A psychological state assessment method based on feature extraction

Abstract

The present invention proposes a psychological state assessment method based on feature extraction, which relates to the fields of mental health and deep learning. The method comprehensively collects individual physiological indicators, behavioral data and language usage characteristics, uses principal component analysis to extract main dynamic features, and constructs a psychological state assessment model based on machine learning; the model can receive the physiological indicators, behavioral data and language usage characteristics of the person to be assessed as input, and output the corresponding psychological state and its confidence; by regularly evaluating the model performance and adjusting and optimizing it, it is ensured that the model accuracy continues to meet the standards. The present invention can improve the real-time, accuracy and intelligence level of psychological state assessment, and provide a new monitoring and early warning method for the field of mental health.

Description

Technical Field

[0001] The present invention provides a psychological state assessment method based on feature extraction, and specifically relates to the fields of mental health and deep learning. Background Art

[0002] In the field of mental health assessment, traditional methods rely heavily on the subjective judgment and experience of psychologists or assessors, and subjectivity may affect the accuracy and reliability of the assessment. The type and quantity of data that can be collected by traditional methods are limited and cannot fully reflect the true mental state of the person being assessed.

[0003] In the process of implementing engineering projects, material supply and equipment stability are key factors affecting the progress of the enterprise's engineering projects. Accurately predicting the engineering status and workload of the enterprise is one of the effective means to enhance the competitiveness of the enterprise. However, factors such as changes in resource supply and equipment failures often interfere with the prediction results. Compared with large-scale standardized engineering models, engineering data in small-scale non-standardized models are more chaotic. The accuracy of the prediction results is low when the project progress is predicted by analyzing the trend of historical data or the average workload of each period.

[0004] In order to solve the above problems, the present invention proposes a psychological state assessment method based on feature extraction. This method uses advanced data acquisition technology and machine learning algorithms to more objectively and accurately assess the psychological state of an individual. By collecting multidimensional feature data of an individual (including physiological data, behavioral data, language usage characteristics, etc.), and using machine learning algorithms to analyze and predict these data, a comprehensive and objective assessment of the individual's psychological state can be achieved. This method not only improves the accuracy and efficiency of the assessment, but also better meets the needs of personalized mental health services. In addition, with the continuous development of big data and artificial intelligence technology, the psychological state assessment method based on feature extraction still has great room for development and potential. Summary of the invention

[0005] In order to solve the above problems, the present invention proposes a psychological state assessment method based on feature extraction, comprising the following steps:

[0006] Step S1, data collection: physiological indicators, behavioral data, language usage characteristics. Data collection can be carried out through wearable devices, smartphone applications, psychological questionnaires, etc.

[0007] Step S2, data preprocessing: based on normal human behavior and expression, data is screened to remove unreasonable data, malicious data, and fabricated data;

[0008] Step S3, constructing a data set: dividing the preprocessed data into a normal operation data set and an abnormal operation data set, and encoding the collected feature data and the corresponding psychological state;

[0009] Step S4, feature extraction: physiological index feature extraction, behavioral data feature extraction and language usage feature extraction;

[0010] Step S5, model construction: using the extracted features, construct a psychological state assessment model based on machine learning, a support vector machine classification deep learning model, and optimize the model through a validation set to ensure that the model mean square error is above 0.95;

[0011] Step S6, psychological state assessment and result output: input the collected data of the individual to be assessed into the model to conduct psychological state assessment, and the model will output one or more psychological state labels and corresponding confidence levels;

[0012] Step S7, system performance update and optimization: Regularly evaluate model performance to ensure that accuracy continues to meet standards, and adjust and optimize model parameters based on actual individual identification data.

