A student psychological state recognition intelligent screening method and system

Abstract

The application discloses a student psychological state recognition intelligent screening method and system, and belongs to the technical field of data analysis. Student psychological log data in a continuous time period is acquired and standardized and cleaned, emotional fluctuation amplitude features and context correlation features are extracted, an emotional coherence index is calculated based on the two features, and the complexity of emotional evolution is quantified. Further, the classification accuracy of a traditional model is evaluated in different emotional coherence index intervals. If the accuracy is lower than a preset threshold in a high complexity interval, the data input quantity reflecting time sequence features is dynamically adjusted, and the adjusted model is iteratively trained and optimized. The application can significantly improve the detection accuracy and intervention timeliness of a psychological state recognition system in a complex emotional change scenario, effectively reduces false negatives, and improves the intelligent and accurate level of psychological health risk screening.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data analysis, in particular to a student psychological state recognition intelligent screening method and system. BACKGROUND

[0002] Student psychological state recognition intelligent screening refers to using artificial intelligence technology (such as big data analysis, machine learning, natural language processing, etc.) to automatically detect and analyze the mental health status of students through questionnaires, voices, texts, behavior data and other ways, to identify potential psychological problem risks in time, to help schools and professionals for early intervention and scientific management, so as to more effectively promote the development of students' mental health.

[0003] The prior art has the following shortcomings:

[0004] In the student psychological state long-term tracking system, if a traditional machine learning model (such as support vector machine SVM) is used to process the psychological log text of a continuous week, since such a model cannot model the complex time sequence and context dependence relationship, the system can only make static decisions based on local short text, ignoring the coherent trend of emotional changes. As a result, although the classification accuracy of the system is high on the surface, a large number of potential psychological crises cannot be identified in time, resulting in serious false negatives, reducing the actual intervention effect, and easily causing the serious consequences of missing long-term emotional deterioration. SUMMARY

[0005] The purpose of the present application is to provide a student psychological state recognition intelligent screening method and system to solve the problems in the background art.

[0006] In order to achieve the above purpose, the present application provides the following technical scheme: a student psychological state recognition intelligent screening method, comprising:

[0007] Obtaining student psychological log data in a continuous time period, and cleaning and processing the psychological log data to form a standardized text data set;

[0008] Extracting emotional fluctuation amplitude features and context correlation features from the text data set;

[0009] Calculating an emotional coherence index based on the emotional fluctuation amplitude features and the context correlation features, the emotional coherence index being used to represent the complexity of emotional evolution;

[0010] Classifying and predicting the student psychological state based on the trained traditional machine learning model, and respectively counting the classification accuracy in different emotional coherence index intervals;

[0011] determine whether the classification accuracy of the traditional machine learning model in the high emotional coherence index interval is lower than a preset accuracy threshold, and if so, determine that the model lacks the ability to capture context features of time-series psychological data;

[0012] When it is determined that the model accuracy is insufficient, the amount of data input reflecting time series features is dynamically adjusted, and the adjusted traditional machine learning model is iteratively trained and evaluated.

[0013] Preferably, the method for obtaining the emotion change amplitude value is: for each log text, an emotion score is generated by applying an emotion analysis model, denoted as , the single emotion change amplitude is calculated according to the absolute value of the difference between the emotion scores of two consecutive time nodes , and the expression is: ; the average emotion change amplitude in a continuous time period is calculated, and the calculation expression is: ; wherein N is the total number of time nodes, and the emotion change amplitude value ME is calculated, and the calculation expression is: .

[0014] Preferably, the method for obtaining the text association similarity is: the log text of each time node is converted into a fixed-dimensional vector representation by an embedding model, denoted as , the cosine similarity between the text vectors of two consecutive time nodes is calculated as the single text association similarity , and the calculation expression is: ; wherein: is the vector norm, and the text association similarity is calculated: ; M is the total number of time points for obtaining text vectors, , and the text association similarity is used to represent the context coherence degree.

[0015] Preferably, the emotional coherence index is calculated based on the emotion fluctuation amplitude feature and the context association feature, and the emotional coherence index is used to represent the complexity of emotional evolution, specifically including:

[0016] The emotion change amplitude value and the text association similarity are normalized to be between 0 and 1, and the emotional coherence index is calculated according to the normalized emotion change amplitude value and the text association similarity.

