human brain feature subsequence in the behavioral data phase is calculated to be 9.8 Hz, with the largest absolute difference being 1.7 Hz. Its corresponding sub-fluctuation coefficient is 1.7 Hz, indicating smaller fluctuations and higher data quality.
[0047] Furthermore, multiple mental partitioning gradients are configured based on multiple sub-fluctuation coefficients.
[0048] The method provided in this application embodiment configures multiple mental partitioning gradients based on multiple sub-fluctuation coefficients, including:
[0049] Obtain the preset mental division gradient;
[0050] The preset mental partitioning gradient is corrected based on the ratio of each sub-fluctuation coefficient to the average of multiple sub-fluctuation coefficients to obtain multiple mental partitioning gradients.
[0051] First, obtain the preset mental partitioning gradient. The preset mental partitioning gradient uses the initial number of binary classification node layers in a general scenario as the adjustment benchmark. The number of binary classification node layers is the threshold number for determining whether the data is normal or abnormal. For example, the preset mental partitioning gradient is 2 layers, meaning the base number of layers is 2.
[0052] Secondly, the preset mental partitioning gradient is corrected based on the ratio of each sub-fluctuation coefficient to the mean of multiple sub-fluctuation coefficients, resulting in multiple mental partitioning gradients. The mental partitioning gradient refers to the number of binary classification node layers adapted to the data quality of each testing stage after correcting the preset gradient based on the sub-fluctuation coefficients. High-quality data requires more lenient judgment criteria, therefore, the low-fluctuation stage corresponds to fewer node layers; low-quality data requires stricter judgment criteria, therefore, the high-fluctuation stage corresponds to more node layers. The ratio of each sub-fluctuation coefficient to the mean of all sub-fluctuation coefficients is calculated. This ratio reflects the degree of deviation of the current stage's fluctuation intensity from the average level. For subsequences with a ratio less than 1, the fluctuation is less than the average level, indicating higher data quality. The preset node layer number is reduced according to the ratio to make the judgment criteria more lenient. For subsequences with a ratio greater than 1, the fluctuation is greater than the average level, indicating lower data quality. The preset node layer number is increased according to the ratio to make the judgment criteria more stringent, ultimately obtaining the mental partitioning gradient corresponding to each stage.
[0053] For example, the preset mental gradient is 2 layers, and the mean of the two sub-fluctuation coefficients is calculated to be (2.3 + 1.7) / 2 = 2.0 Hz. In the behavioral data testing phase, the sub-fluctuation coefficient ratio is 1.7 / 2.0 = 0.85, which is less than 1, indicating small fluctuations and high data quality. Therefore, the corresponding node layer number is adjusted to 2 × 0.85 ≈ 1 layer. In the report data testing phase, the sub-fluctuation coefficient ratio is 2.3 / 2.0 = 1.15, which is greater than 1, indicating large fluctuations and low data quality. Therefore, the corresponding node layer number is adjusted to 2 × 1.15 ≈ 3 layers, resulting in 3 decision nodes. The more times the division is performed, the more stringent the decision becomes.
[0054] In this embodiment, precise segmentation of human brain feature sequences is achieved through test time correlation, ensuring a one-to-one correspondence between human brain feature sub-sequences and each test stage, thus guaranteeing the targeted nature of fluctuation analysis. The calculation of sub-fluctuation coefficients quantifies the differences in data quality across different test stages. Based on the correlation between fluctuation intensity and data quality, the mental classification gradient is configured by adjusting the number of binary classification node layers, making anomaly detection more lenient for high-quality data and more stringent for low-quality data. This step provides a personalized judgment framework adapted to the quality characteristics of data from different test stages, effectively solving the problem of insufficient adaptability of general judgment standards and laying a crucial foundation for further gradient correction and precise data selection.
[0055] S300: Analyze the total fluctuation coefficient of the human brain feature sequence, correct the multiple mental segmentation gradients, and obtain multiple corrected mental segmentation gradients.
[0056] In this embodiment, the total fluctuation coefficient of the human brain feature sequence is analyzed, and the multiple mental segmentation gradients are corrected to obtain multiple corrected mental segmentation gradients. S200 has configured mental segmentation gradients adapted to different stages within a single test based on the fluctuation characteristics of each stage. However, these gradients only reflect the immediate fluctuation patterns of a single test and may be affected by the user's temporary state on that day, causing the gradients to deviate from the user's long-term stable neural activity characteristics. The total fluctuation coefficient quantifies the overall fluctuation intensity throughout a single test, while the historical total fluctuation coefficient reflects the long-term baseline level of neural activity. By correcting the gradient through the ratio of these two, the mental segmentation gradient can adapt to both the immediate characteristics of a single test and the user's long-term stable fluctuation patterns, eliminating the interference of occasional fluctuations in a single test, further improving the stability and accuracy of the gradient, and providing a more reliable judgment standard for subsequent abnormal data classification.