[0013] Furthermore, the specific steps of step S1 are:

[0014] S11: Determine the data collection objectives and scope to obtain a large sample set from comprehensive questionnaires, wearable devices, smartphone app tracking, and social media data, including physiological indicators, behavioral data, and language use characteristics;

[0015] S12: Clarify the content of physiological index collection, determine the physiological indicators that need to be collected, including heart rate, blood pressure, and respiratory rate, and select appropriate wearable devices, such as smart bracelets and health monitoring watches, to ensure that these physiological indicators can be accurately measured;

[0016] S13. Clarify the content of behavioral data collection and list the behavioral data that need to be collected, including daily activity, sleep patterns, and eating habits. Use existing smartphone applications to track and record these behavioral data, or design specific diaries or record sheets for participants to fill out;

[0017] S14. Clarify the content of language usage characteristics collection and determine the language usage characteristics that need to be analyzed, including speech emotional tendency, speech attitude and expression mode;

[0018] S15. Prepare data collection tools and environment. Prepare corresponding data collection tools and ensure the stability of the network environment so that data can be transmitted and stored in real time;

[0019] S16. Obtain consent from participants and conduct data collection training, explain the purpose and process of data collection, as well as the possible risks and privacy protection measures involved, obtain explicit consent from participants, and have them sign relevant informed consent forms, and train participants to ensure that they can use data collection tools correctly and understand the requirements of data collection;

[0020] S17. Perform data collection, instruct participants to start using wearable devices and smartphone applications for data collection, and regularly monitor the progress and quality of data collection to ensure the integrity and accuracy of the data;

[0021] S18. Preliminary data processing and verification: preliminary processing of the collected raw data to ensure the integrity and objectivity of the data;

[0022] S19. Data storage and backup: store the processed data on a secure and reliable server or storage device and ensure data encryption and security. Back up data regularly to prevent data loss or damage, and develop a data recovery plan to deal with possible data loss.

[0023] Furthermore, the specific steps of step S2 are:

[0024] S21. Preliminary data screening: preliminarily browse the collected raw data to understand the overall situation of the data and possible problems, and preliminarily identify obviously unreasonable, abnormal or erroneous data records based on common sense and normal human behavior patterns;

[0025] S22. Set filtering rules. Based on normal human behavior and language expression habits, formulate a series of data filtering rules, including numerical range, frequency of occurrence, logical association, etc., to automatically identify and filter unreasonable data;

[0026] S23, automated screening, automatically removing data that obviously does not conform to the rules through conditional judgment, regular expression matching and other methods;

[0027] S24. Manual review and adjustment: Manual review of the data after automated screening to ensure the accuracy and completeness of the screening. For ambiguous or complex data, manual judgment and adjustment are performed to avoid accidental or missed deletions.

[0028] S25. Identification and removal of malicious and fabricated data, combined with other information sources, including user behavior logs and IP addresses, to analyze abnormal patterns, sudden data spikes and unreasonable distributions in the data, and remove fabricated or maliciously submitted data;

[0029] Furthermore, the specific steps of step S3 are:

[0030] S31. Data classification: according to the preprocessed data, the data are divided into two categories: a normal operating data set and an abnormal operating data set. The normal operating data set contains data showing normal behavior or psychological state, while the abnormal operating data set contains abnormal or deviating behavior or psychological state data;

[0031] S32, feature encoding, including physiological indicators, behavioral data, language usage features, for categorical features including sleep patterns, eating habits, verbal emotional tendencies, verbal attitudes and expressions, one-hot encoding is performed for conversion, for continuous features including heart rate, blood pressure and respiratory rate, normalization is performed to ensure that they are on the same scale and improve the training efficiency of the model;

[0032] S33, psychological state coding, encode each psychological state, psychological state includes cognitive psychological state, emotional psychological state, volitional psychological state and motivational psychological state, which are simply divided into optimal psychological state, general psychological state and bad psychological state. Each psychological state can be subdivided. In order to identify in the data set, the above three psychological states are subdivided, "1" represents stable state, "2" represents anxious state, "3" represents depressed state, "4" represents stressed state, "5" represents relaxed state, "6" represents excited state, "7" represents angry state, "8" represents confused state, "9" represents happy state and "10" represents sad state. The optimal psychological state includes "5" and "9", the general psychological state includes "1" and "5", and the bad psychological state includes "2", "3", "4", "7", "8", "10";

[0033] S34, data corresponds to psychological state, ensuring that each piece of data corresponds to the corresponding psychological state code, which is achieved by adding a new column to the data set, which contains the psychological state code corresponding to each piece of data;

[0034] S35, data set segmentation, storing the constructed data set in .dat or .mat format, and creating a backup to prevent data loss, ensuring the security and accessibility of the data set;

[0035] Furthermore, the feature extraction in step S4 is specifically as follows:

[0036] Extract continuous features, including heart rate, blood pressure, and respiratory rate. Apply principal component analysis to extract the main dynamic features of the data, reduce the dimension of the features, and retain the main dynamic data in the data;

[0037] Preferably, the application of the principal component analysis method is specifically manifested as follows: first, the original continuous feature data is standardized to eliminate the impact of different features due to dimensional and order of magnitude differences; then, the eigenvalues ​​and eigenvectors are analyzed by calculating the covariance matrix of the data; then, the principal components are selected according to the size of the eigenvalues, usually the first few principal components with larger eigenvalues ​​are selected, and these principal components can retain most of the variation information of the original data; finally, the original data is projected onto the selected principal components to obtain the reduced-dimensional data, which will be used as input for subsequent model training and psychological state assessment.

[0038] Furthermore, the specific steps of step S5 are:

[0039] S51, data preparation, taking the main features after feature extraction of continuous features and hot encoding of categorical features as input of the training model, and the psychological state as output;

[0040] S52, constructing a support vector machine classification (SVC) model, selecting a radial basis function (RBF) kernel function, and setting parameters of the SVC model, a penalty coefficient C, and a γ parameter;

[0041] S53, model training, using the training set data to train the SVC model, and using the optimization algorithm sequence minimum optimization algorithm SMO to find the support vector and determine the decision boundary;

[0042] S54, model verification and optimization, use the verification set to evaluate the performance of the SVC model, and adjust the model parameter penalty coefficient C and γ parameter;

[0043] S55, performance metrics. For classification problems, the performance metrics include recall rate above 90% and F1 score above 0.85, ensuring that the model meets the predetermined performance indicators on the validation set and test set.

[0044] Furthermore, the psychological state assessment and its corresponding confidence level in step S6 are specifically:

[0045] According to the model of the deep learning model, the physiological indicators, behavioral data, and language usage characteristics of the person to be evaluated are input into the model, and the corresponding psychological state is output; there are many psychological states of the person to be evaluated, and the psychological states evaluated have their own confidence levels. The output with a higher confidence level is the main psychological characteristic of the person to be evaluated, and the psychological state with a low confidence level is the secondary psychological state of the person to be evaluated; in the evaluation process using the deep learning model, a confidence threshold is set. When the confidence level of a certain psychological state exceeds this threshold, the model's judgment on the state is considered to be credible, and the main and secondary psychological states are distinguished according to the confidence level.

[0046] Furthermore, the specific steps of step S7 are:

[0047] S71. Regularly evaluate model performance, set the time interval for regular evaluation, use recall rate and F1 score as evaluation indicators, and conduct quantitative evaluation of model performance;

[0048] S72. Ensure that accuracy continues to meet standards. According to the requirements of the evaluation indicators, determine whether the current model is invalid, monitor model performance, and promptly detect and handle performance degradation. If the model performance is lower than the set standard, the optimization process is triggered;

[0049] S73, data re-collection and feature extraction, the individuals obtained during this period are also used as data sets to re-divide the data set and extract features, and optimize the features to adapt to the dynamic changes of the current application scenarios;

[0050] S74, model retraining and validation, retrain the model using optimized features and parameters, test the performance of the new model on an independent validation set to ensure that performance is improved and no new bias is introduced;

[0051] S75. Repeat the above optimization steps regularly to adapt to changes in data distribution and dynamic changes in turntable evaluation requirements.

[0052] The present invention has the following beneficial effects:

[0053] The present invention proposes a psychological state assessment method based on feature extraction. The method constructs a comprehensive and objective psychological state assessment system by comprehensively collecting individual physiological indicators, behavioral data and language usage characteristics. Through seven steps including data collection, preprocessing, data set construction, feature extraction, model construction, psychological state assessment and result output, and system performance update and optimization, an accurate assessment of the individual's psychological state is achieved. Compared with the traditional manual psychological state assessment method, the psychological state assessment method of the present invention has the characteristics of comprehensiveness, objectivity and accuracy, and can provide strong support and help for the field of mental health. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the structures shown in these drawings without paying creative work.