[0017] Preferably, the byte comparison bias index and the check value comparison anomaly index are normalized to be between 0 and 1, and the normalized byte comparison bias index and the check value comparison anomaly index are weighted and averaged to obtain the accuracy score of the current message parsing.

[0018] Preferably, the student psychological state is classified and predicted based on the trained traditional machine learning model, and the classification accuracy is respectively counted in different emotion coherence index intervals, specifically including:

[0019] A data set with labeled psychological state categories is used to train a traditional machine learning model; the input features include emotion scores, emotional tendencies, and behavior pattern indicators, and the output is the corresponding psychological state category label;

[0020] The psychological log data of the student to be detected is input into the trained model to automatically classify and predict the psychological state, and the predicted label of each student in each time period is obtained;

[0021] For each student, the emotion coherence index ECI is calculated based on the psychological log of the continuous time period;

[0030]

[0022] All students are divided into different intervals according to their ECI values, including: low complexity interval: ECI <0.3; medium complexity interval: 0.3≤ECI <0.6; high complexity interval: ECI≥0.6; each interval contains a number of student samples;

[0023] In each ECI interval, the classification accuracy of the model is calculated respectively:

[0024] Define whether the prediction of each sample is correct;

[0025] The number of correctly predicted samples in each interval is counted ;

[0026] The total number of samples in each interval is counted ;

[0027] The interval accuracy is calculated , the expression is: .

[0028] Preferably, it is judged whether the classification accuracy of the traditional machine learning model in the high emotion coherence index interval is lower than the preset accuracy threshold, and if it is lower, it is determined that the model lacks the ability to capture the context features of the time-series psychological data, specifically including:

[0029] A preset classification accuracy threshold θ, in all student samples, a sample set with emotion coherence index ECI≥0.6 is selected, denoted as high ECI interval data set Dhigh-ECI;

[0030] In Dhigh-ECI, the actual accuracy of the traditional machine learning model for psychological state classification prediction is counted , the calculation formula is: ; wherein: represents the number of samples correctly classified in the high ECI interval, represents the total number of samples in the high ECI interval;

[0031] comparing the classification accuracy in the high ECI interval with a preset threshold θ: if ≥ θ, it is considered that the classification ability of the model under complex emotional evolution meets the standard; if < θ, it is determined that the model is insufficient in capturing the context features of the time-series psychological data.

[0032] Preferably, when it is determined that the model accuracy is insufficient, the amount of data input reflecting the time series features is dynamically adjusted, specifically including:

[0033] applying DTW calculation to the continuous emotional score sequence or text emotional tendency sequence of each student, and dynamically adjusting the input features according to the degree of change of the DTW features:

[0034] if the DTW distance is greater than a preset distance threshold, the data input dimension is increased;

[0035] if the DTW distance is less than or equal to the preset distance threshold, the input amount is reduced;

[0036] combining the new time series features adjusted based on DTW with the original static features to form a new training data set;

[0037] retraining the traditional machine learning model using the updated data set.

[0038] The present application also provides a student psychological state recognition intelligent screening system, comprising a data acquisition and preprocessing module, a feature extraction module, an emotional coherence calculation module, a psychological state recognition and performance evaluation module, an accuracy detection and insufficiency judgment module, and a dynamic adjustment and iterative optimization module.

[0039] The data acquisition and preprocessing module: acquires student psychological log data in a continuous time period, and cleans and processes the psychological log data to form a standardized text data set;

[0040] The feature extraction module: extracts emotional fluctuation amplitude features and context correlation features from the text data set;

[0041] The emotional coherence calculation module: calculates an emotional coherence index based on the emotional fluctuation amplitude features and the context correlation features, the emotional coherence index being used to represent the complexity of emotional evolution;

[0042] The psychological state recognition and performance evaluation module: classifies and predicts the student psychological state based on the trained traditional machine learning model, and respectively calculates the classification accuracy in different emotional coherence index intervals;

[0043] An accuracy detection and deficiency judgment module is configured to judge whether the classification accuracy of the traditional machine learning model in the high emotional coherence index interval is lower than a preset accuracy threshold, and if so, determine that the model has insufficient ability to capture context features of time-series psychological data;

[0044] A dynamic adjustment and iterative optimization module is configured to dynamically adjust the amount of data input reflecting time-series features and iteratively train and evaluate the adjusted traditional machine learning model when it is determined that the model accuracy is insufficient.