[0057] Step S300 in the method provided in this application embodiment includes:
[0058] Analyze the total fluctuation coefficient of the aforementioned human brain feature sequence;
[0059] The average total fluctuation coefficient of the user's brain characteristics over a historical period is obtained as the historical total fluctuation coefficient.
[0060] The ratio of the total volatility coefficient to the historical total volatility coefficient is calculated, and multiple mental partitioning gradients are corrected to obtain multiple corrected mental partitioning gradients.
[0061] First, the total fluctuation coefficient of the human brain feature sequence is analyzed. The total fluctuation coefficient is an indicator that quantifies the overall fluctuation intensity of neural activity throughout a single test. Specifically, it is the maximum deviation between any human brain feature parameter and the mean value of all human brain feature parameters within the complete human brain feature sequence. The larger the total fluctuation coefficient, the more unstable the overall neural activity throughout the single test; conversely, the smaller the coefficient, the more stable the neural activity. Based on the complete human brain feature sequence obtained by S100, the mean value of all human brain feature parameters within the sequence is first calculated. Then, the absolute value of the difference between each parameter and the mean value is calculated one by one. The maximum value is the total fluctuation coefficient of the sequence. The calculation logic is consistent with that of the sub-fluctuation coefficients, ensuring a unified standard for fluctuation quantification and facilitating subsequent comparison with historical data. For example, the complete human brain feature sequence covers minutes 1 to 8, with one set collected every 100 milliseconds, containing 4800 sets of alpha wave frequency data. The mean of these data is calculated to be 10.1 Hz. Then, the absolute value of the difference between each set of data and 10.1 Hz is calculated, with the largest absolute value of the difference being 2.5 Hz. Therefore, the total fluctuation coefficient of this sequence is 2.5 Hz, reflecting the overall brainwave fluctuation intensity of the user throughout this test.
[0062] Secondly, the average total fluctuation coefficient of the user's brain characteristics over a historical period is obtained as the historical total fluctuation coefficient. The historical total fluctuation coefficient refers to the average of the total fluctuation coefficients of the brain characteristic sequences obtained from each test when the user participated in the same type of mental test over a historical period. It reflects the baseline level of the user's overall long-term neural activity fluctuations and serves as a reference for judging whether a single fluctuation deviates from the norm. From the user's historical test database, all historical test records consistent with the current test objective and the same test procedure are retrieved. The total fluctuation coefficient corresponding to each record is extracted, and the average of these total fluctuation coefficients is calculated using the arithmetic mean method, thus obtaining the historical total fluctuation coefficient. For example, retrieving the user's historical records of the past three times participating in the same attention concentration test, their total fluctuation coefficients are 2.2Hz, 2.4Hz, and 2.3Hz respectively, with an average of (2.2+2.4+2.3) / 3=2.3Hz. This value is the user's historical total fluctuation coefficient, reflecting their long-term overall fluctuation baseline.
[0063] Finally, the ratio of the total fluctuation coefficient to the historical total fluctuation coefficient is calculated to correct multiple mental segmentation gradients, resulting in multiple corrected mental segmentation gradients. First, the ratio of the current total fluctuation coefficient to the historical total fluctuation coefficient is calculated, i.e., the overall correction coefficient. Then, each mental segmentation gradient obtained in S200 is multiplied by this ratio to obtain the corrected mental segmentation gradient. If the ratio is greater than 1, it indicates that the overall situation is less stable, and the number of node layers in the corrected gradient increases, resulting in stricter judgment. If the ratio is less than 1, it indicates that the overall situation is more stable, and the number of node layers in the corrected gradient decreases, resulting in more lenient judgment and ensuring that the gradient conforms to the user's long-term characteristics. For example, if the current total fluctuation coefficient is 2.5Hz and the historical total fluctuation coefficient is 2.3Hz, the calculated ratio of the total fluctuation coefficient is 2.5 / 2.3≈1.087. S200 has obtained two mental segmentation gradients: one layer for the behavioral data stage and three layers for the reporting data stage. Using ratio correction: The correction gradient for the behavioral data stage is approximately 1 × 1.087 ≈ 1 layer, maintaining a relatively low layer number. While the overall volatility is slightly high, the volatility in this stage itself is low, making it moderately strict. The correction gradient for the report data stage is approximately 3 × 1.087 ≈ 3 layers, maintaining a relatively high layer number. Furthermore, due to the overall high volatility, the judgment strictness is further adjusted. Ultimately, two correction gradients for the mental classification are obtained: 1 layer and 3 layers.