[0055] Figure 1 is a flow chart of a method for evaluating a psychological state based on feature extraction in an embodiment of the present invention;

[0056] Figure 2The present invention is a flowchart of a principal component analysis method of a psychological state assessment method based on feature extraction in an embodiment of the present invention. DETAILED DESCRIPTION

[0057] In order to make the purpose, technical scheme and advantages of the present application clearer, some embodiments of the present application are further described in detail below in conjunction with the accompanying drawings and Examples. It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application. However, it will be appreciated by those skilled in the art that in the various embodiments of the present application, many technical details are proposed in order to enable the reader to better understand the present application. However, even without these technical details and various changes and modifications based on the following embodiments, the technical scheme claimed in the present application can also be implemented.

[0058] Example 1: See Figure 1 , is a flow chart of a psychological state assessment method based on feature extraction in an embodiment of the present invention, a psychological state assessment method based on feature extraction, comprising the following steps:

[0059] S1. Data collection. Data collection is the first step in psychological state assessment. It involves comprehensive collection of individual physiological indicators such as heart rate, blood pressure, respiratory rate, behavioral data such as daily activity, sleep patterns, and language usage characteristics, including verbal emotional tendencies and expressions, through various methods such as wearable devices, smartphone applications, and psychological questionnaires. This step ensures that the basic data for the assessment is comprehensive and accurate.

[0060] S2. Data preprocessing. In the data preprocessing stage, the collected raw data is screened based on people’s normal behavior and expression. This process aims to remove unreasonable data, malicious data, and fabricated data, ensure the quality and authenticity of the data, and provide a reliable basis for subsequent data analysis.

[0061] S3. Build a dataset. The dataset building stage is to divide the preprocessed data into a normal operating dataset and an abnormal operating dataset. At the same time, the collected feature data and the corresponding psychological state are encoded to provide standardized input for subsequent model training.

[0062] S4. Feature extraction. Feature extraction is the process of extracting key information from physiological indicators, behavioral data, and language usage characteristics. This information is crucial for training deep learning models, helping the models to more accurately identify and understand patterns in the data, thereby improving the accuracy of psychological state assessment.

[0063] S5. Model construction. In the model construction stage, the extracted features are used to build a psychological state assessment model based on machine learning. By selecting an appropriate algorithm, a support vector machine classification deep learning model is used, and model optimization is performed to ensure the accuracy and efficiency of the model. The validation set is used to adjust the model parameters to achieve the best assessment performance.

[0064] S6. Psychological state assessment and result output. During the psychological state assessment stage, the collected data of the individual to be assessed is input into the trained model. The model will output one or more psychological state labels and corresponding confidence levels, which helps to fully understand the individual's psychological state and provide strong support for subsequent mental health interventions.

[0065] S7. System performance update and optimization. System performance update and optimization is an important part of ensuring the continued effectiveness of the model. By regularly evaluating model performance, performance degradation can be discovered and handled in a timely manner. At the same time, model parameters are adjusted and optimized based on actual individual identification data to maintain the accuracy and adaptability of the model. This step ensures the continuous improvement and enhancement of the evaluation system.

[0066] The psychological state assessment method based on feature extraction proposed in the embodiment of the present invention is an innovative, technology-driven psychological assessment method, which comprehensively and objectively assesses the individual's psychological state through multi-dimensional data collection, including physiological indicators, behavioral data and language usage characteristics; the method first performs data preprocessing to remove unreasonable and malicious data to ensure the quality and authenticity of the data; then, feature extraction is performed using methods such as principal component analysis to extract key information from a large amount of data; finally, the psychological state is accurately assessed through a machine learning model, and the specific psychological state and its confidence are output; in addition, the method also includes regular evaluation and optimization of system performance to ensure the continued accuracy and reliability of the evaluation results.

[0067] Example 2: See Figure 2 , is a flow chart of a principal component analysis method of a psychological state assessment method based on feature extraction in an embodiment of the present invention, specifically comprising:

[0068] Preprocessing: In this stage, the raw continuous feature data is standardized and normalized. Standardization ensures that the data has zero mean and unit variance, while normalization scales the data to the range of [0,1], which facilitates subsequent data processing and analysis.

[0069] Data matrix construction and covariance matrix calculation: The preprocessed data will be assembled into a data matrix, and the covariance matrix of the matrix will be calculated. The covariance matrix can reveal the correlation between different features and their respective variances, providing an important basis for subsequent feature selection.