[0045] In the above technical solution, the present application provides the following technical effects and advantages:

[0046] 1. The present application introduces the emotional fluctuation amplitude feature and the context correlation feature, innovatively proposes the emotional coherence index as an important indicator for measuring the complexity of emotional evolution, and dynamically detects the performance bottleneck of the traditional machine learning model in processing complex time-series psychological data by combining the classification accuracy in the emotional coherence index interval. When it is detected that the accuracy of the model in high complexity emotional evolution samples is insufficient, the present application uses the dynamic time warping algorithm to intelligently adjust the input amount of time-series features, and iteratively optimizes the traditional model, thereby effectively improving the sensitivity and accuracy of the psychological state recognition system to potential psychological crisis, significantly reducing the false negative rate, and enhancing the early warning ability and practical value of psychological health screening.

[0047] 2. The student psychological state recognition intelligent screening system proposed by the present application integrates the data acquisition and preprocessing module, the feature extraction module, the emotional coherence calculation module, the psychological state recognition and performance evaluation module, the accuracy detection and deficiency judgment module, and the dynamic adjustment and iterative optimization module, and can dynamically and adaptively adjust during data interaction and frame analysis, improve the accuracy and stability of data processing, and ensure the integrity and reliability of the student psychological state data transmission process. Overall, the present application not only makes up for the defects of the existing system in modeling the dynamic evolution of emotions, but also improves the intelligent level, dynamic adaptability and security of the psychological state intelligent screening system through system cascade optimization. BRIEF DESCRIPTION OF DRAWINGS

[0048] In order to more clearly illustrate the technical solutions in the embodiments or the prior art, the drawings needed in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments described in the present application, and other drawings can also be obtained by those skilled in the art based on these drawings.

[0049] Figure 1 The method mind map of the present application.

[0050] Figure 2The system module mind map of the present application. DETAILED DESCRIPTION

[0051] For the purpose, technical solutions and advantages of the embodiments of the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0052] Embodiment 1, please refer to Figure 1 The student psychological state recognition intelligent screening method described in the embodiment includes:

[0053] Obtain student psychological log data in a continuous time period, and clean and process the psychological log data to form a standardized text data set;

[0054] Extract emotion fluctuation amplitude features and context correlation features from the text data set;

[0055] Calculate an emotion coherence index based on the emotion fluctuation amplitude features and the context correlation features, the emotion coherence index being used to represent the complexity of emotion evolution;

[0056] Classify and predict the student psychological state based on the trained traditional machine learning model, and respectively count the classification accuracy in different emotion coherence index intervals;

[0057] Determine whether the classification accuracy of the traditional machine learning model in the high emotion coherence index interval is lower than a preset accuracy threshold, and if it is lower, determine that the model has insufficient ability to capture the context features of the time series psychological data;

[0058] When it is determined that the model accuracy is insufficient, dynamically adjust the data input amount reflecting the time series features, and iteratively train and evaluate the adjusted traditional machine learning model.

[0059] Set a fixed data collection period, for example, in units of days, continuously collect student psychological log data within 7 days, 14 days or 30 days. Each data point contains timestamp information, ensuring that the log has a clear time sequence.

[0060] Collect student self-reported psychological logs through designated channels, such as: online psychological diary systems (web pages, applets); mobile application psychological recording modules; campus psychological health management platforms.

[0061] Standardize the submission of content, and require the use of natural language to write psychological feelings, daily emotions, stress experiences, and important event descriptions. Set a minimum word count requirement (e.g., no less than 100 words per daily log) to ensure sufficient data volume.

[0062] Store the log data uniformly as a standard text format (e.g., UTF-8 encoded txt or json documents). Bind the text content with the corresponding timestamp to generate structured data units;

[0063] Remove the following invalid logs: empty text or insufficient content (e.g., fewer than 30 characters); obviously irrelevant content (e.g., only contains emojis, a large number of garbled characters, or advertising information); non-natural language content (e.g., pure numbers or meaningless letter strings).