[0064] In this embodiment, the overall volatility intensity of a single test is quantified by the total volatility coefficient. A long-term volatility baseline for the user is established by combining historical total volatility coefficients. The ratio of the total volatility coefficient accurately reflects the degree of deviation of a single fluctuation from the long-term norm. Based on this ratio, the gradient is modified to retain the advantage of adapting to the characteristics of each stage of a single test in S200, while also incorporating the user's long-term stable neural activity patterns, effectively eliminating the interference of accidental fluctuations in a single test on the judgment criteria. The modified gradient's stability and individual adaptability are further improved, providing a scientific judgment framework that considers both immediate and long-term characteristics for subsequent data classification and screening, effectively reducing abnormal judgment bias caused by short-term fluctuations deviating from the long-term baseline.
[0065] S400: Using the multiple modified mental division gradients, the multiple mental test data are classified as abnormal to obtain a first degree of abnormality, and a second degree of abnormality is obtained based on the total fluctuation coefficient. The degree of abnormality is calculated, and the multiple mental test data are screened and managed.
[0066] In this embodiment, multiple modified mental gradations are used to classify multiple mental test data into anomalies, obtaining a first anomaly score. A second anomaly score is obtained based on the total fluctuation coefficient analysis. The anomaly score is then calculated, and the multiple mental test data are filtered and managed. Personalized, multi-scale indicators generated in the preceding steps are comprehensively used to perform a final, quantitative reliability assessment and filtering of the collected mental test data. Anomaly judgment from a single perspective may have limitations: relying solely on task-specific data (i.e., the first anomaly score) may not reflect the user's overall state drift; relying solely on overall physiological fluctuations (i.e., the second anomaly score) cannot pinpoint which specific test data is problematic. By fusing these two complementary anomaly indicators, a more comprehensive and robust anomaly score can be obtained, thereby determining the data's reference value and making scientific and automated management decisions.
[0067] Step S400 in the method provided in this application embodiment includes:
[0068] The method provided in this application embodiment uses the multiple modified mental partitioning gradients to classify multiple mental test data into anomalies and obtain a first anomaly degree, including:
[0069] A mental data classifier is obtained, wherein the mental data classifier includes multiple mental data classification branches corresponding to multiple test methods, and each mental data classification branch includes multiple classification nodes;
[0070] Obtain the number of multiple key node layers corresponding to multiple corrected mental partition gradients within multiple mental data classification branches;
[0071] The multiple mental health test data are combined with multiple normal mental health test datasets of the target user, and then input into the mental health data classifier for multi-level segmentation. The classifier determines whether the multiple mental health test data are classified as isolated data by multiple key node layers and below, and obtains multiple abnormal classification results, wherein each abnormal classification result includes yes or no.
[0072] Calculate the proportion of "yes" among multiple anomaly classification results to obtain the first anomaly score.
[0073] First, a mental data classifier is obtained, wherein the mental data classifier includes multiple mental data classification branches corresponding to multiple test methods, and each mental data classification branch includes multiple layers of classification nodes.
[0074] The method provided in this application embodiment for obtaining a mental data classifier includes:
[0075] Obtain multiple sample mental test datasets obtained by testing multiple users using multiple test methods within a historical period;
[0076] Sample mental test data were randomly selected from multiple sample mental test datasets to construct multiple multi-level classification nodes. Each classification node performs binary classification on the input mental test data.
[0077] Based on multiple multi-level classification nodes, multiple mental data classification branches are constructed to obtain a mental data classifier.
[0078] First, we acquire multiple sample mental health test datasets obtained from testing multiple users using multiple testing methods over a historical period. These sample mental health test datasets refer to the quantitative and qualitative data collection, including normal and abnormal states, gathered over a historical period by conducting mental health tests on a large number of different users using multiple testing methods consistent with the current test. This dataset serves as the foundational data source for building the classifier. For example, if the current test objective is to assess attention concentration, the testing methods include report data, behavioral data, and physiological data. We retrieve historical test data from the database for 1000 users and split it according to the testing method: we compile 1000 users' attention self-assessment scale data to form a report data sample set; we compile 1000 users' digit cancellation task data to form a behavioral data sample set; and we compile 1000 users' heart rate variability and skin conductance data to form a physiological data sample set.