[0070] Feature retention: Based on the eigenvalues ​​and eigenvectors calculated based on the covariance matrix, the principal components to be retained are selected according to the size of the eigenvalues. The size of the eigenvalues ​​reflects the variance contribution of the corresponding eigenvectors in the data. Therefore, selecting principal components with larger eigenvalues ​​can retain the main information in the data.

[0071] Principal component selection: After determining the principal components that need to be retained, the original data is projected onto these selected principal components to obtain a reduced-dimensional data representation. This step is achieved by multiplying the original data by the selected eigenvectors, which effectively reduces the dimensionality of the data while retaining the main features of the data.

[0072] Model output: Finally, the selected principal components and their corresponding eigenvalues ​​and eigenvectors are used as output. These outputs will serve as inputs to the deep learning model and provide basic data for subsequent model training.

[0073] This flowchart describes a data processing process that starts from raw data, goes through preprocessing, feature selection, dimensionality reduction and other steps, and finally obtains data suitable for deep learning model input.

[0074] It should be understood that those skilled in the art can make improvements or changes based on the above description, and all such improvements and changes should fall within the scope of protection of the appended claims of the present invention. Parts not described in detail in this specification belong to the prior art known to those skilled in the art.

Claims

1. A psychological state assessment method based on feature extraction, characterized in that: include: Step S1: Data collection, through group data collection, comprehensively collect individual physiological indicators, behavioral data and language usage characteristics; Step S2: Data preprocessing: based on normal human behavior and expression, the data is screened to remove unreasonable data, malicious data, and fabricated data; Step S3: construct a data set, classify and encode the preprocessed data, and the specific steps of step S3 are: S31. The pre-processed data are divided into two categories: normal operation and abnormal operation. The normal operation data set covers data showing normal behavior or psychological state, while the abnormal operation data set specifically collects abnormal or deviating behavior or psychological state data; S32, perform one-hot encoding on categorical features and normalize continuous features. For categorical features, sleep patterns, eating habits, verbal emotional tendencies, verbal attitudes, and expressions are converted using one-hot encoding. For continuous features, physiological indicators such as heart rate, blood pressure, and respiratory rate are normalized to ensure that they are on the same scale. S33. Code each mental state, "1" represents a stable state, "2" represents an anxious state, "3" represents a depressed state, "4" represents a stressed state, "5" represents a relaxed state, "6" represents an excited state, "7" represents an angry state, "8" represents a confused state, "9" represents a happy state, and "10" represents a sad state, where "5" and "9" represent the best mental state, "1" and "5" represent a general mental state, and "2", "3", "4", "7", "8", and "10" represent a bad mental state; S34, data corresponds to psychological state, ensuring that each piece of data corresponds to the corresponding psychological state code, which is achieved by adding a new column to the data set, which contains the psychological state code corresponding to each piece of data; S35, data set segmentation, storing the constructed data set in .dat or .mat format, and creating a backup to prevent data loss, ensuring the security and accessibility of the data set; Step S4: feature extraction, using a feature extraction method to extract main dynamic features from continuous features, the feature extraction of step S4 is specifically as follows: Extract continuous features, including heart rate, blood pressure, and respiratory rate. Use principal component analysis to extract and retain the dominant dynamic features of the data, ignore dynamic features with small eigenvalues, reduce the dimension of the features, and retain the main dynamic data in the data. Step S5: Model construction, using the extracted features to build a psychological state assessment model based on machine learning, and optimizing the model through a validation set. The model construction of step S5 is specifically as follows: S51. Data preparation: Use the features after feature extraction and the encoded psychological state as model input and output; S52, constructing an SVC model: selecting the RBF kernel function and setting the parameters C and γ, where C is a regularization parameter used to control the complexity of the model and the degree of penalty for misclassification, and γ defines the width and influence range of the kernel function, which is used to control the accuracy of the kernel function boundary division; S53, model training: use the training set data to train the SVC model and optimize it through the SMO algorithm; S54, Model validation and optimization: Use the validation set to evaluate and adjust model parameters; S55, Performance measurement: Ensure that the recall rate is ≥ 90%, the F1 score is ≥ 0.85, and ensure that the model meets the predetermined performance indicators on the validation set and the test set; Step S6: Psychological state assessment and result output, inputting the data of the individual to be assessed into the model to obtain the psychological state assessment result; Step S7: Update and optimize system performance, regularly evaluate model performance and make adjustments and optimizations to ensure that model accuracy continues to meet standards.