[0064] Segmentation and basic processing (for Chinese / English respectively): For Chinese, use the jieba or BERT tokenizer to segment sentences into meaningful words or phrases. For English, use standard natural language tools (e.g., NLTK, spaCy) for Tokenization. Remove stop words (e.g., "de", "and", "the", and other words that have no actual emotional significance).

[0065] Calculate the basic statistical features of each log text (e.g., word count, sentiment score, and sentence count). Detect extreme outliers in the feature values (e.g., word count far beyond the normal range or extreme emotion scores) for manual review or automatic removal.

[0066] Align the text logs by day, hour, or other specified granularity, and fill in missing time nodes (e.g., fill in "no data" for dates without logs).

[0067] Form a unified data table or data packet format, and desensitize student personal information (e.g., name, student ID) using anonymous ID identification. Ensure that the log data complies with data protection regulations.

[0068] Extract emotion fluctuation amplitude features and context correlation features from the text data set. The emotion fluctuation amplitude feature is calculated based on the emotion score at each time node, and the context correlation feature is calculated based on the text vectorization processing of adjacent time node log texts.

[0069] The method for obtaining the emotion change amplitude value is as follows: For each log text, apply a sentiment analysis model (e.g., based on an emotion dictionary or a deep sentiment classification model) to generate an emotion score, denoted as where t represents the time node (e.g., each day). The single emotion change amplitude is calculated based on the absolute value of the difference between the emotion scores of two consecutive time nodes, denoted as , and the expression is: ; the average emotional change amplitude in the continuous time period is calculated, and the calculation expression is: ; wherein N is the total number of time nodes, and the emotional change amplitude value ME (measuring emotional fluctuation discreteness) is calculated, and the calculation expression is: .

[0070] The greater the emotional change amplitude value, the more intense the emotional fluctuation of the student in the continuous time period, the greater the emotional state fluctuation, and there is a significant emotional instability phenomenon. This usually indicates that the psychological state is complex, and the student may have experienced a variety of different emotional experiences, such as from anxiety to joy, and then to depression, and the emotional evolution process is relatively complex and unpredictable, which indicates that further attention is needed to the risk of potential psychological problems.

[0071] On the contrary, the smaller the emotional change amplitude value, the more gentle the change of the student's emotional score between time nodes, the more stable the emotional state, and the evolution process is simple and controllable. In this case, the student's emotional state presents a continuous and consistent feature, whether it is continuously positive or continuously negative, and the emotional evolution complexity is low, but if it is continuously in a negative emotion, it may also indicate a long-term psychological health problem, which needs to be further analyzed in combination with other characteristics.

[0072] The method for obtaining the text association similarity is: the log text of each time node is converted into a fixed-dimensional vector representation through an embedding model (such as BERT, Word2Vec, Sentence-BERT), denoted as . The cosine similarity between the text vectors of two consecutive time nodes is calculated as the single text association similarity , and the calculation expression is: ; wherein: is the vector norm, and the text association similarity is calculated as: ; M is the total number of text vector acquisition time points, , and the text association similarity is used to represent the degree of context coherence.

[0073] The greater the text association similarity, the more consistent the student's psychological log content between adjacent time nodes in semantics, the stronger the coherence of emotional state and psychological expression, and the relatively stable and orderly evolution process. This usually indicates that the student's emotional change is small, and the psychological experience is continuous, and the student may be in a stable emotional state, whether positive or negative, the emotional trajectory is relatively single and predictable, and the evolution complexity is low.

[0074] On the contrary, the smaller the text correlation similarity is, the greater the semantic difference between the continuous time node log texts is, the obvious content jump is, the frequent emotional state change is or the great change in psychological expression style is. In this case, the student may experience emotional rupture, psychological state mutation or complex evolution process of multiple emotions interlaced, the emotional coherence is weak, the evolution complexity is high, and the risk of abnormal psychological fluctuation needs special attention.

[0075] The emotion coherence index is calculated based on the emotion fluctuation amplitude feature and the context correlation feature, and the emotion coherence index is used to represent the evolution complexity of emotion, and specifically includes:

[0076] The emotion change amplitude value and the text correlation similarity are normalized to be between 0 and 1, and the emotion coherence index is calculated according to the normalized emotion change amplitude value and the text correlation similarity.