[0079] Secondly, sample mental health test data were randomly selected from multiple sample mental health test datasets to construct multiple multi-level classification nodes. Each classification node performs binary classification on the input mental health test data. The multi-level classification nodes are binary classification units set in a tiered manner, from lenient to strict, for sample datasets of a single test method. By dividing layer by layer, the data that is separated out is identified. Each node sets a threshold based on corresponding features to divide the data into two groups. The two groups are then passed to the next higher level for further division. The greater the difference between the data and the normal data, the easier it is to be separated out at lower-level nodes. For example, the normal score for reported data is 80-100 points, so a 4-layer node structure is constructed, with thresholds of 60 points for layer 1, 70 points for layer 2, 75 points for layer 3, and 80 points for layer 4; the accuracy rate of behavioral data is 90-100%, so a 4-layer node structure is constructed, with thresholds of 70% for layer 1, 80% for layer 2, 85% for layer 3, and 90% for layer 4; the normal resting heart rate for physiological data is 60-80 beats / min, so a 4-layer node structure is constructed, with thresholds of 40 beats / min for layer 1, 50 beats / min for layer 2, 55 beats / min for layer 3, and 60 beats / min for layer 4.
[0080] Furthermore, based on multiple multi-level classification nodes, multiple mental data classification branches are constructed to obtain a mental data classifier. The mental data classification branch is a decision path formed by concatenating the multi-level classification nodes corresponding to a single test method in the order of layer 1 to layer 2. It specifically processes this type of test data, determining whether to isolate and identify anomalous data through layer-by-layer partitioning. Isolation forest is an algorithm that isolates outliers by randomly partitioning data. Anomalous data, due to its unique characteristics, is more easily and quickly isolated, while normal data requires more partitioning steps.
[0081] For example, the steps for constructing a mental health data classifier based on isolated forests are as follows: Preparing sample data: Collect sample sets of report data, behavioral data, and physiological data from multiple users over a historical period, including normal and abnormal data. Split these data according to the testing method to form three independent training datasets; Training the isolated forest model: Train an isolated forest model, i.e., the corresponding classification branch, for each test method's dataset. The model repeatedly segments the data by randomly selecting features and splitting thresholds, recording the path length required for each sample to be isolated, thus forming an anomaly detection model for that test method; Integrating the classification branches: Integrate the isolated forest models corresponding to the three test methods to form a mental health data classifier, which can separately determine anomalies in report, behavioral, and physiological data. For example, for report data, train an isolated forest model with 1000 sample data points, including normal rating distribution and abnormal fluctuation data. The model splits the data by randomly selecting features such as the mean of the ratings and the standard deviation of the fluctuation, recording the isolated path length for each data point; similarly, train isolated forest models separately for behavioral and physiological data, and then integrate them to obtain the mental health data classifier.
[0082] Secondly, the number of key node layers corresponding to multiple revised mental classification gradients within multiple mental data classification branches is obtained. For example, the number of key node layers is 3 for the report data classification branch; 1 for the behavioral data classification branch; and 2 for the physiological data classification branch.
[0083] Furthermore, the multiple mental health test data are combined with multiple normal mental health test datasets of the target user, and input into the mental health data classifier for multi-level segmentation. The classifier determines whether the multiple mental health test data are classified as isolated data by multiple key node layers and below, obtaining multiple anomaly classification results. Each anomaly classification result includes yes or no. The normal mental health test dataset of the target user is a collection of data obtained from similar tests conducted by the target user when they were in a normal mental state historically, serving as a benchmark for determining the difference between the current data and normal data. The anomaly classification result is the judgment conclusion of each node on the input data, including yes or no. [Yes] indicates that the data is isolated anomaly data; [No] indicates that the data belongs to the normal category.