2. A method for evaluating a psychological state based on feature extraction as claimed in claim 1, characterized in that: The specific steps of step S1 are: S11. Determine the data collection objectives and scope, taking into account comprehensive questionnaires, wearable devices, smartphone app tracking, and social media data to obtain a large sample set containing physiological indicators, behavioral data, and language usage characteristics; S12. Clarify and refine the content of physiological indicators to be collected, select the key physiological indicators that need to be collected, such as heart rate, blood pressure, and respiratory rate, and select wearable devices to ensure that these physiological indicators can be accurately measured and recorded; S13. Determine the details of behavioral data collection and list the behavioral data that need to be tracked, covering daily activity, sleep patterns, and eating habits, using smart device programs or logs for participants to fill in; S14. Clarify the content of language usage characteristics collection and determine the language usage characteristics that need to be analyzed, including speech emotional tendency, speech attitude and expression mode; S15. Prepare data collection tools and equipment operating environment, and ensure a stable network environment; S16. Participants can correctly use the collection tools and understand the relevant requirements, explain the purpose of the collection, specific procedures, potential risks and privacy protection policies, and sign the informed consent form; S17. Regularly monitor progress and data quality to ensure the accuracy and completeness of the collected data; S18. Perform preliminary processing and verification on the collected raw data, remove invalid and erroneous information, and ensure data quality; S19. Back up data regularly, save it in encrypted form, and develop a data recovery plan to prevent the risk of data loss.

3. A method for evaluating a psychological state based on feature extraction as claimed in claim 1, characterized in that: The specific steps of step S2 are: S21. Preliminary data screening: quickly review the raw data and mark unreasonable or abnormal records; S22. Formulate screening rules: Determine the reasonable range, frequency and logical association rules of data based on normal behavior and language habits; S23, Automated screening: Use programming technology to automatically filter unreasonable data according to rules; S24, Manual review: Detailed inspection of automated screening results to ensure data accuracy; S25. Eliminate malicious data: Combine multiple information sources to identify and remove forged or malicious data.

4. A method for evaluating a psychological state based on feature extraction as claimed in claim 1, characterized in that: The step S6 is specifically as follows: The model receives the physiological indicators, behavioral data, and language usage characteristics of the person to be evaluated as input, and outputs the corresponding psychological state and its confidence, where the psychological state with high confidence is regarded as the main psychological characteristic, while the psychological state with low confidence is regarded as the secondary psychological characteristic.

5. A method for evaluating a psychological state based on feature extraction as claimed in claim 1, characterized in that: The specific steps of step S7 are: S71. Evaluate the model regularly by setting time intervals and using recall and F1 score as key evaluation metrics. S72. Monitor the accuracy of the model and initiate the optimization process when performance degradation or model failure is detected and the performance is below the standard; S73. As time goes by, more new individual data is collected, and data is re-collected and features are extracted to adapt to the dynamic changes of the current application environment; S74. Retrain the model using the optimized features and parameters and test it on an independent validation set. S75. To cope with the changes in data distribution and the dynamic adjustment of evaluation requirements, repeat the above optimization steps in a cycle of 3 months.

6. A method for evaluating a psychological state based on feature extraction as claimed in claim 1, characterized in that: The principal component analysis method is specifically as follows: By standardizing the original continuous feature data, the influence caused by the differences in dimensions and magnitudes between different features can be eliminated; By calculating the covariance matrix of the data, its eigenvalues ​​and eigenvectors are obtained; Select the principal components by the size of the eigenvalues, and select the first few principal components with large eigenvalues. These principal components can retain most of the dynamic information of the original data; By projecting the original data onto the selected principal components, the reduced-dimensional data are obtained, which serve as the input for subsequent model training and psychological state assessment.

7. A method for evaluating a psychological state based on feature extraction as claimed in claim 4, characterized in that: The confidence index is specifically: The confidence value is between 0 and 1, where the closer the value is to 1, the more important the model's judgment of the psychological state is, and vice versa; When using deep learning model evaluation, a confidence threshold is set. When the confidence of a certain psychological state exceeds this threshold, the model distinguishes between primary and secondary psychological states based on the size of the confidence.

Images