[0077] For example, the emotion coherence index can be calculated by the following formula, and the calculation expression is: In the formula, is the emotion coherence index, is the emotion change amplitude value, is the text correlation similarity, is the weight coefficient of the emotion change amplitude value and the text correlation similarity (which can be optimized according to experimental experience or machine learning), and are all greater than 0.

[0078] The student psychological state is classified and predicted based on the trained traditional machine learning model, and the classification accuracy is respectively counted in different emotion coherence index intervals, and specifically includes:

[0079] The data set with labeled psychological state categories (such as healthy, anxious and depressed) is used, and a traditional machine learning model (such as support vector machine SVM, random forest RF, XGBoost, etc.) is used for model training. The input features can include emotion score, emotional tendency, behavior pattern index, etc., and the output is the corresponding psychological state category label.

[0080] The psychological log data (after cleaning and feature extraction processing) of the student to be detected is input into the trained model to automatically classify and predict the psychological state. The predicted label of each student in each time period is obtained.

[0081] For each student, the emotion coherence index (ECI) is calculated based on the continuous time period psychological log according to the established method (comprehensive of emotion change amplitude value and text correlation similarity).

[0082] All students are divided into different intervals according to their ECI values, for example:

[0083] Low complexity interval: ECI < 0.3;

[0084] Medium complexity interval: 0.3 ≤ ECI < 0.6;

[0085] High complexity interval: ECI ≥ 0.6;

[0086] Each interval contains a number of student samples.

[0087] In each ECI interval, the classification accuracy of the model is calculated respectively, and the specific method is as follows:

[0088] Define whether the prediction of each sample is correct;

[0089] Statistical number of samples in each interval that are correctly predicted ;

[0090] Statistical number of total samples in each interval ;

[0091] Calculate the interval accuracy , the expression is: .

[0092] By comparing the classification accuracy in different emotion coherence index intervals, the performance of the traditional machine learning model under different complexity of emotion evolution can be evaluated. If the classification accuracy in the high ECI interval decreases significantly, it can be further inferred that the model lacks the ability to capture complex emotion changes.

[0093] Determine whether the classification accuracy of the traditional machine learning model in the high emotion coherence index interval is lower than the preset accuracy threshold, and if it is lower, it is determined that the model lacks the ability to capture the context features of the time-series psychological data, which specifically includes:

[0094] The preset classification accuracy threshold θ, for example, 70%, 75% or other standards set according to business needs. The threshold is used to determine whether the performance of the model under a specific emotion complexity meets the standard.

[0095] In all student samples, filter out the sample set with emotion coherence index ECI higher than the set limit (such as ECI ≥ 0.6), denoted as high ECI interval dataset Dhigh-ECI.

[0096] In Dhigh-ECI, the actual accuracy of the traditional machine learning model for psychological state classification prediction is calculated , and the calculation formula is: ; Wherein: represents the number of samples correctly classified in the high ECI interval, represents the total number of samples in the high ECI interval.

[0097] Compare the classification accuracy in the high ECI interval with the preset threshold θ:

[0098] If ≥ θ, it is considered that the classification ability of the model under complex emotional evolution meets the standard;

[0099] If < θ, it is determined that the model's ability to capture the context features of time-series psychological data is insufficient.

[0100] When it is determined that the model accuracy is insufficient, the amount of data input reflecting the time series features is dynamically adjusted, and the adjusted traditional machine learning model is iteratively trained and evaluated, including:

[0101] Determine whether the classification accuracy of the traditional machine learning model in the high emotional coherence index interval is lower than the preset threshold. If it is lower, it is determined that the model's ability to capture time-series features is insufficient, and the dynamic adjustment process is entered.

[0102] Select the Dynamic Time Warping (DTW) algorithm as the basic algorithm for dynamic adjustment. DTW can align two time series non-linearly and measure their similarity, even if the speed changes.

[0103] Apply DTW calculation to each student's continuous emotional score sequence or text sentiment tendency sequence. Based on the DTW distance feature, reflect the overall shape and evolution pattern of the student's emotional change trajectory.

[0104] Dynamically adjust the input features according to the degree of change in the DTW feature:

[0105] If the DTW distance is greater than the preset distance threshold (indicating that the emotional change is severe), increase the data input dimension, such as introducing the emotional sequence within a longer time window;

[0106] If the DTW distance is less than or equal to the preset distance threshold (indicating that the emotional change is smooth), appropriately reduce the input amount, retain the core change segment, and improve data processing efficiency.