[0084] For example, the test data in this case is as follows: Report data 75 points, behavioral data 92%, and physiological data 55 times / minute. Report data processing: Input report data classification branch, with three threshold levels: 70, 75, and 80 points respectively. Level 1 classification based on 70 points: 75 points ≥ 70 points, not classified as isolated data; Level 2 classification based on 75 points: 75 points = 75 points, still not classified as isolated data; Level 3 classification based on 80 points: 75 points < 80 points, deviating from the normal range, classified as isolated data. The abnormal classification result for the report data is [Yes]. Behavioral data processing: Input behavioral data classification branch, with one key node level and a threshold of 90%. Level 1 classification based on 90 points: 92% ≥ 90%, within the normal range, not classified as isolated data. The abnormal classification result for the behavioral data is [No]. Physiological data processing: Input physiological data classification branch, with two key node levels, and threshold levels of 50 times / minute and 60 times / minute respectively. The first layer is divided according to 50 beats / min: 55 beats / min ≥ 50 beats / min is not classified as isolated data; the second layer is divided according to 60 beats / min: 55 beats / min < 60 beats / min deviates from the normal range and is classified as isolated data. The abnormal classification result for physiological data is [Yes]. Finally, three abnormal classification results are obtained: [Yes, No, Yes].
[0085] Finally, the proportion of "yes" results within multiple outlier classifications is calculated to obtain the first anomaly score. The first anomaly score is an indicator that quantifies the degree of anomaly in the current mental health test data. It is obtained by statistically analyzing the proportion of "yes" results in all outlier classifications to the total number of judgments. The value ranges from 0 to 1, with a higher value indicating a higher probability of data anomaly. In the calculation, the number of "yes" results in all outlier classifications is first counted, and then divided by the total number of outlier classifications to obtain the first anomaly score. For example, if the outlier classification results are [yes, no, yes], with a total of 3 counts, and "yes" occurs 2 times, the first anomaly score = 2 / 3 ≈ 0.67.
[0086] The method provided in this application embodiment includes obtaining a second anomaly degree based on the total fluctuation coefficient analysis, calculating the anomaly degree, and filtering and managing the multiple mental health test data, including:
[0087] The total fluctuation coefficient is input into the fluctuation anomaly classification table, and the output is the second anomaly degree. The fluctuation anomaly classification table is constructed based on the mapping relationship between the sample total fluctuation coefficient set and the sample second anomaly degree set. Each sample second anomaly degree includes the proportion of abnormal mental test data under different sample total fluctuation coefficients.
[0088] The anomaly score is calculated based on the first and second anomalies.
[0089] Based on the degree of abnormality, the multiple mental health test data are filtered and managed, wherein the filtering and management includes recording or discarding records.
[0090] First, the total fluctuation coefficient is input into a fluctuation anomaly classification table, and the output is the second anomaly degree. This fluctuation anomaly classification table is constructed based on the mapping relationship between the sample total fluctuation coefficient set and the sample second anomaly degree set. Each sample second anomaly degree includes the proportion of abnormal data in mental health test data under different sample total fluctuation coefficients. A large number of historical total fluctuation coefficients of similar mental health tests are collected to form a sample total fluctuation coefficient set. Simultaneously, the proportion of abnormal data in the test data corresponding to each total fluctuation coefficient is statistically analyzed to form a sample second anomaly degree set. A fluctuation anomaly classification table is constructed based on the mapping relationship between the sample total fluctuation coefficient set and the sample second anomaly set for quickly querying the anomaly probability corresponding to a specific total fluctuation coefficient. The total fluctuation coefficient calculated in this test is compared with the value in the fluctuation anomaly classification table to find the anomaly proportion corresponding to the matching total fluctuation coefficient; this proportion is the second anomaly degree. For example, the total fluctuation coefficient set of the sample includes the total fluctuation coefficients of historical tests: 2.0Hz, 2.2Hz, 2.5Hz, and 2.8Hz, with corresponding sample second anomaly sets of: 0.2, 0.25, 0.3, and 0.4; a fluctuation anomaly classification table is constructed: 2.0Hz: 0.2; 2.2Hz: 0.25; 2.5Hz: 0.3; 2.8Hz: 0.4. In this test, the total fluctuation coefficient is 2.5Hz, and after querying the table, the output second anomaly value is 0.3.
[0091] Secondly, the anomaly score is calculated based on the first and second anomaly scores. The first anomaly score represents the anomalous characteristics of a local stage, while the second anomaly score represents the overall probability of anomaly in the fluctuation. The anomaly score is the final anomaly determination value obtained by combining the first and second anomaly scores, ranging from 0 to 1. The larger the value, the higher the overall probability of the data being anomalous. Weights are assigned to the first and second anomaly scores, and their average is calculated through weighted summation to obtain the anomaly score. For example, if the first anomaly score is 0.67, the second anomaly score is 0.3, and both are weighted at 0.5, the anomaly score is calculated as: 0.67 × 0.5 + 0.3 × 0.5 = 0.335 + 0.15 = 0.485.