[0107] Combine the new time series features based on DTW adjustment with the original static features (such as single-point emotional scores, text similarity) to form a new training data set.

[0108] Use the updated data set to retrain the traditional machine learning model (such as SVM, random forest). In the training process, perform hyperparameter optimization, such as adjusting the regularization coefficient, decision tree depth, etc.

[0109] In the validation set and high emotional coherence index interval, the classification accuracy of the adjusted model is re-evaluated. If it reaches or exceeds the preset accuracy threshold, optimization is completed; otherwise, the feature input quantity is adjusted or another deep time series model (such as LSTM) is switched to.

[0110] The embodiment provides a student psychological state recognition intelligent screening method. In view of the problem that a traditional machine learning model has insufficient capture ability when processing psychological log data with time sequence and context dependence characteristics, student psychological log data in a continuous time period is collected, emotion fluctuation amplitude features and context correlation features are extracted, and an emotional coherence index is comprehensively calculated to represent the complexity of emotional evolution. Further, the classification accuracy of the traditional model is evaluated in different emotional coherence index intervals. When the accuracy in the high emotional coherence interval is lower than the preset threshold, a dynamic time warping (DTW) algorithm is used to dynamically adjust the data input quantity reflecting the time sequence characteristics, and the adjusted model is iteratively trained and evaluated, thereby effectively improving the accuracy and robustness of the psychological state recognition system under complex emotional evolution samples, and realizing intelligent and accurate psychological health screening.

[0111] Embodiment 2, please refer to Figure 2 The student psychological state recognition intelligent screening system described in the embodiment includes a data acquisition and preprocessing module, a feature extraction module, an emotional coherence calculation module, a psychological state recognition and performance evaluation module, an accuracy detection and deficiency judgment module, and a dynamic adjustment and iterative optimization module.

[0112] Data acquisition and preprocessing module: obtain student psychological log data in a continuous time period, and clean the psychological log data to form a standardized text data set;

[0113] Feature extraction module: extract emotion fluctuation amplitude features and context correlation features from the text data set;

[0114] Emotional coherence calculation module: calculate an emotional coherence index based on the emotion fluctuation amplitude features and the context correlation features, the emotional coherence index being used to represent the complexity of emotional evolution;

[0115] Psychological state recognition and performance evaluation module: classifies and predicts the student psychological state based on the trained traditional machine learning model, and respectively calculates the classification accuracy in different emotional coherence index intervals;

[0116] Accuracy detection and deficiency judgment module: judges whether the classification accuracy of the traditional machine learning model in the high emotional coherence index interval is lower than the preset accuracy threshold. If it is lower, it is determined that the model has insufficient capture ability for the context features of the time sequence psychological data;

[0117] Dynamic adjustment and iterative optimization module: when the accuracy of the model is determined to be insufficient, the dynamic adjustment reflects the amount of data input reflecting the time series characteristics, and the adjusted traditional machine learning model is iteratively trained and evaluated.

[0118] The above formulas are dimensionless numerical calculations, and the formulas are obtained by collecting a large amount of data to simulate the current real situation, and the preset parameters in the formula are set by the person skilled in the art according to the actual situation.

[0119] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be covered within the protection scope of the present application.