[0092] Finally, based on the anomaly score, the multiple mental health test data are filtered and managed. This filtering and management includes recording or discarding records. A preset anomaly score threshold is used, for example, 0.5. The calculated anomaly score is compared to this threshold: if the anomaly score is less than the threshold, the data is considered valid, recorded, and retained for subsequent analysis; if the anomaly score is greater than or equal to the threshold, the data is considered invalid, discarded, and invalid data is removed to avoid interfering with the analysis. For example, if the preset anomaly score threshold is 0.5, and the anomaly score of this test is 0.485, which is less than 0.5, the data is considered valid, recorded, and the report data, behavioral data, and physiological data are saved for subsequent mental health analysis. If the anomaly score is 0.52, which is greater than 0.5, the data is considered invalid, discarded, and the data is not saved to avoid affecting the mental health analysis results.
[0093] In this embodiment, by integrating isolated detection based on local task data and volatility risk assessment based on overall physiological state, a multi-dimensional and quantitative fusion assessment of the reliability of multimodal mental health test data is achieved. The final output is a comprehensive anomaly index, providing an objective and unified decision-making basis for data screening and management, improving the overall quality and reliability of the final data stored, and effectively reducing the risk of inaccurate subsequent analysis and judgment due to the use of contaminated data.
[0094] The embodiments of this application, through the specific implementation methods described above, achieve the following technical effects:
[0095] This application provides a method and system for screening and managing mental state data that integrates machine learning. By simultaneously acquiring multimodal data, configuring dynamic gradients, and performing multi-dimensional anomaly fusion analysis, it effectively solves the problem of data distortion in mental state tests caused by temporary user states, significantly improving the authenticity and reliability of the final dataset used for analysis. First, the simultaneous acquisition of behavioral data and EEG physiological parameters provides objective evidence for data authenticity verification. Then, through an adaptive gradient adjustment mechanism based on local and global physiological fluctuation characteristics, a dynamic classification standard accurately reflecting the user's real-time state is constructed. Finally, by integrating local isolation detection and overall volatility assessment, a multi-dimensional quantitative evaluation of data reliability is achieved. The entire process from data acquisition and quality assessment to final screening is optimized, ensuring high quality and high reliability of the output data, providing a solid and reliable data foundation for subsequent mental state analysis.
[0096] Example 2, as Figure 2 As shown, this application provides a mental state data filtering and management system that integrates machine learning, the system comprising:
[0097] The feature acquisition module 11 is used to collect multiple mental test data of the target user through multiple test methods, and to collect the target user's human brain feature parameters during the test to obtain the human brain feature sequence;
[0098] Gradient partitioning module 12 is used to partition the human brain feature sequence according to the test time of the multiple mental test data, obtain multiple human brain feature sub-sequences, analyze multiple sub-fluctuation coefficients respectively, and configure multiple mental partitioning gradients.
[0099] Gradient correction module 13 is used to analyze the total fluctuation coefficient of the human brain feature sequence, correct the multiple mental segmentation gradients, and obtain multiple corrected mental segmentation gradients.
[0100] The anomaly assessment module 14 is used to classify multiple mental test data into anomalies using the multiple modified mental division gradients, obtain a first anomaly degree, and obtain a second anomaly degree based on the total fluctuation coefficient analysis, calculate the anomaly degree, and screen and manage the multiple mental test data.
[0101] In one embodiment, the feature acquisition module 11 is further configured to:
[0102] Multiple psychological test data were collected from the target users through various testing methods, including report data, behavioral data, and physiological data.
[0103] During the testing process, human brain feature parameters of the target user are collected to obtain human brain feature sequences, including brainwave parameters.
[0104] In one embodiment, the gradient partitioning module 12 is further configured to:
[0105] Based on the testing time of the multiple mental test data, the human brain feature sequence is divided to obtain human brain feature sub-sequences during the testing process of the multiple mental test data;
[0106] The fluctuation amplitude of the multiple human brain feature sub-sequences is analyzed and calculated as multiple sub-fluctuation coefficients, wherein the fluctuation amplitude is the maximum deviation between any human brain feature within the human brain feature sub-sequence and the mean of human brain features.
[0107] Multiple mental partitioning gradients are configured based on multiple sub-fluctuation coefficients.
[0108] Among them, multiple mental partitioning gradients are configured based on multiple sub-fluctuation coefficients, including:
[0109] Obtain the preset mental division gradient;
[0110] The preset mental partitioning gradient is corrected based on the ratio of each sub-fluctuation coefficient to the average of multiple sub-fluctuation coefficients to obtain multiple mental partitioning gradients.
[0111] In one embodiment, the gradient correction module 13 is further configured to:
[0112] Analyze the total fluctuation coefficient of the aforementioned human brain feature sequence;
[0113] The average total fluctuation coefficient of the user's brain characteristics over a historical period is obtained as the historical total fluctuation coefficient.
[0114] The ratio of the total volatility coefficient to the historical total volatility coefficient is calculated, and multiple mental partitioning gradients are corrected to obtain multiple corrected mental partitioning gradients.
[0115] In one embodiment, the anomaly assessment module 14 is further configured to:
[0116] Specifically, the multiple modified mental partitioning gradients are used to classify multiple mental test data into anomalies, obtaining a first anomaly degree, including:
[0117] A mental data classifier is obtained, wherein the mental data classifier includes multiple mental data classification branches corresponding to multiple test methods, and each mental data classification branch includes multiple classification nodes;
[0118] Obtain the number of multiple key node layers corresponding to multiple corrected mental partition gradients within multiple mental data classification branches;
[0119] The multiple mental health test data are combined with multiple normal mental health test datasets of the target user, and then input into the mental health data classifier for multi-level segmentation. The classifier determines whether the multiple mental health test data are classified as isolated data by multiple key node layers and below, and obtains multiple abnormal classification results, wherein each abnormal classification result includes yes or no.
[0120] Calculate the proportion of "yes" among multiple anomaly classification results to obtain the first anomaly score.
[0121] Among them, obtaining the mental data classifier includes:
[0122] Obtain multiple sample mental test datasets obtained by testing multiple users using multiple test methods within a historical period;
[0123] Sample mental test data were randomly selected from multiple sample mental test datasets to construct multiple multi-level classification nodes. Each classification node performs binary classification on the input mental test data.
[0124] Based on multiple multi-level classification nodes, multiple mental data classification branches are constructed to obtain a mental data classifier.
[0125] Specifically, based on the analysis of the total fluctuation coefficient, a second anomaly degree is obtained; the anomaly degree is calculated; and the multiple mental health test data are screened and managed, including:
[0126] The total fluctuation coefficient is input into the fluctuation anomaly classification table, and the output is the second anomaly degree. The fluctuation anomaly classification table is constructed based on the mapping relationship between the sample total fluctuation coefficient set and the sample second anomaly degree set. Each sample second anomaly degree includes the proportion of abnormal mental test data under different sample total fluctuation coefficients.
[0127] The anomaly score is calculated based on the first and second anomalies.
[0128] Based on the degree of abnormality, the multiple mental health test data are filtered and managed, wherein the filtering and management includes recording or discarding records.
[0129] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.
[0130] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.
[0131] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.
Claims
1. A mental state data screening management method fused with machine learning, characterized by, The method includes: Multiple mental test data of target users were collected through multiple testing methods, and human brain feature parameters of target users were collected during the testing process to obtain human brain feature sequences; Based on the test time of the multiple mental test data, the human brain feature sequence is divided to obtain multiple human brain feature sub-sequences. Multiple sub-fluctuation coefficients are analyzed respectively, and multiple mental segmentation gradients are configured. Analyze the total fluctuation coefficient of the human brain feature sequence, correct the multiple mental segmentation gradients, and obtain multiple corrected mental segmentation gradients; The multiple modified mental division gradients are used to classify multiple mental test data into anomalies, obtain a first anomaly degree, and obtain a second anomaly degree based on the total fluctuation coefficient. The anomaly degree is calculated, and the multiple mental test data are then filtered and managed. Based on the testing time of the multiple mental test data, the human brain feature sequence is divided to obtain multiple human brain feature sub-sequences, and multiple mental segmentation gradients are configured for each, including: Based on the testing time of the multiple mental test data, the human brain feature sequence is divided to obtain human brain feature sub-sequences during the testing process of the multiple mental test data; The fluctuation amplitude of the multiple human brain feature sub-sequences is analyzed and calculated as multiple sub-fluctuation coefficients, wherein the fluctuation amplitude is the maximum deviation between any human brain feature within the human brain feature sub-sequence and the mean of human brain features. Multiple mental partitioning gradients are configured based on multiple sub-fluctuation coefficients; Analyze the total fluctuation coefficient of the human brain feature sequence, correct the multiple mental segmentation gradients, and obtain multiple corrected mental segmentation gradients, including: Analyze the total fluctuation coefficient of the aforementioned human brain feature sequence; The average total fluctuation coefficient of the user's brain characteristics over a historical period is obtained as the historical total fluctuation coefficient. The ratio of the total volatility coefficient to the historical total volatility coefficient is calculated, and multiple mental partitioning gradients are corrected to obtain multiple corrected mental partitioning gradients.
2. The mental state data screening management method fused with machine learning according to claim 1, characterized in that, Multiple mental health test data were collected from the target users through various testing methods. During the testing process, human brain characteristic parameters of the target users were also collected to obtain human brain characteristic sequences, including: Multiple psychological test data were collected from the target users through various testing methods, including report data, behavioral data, and physiological data. During the testing process, human brain feature parameters of the target user are collected to obtain human brain feature sequences, including brainwave parameters.
3. The mental state data screening management method fused with machine learning according to claim 1, characterized in that, Multiple mental partitioning gradients are configured based on multiple sub-fluctuation coefficients, including: Obtain the preset mental division gradient; The preset mental partitioning gradient is corrected based on the ratio of each sub-fluctuation coefficient to the average of multiple sub-fluctuation coefficients to obtain multiple mental partitioning gradients.
4. The method for filtering and managing mental state data incorporating machine learning according to claim 1, characterized in that, Using the aforementioned multiple modified mental partitioning gradients, anomaly classification is performed on multiple mental test data to obtain a first anomaly score, including: A mental data classifier is obtained, wherein the mental data classifier includes multiple mental data classification branches corresponding to multiple test methods, and each mental data classification branch includes multiple classification nodes; Obtain the number of multiple key node layers corresponding to multiple corrected mental partition gradients within multiple mental data classification branches; The multiple mental health test data are combined with multiple normal mental health test datasets of the target user, and then input into the mental health data classifier for multi-level segmentation. The classifier determines whether the multiple mental health test data are classified as isolated data by multiple key node layers and below, and obtains multiple abnormal classification results, wherein each abnormal classification result includes yes or no. Calculate the proportion of "yes" among multiple anomaly classification results to obtain the first anomaly score.
5. The method for filtering and managing mental state data incorporating machine learning according to claim 4, characterized in that, Obtain a mental data classifier, including: Obtain multiple sample mental test datasets obtained by testing multiple users using multiple test methods within a historical period; Sample mental test data were randomly selected from multiple sample mental test datasets to construct multiple multi-level classification nodes. Each classification node performs binary classification on the input mental test data. Based on multiple multi-level classification nodes, multiple mental data classification branches are constructed to obtain a mental data classifier.
6. The method for filtering and managing mental state data incorporating machine learning according to claim 1, characterized in that, Based on the analysis of the total fluctuation coefficient, a second anomaly degree is obtained. The anomaly degree is then calculated, and the multiple mental health test data are filtered and managed, including: The total fluctuation coefficient is input into the fluctuation anomaly classification table, and the output is the second anomaly degree. The fluctuation anomaly classification table is constructed based on the mapping relationship between the sample total fluctuation coefficient set and the sample second anomaly degree set. Each sample second anomaly degree includes the proportion of abnormal mental test data under different sample total fluctuation coefficients. The anomaly score is calculated based on the first and second anomalies. Based on the degree of abnormality, the multiple mental health test data are filtered and managed, wherein the filtering and management includes recording or discarding records.
7. A mental state data screening and management system integrating machine learning, characterized in that, The system is used to implement the mental state data screening and management method incorporating machine learning as described in any one of claims 1-6, the system comprising: The feature acquisition module is used to collect multiple mental test data of the target user through multiple test methods, and to collect the target user's human brain feature parameters during the test to obtain the human brain feature sequence; The gradient partitioning module is used to partition the human brain feature sequence according to the test time of the multiple mental test data, obtain multiple human brain feature sub-sequences, analyze multiple sub-fluctuation coefficients respectively, and configure multiple mental partitioning gradients. The gradient correction module is used to analyze the total fluctuation coefficient of the human brain feature sequence, correct the multiple mental segmentation gradients, and obtain multiple corrected mental segmentation gradients. The anomaly assessment module is used to classify multiple mental test data into anomalies using the multiple modified mental division gradients, obtain a first anomaly degree, and obtain a second anomaly degree based on the total fluctuation coefficient analysis, calculate the anomaly degree, and filter and manage the multiple mental test data.