Claims

1. A student mental state recognition intelligent screening method, characterized in that: The method comprises the following steps: obtain student psychological diary data in a continuous time period, and clean and process the psychological diary data to form a standardized text data set; extract emotion fluctuation amplitude features and context correlation features from the standardized text data set; calculate an emotion continuity index based on the emotion fluctuation amplitude features and the context correlation features, the emotion continuity index being used to represent the complexity of emotion evolution, specifically comprising: The method for obtaining the emotion fluctuation amplitude feature is: for each log text, an emotion analysis model is applied to generate an emotion score, denoted as Where t represents a time node, the single emotion change amplitude is calculated according to the absolute value of the difference between the emotion scores of two consecutive time nodes , and the expression is: The average emotion change amplitude in a continuous time period is calculated, and the calculation expression is: ; wherein N is the total number of time nodes, and the emotion change amplitude value ME is calculated, and the calculation expression is: ; The acquisition method of the context relevance feature is: converting the log text of each time node into a fixed-dimensional vector representation through an embedding model, denoted as , calculating the cosine similarity between the text vectors of two consecutive time nodes as the single text association similarity , and the calculation expression is: ; wherein: is the respective vector length, and the text association similarity is calculated as ; M is the total number of text vector acquisition time points, is the text association similarity, which is used to represent the context coherence degree; calculate the emotion continuity index based on the normalized emotion fluctuation amplitude features and the context correlation features; classify and predict the student psychological state based on the trained traditional machine learning model, and respectively calculate the classification accuracy in different emotion continuity index intervals; determine whether the classification accuracy of the traditional machine learning model in the high emotion continuity index interval is lower than a preset accuracy threshold, and if it is lower, determine that the model has insufficient ability to capture the context features of time-series psychological data; when it is determined that the model accuracy is insufficient, dynamically adjust the data input quantity reflecting the time series features, and iteratively train and evaluate the adjusted traditional machine learning model, specifically comprising: apply DTW calculation to the continuous emotion score sequence or text sentiment tendency sequence of each student, reflect the overall shape and evolution mode of the student emotion change trajectory based on the DTW distance features, and dynamically adjust the input features according to the change degree of the DTW features: if the DTW distance is greater than a preset distance threshold, increase the data input dimension; if the DTW distance is less than or equal to the preset distance threshold, reduce the input quantity; combine the new time series features adjusted based on DTW with the original static features to form a new training data set; retrain the traditional machine learning model using the updated data set.

2. The student mental state recognition intelligent screening method according to claim 1, characterized in that: classify and predict the student psychological state based on the trained traditional machine learning model, and respectively calculate the classification accuracy in different emotion continuity index intervals, specifically comprising: use the data set with labeled psychological state categories to train the traditional machine learning model; the input features include emotion scores, sentiment tendencies, and behavior pattern indicators, and the output is the corresponding psychological state category label; input the psychological diary data of the student to be detected into the trained model to automatically classify and predict the psychological state, and obtain the prediction label of each student in each time period; for each student, calculate the emotion continuity index ECI based on the psychological diary in the continuous time period; divide all students into different intervals according to their ECI values, including: low complexity interval: ECI<0.3; medium complexity interval: 0.3≤ECI<0.6; high complexity interval: ECI≥0.6; each interval contains a number of student samples; calculate the classification accuracy of the model in each ECI interval: define whether the prediction of each sample is correct; counting the number of samples in each interval that are correctly predicted ; counting the total number of samples in each interval ; Computing interval accuracy , where the expression is: .

3. The student mental state recognition intelligent screening method according to claim 2, characterized in that: determine whether the classification accuracy of the traditional machine learning model in the high emotion continuity index interval is lower than the preset accuracy threshold, and if it is lower, determine that the model has insufficient ability to capture the context features of time-series psychological data, specifically comprising: A preset classification accuracy threshold θ, in all student samples, a sample set with an emotional coherence index ECI ≥ 0.6 is screened out, and is recorded as a high ECI interval data set Dhigh-ECI; In Dhigh-ECI, the actual accuracy of the traditional machine learning model in predicting the classification of the psychological state , the calculation formula of which is: ; wherein: represents the number of samples classified correctly in the high ECI interval, represents the total number of samples in the high ECI interval; Compare the classification accuracy in the high ECI interval with the preset threshold θ: if ≥ θ, it is considered that the classification ability of the model under complex emotional evolution meets the standard; if < θ, it is determined that the model is insufficient in capturing the context features of the time-series psychological data.

4. A student mental state recognition intelligent screening system for implementing the student mental state recognition intelligent screening method of any one of claims 1-3, characterized in that: It comprises a data acquisition and preprocessing module, a feature extraction module, an emotional coherence calculation module, a psychological state recognition and performance evaluation module, an accuracy detection and deficiency judgment module, and a dynamic adjustment and iterative optimization module. The data acquisition and preprocessing module: acquires student psychological log data in a continuous time period, and cleans and processes the psychological log data to form a standardized text data set; The feature extraction module: extracts emotional fluctuation amplitude features and context correlation features from the standardized text data set; The emotional coherence calculation module: calculates an emotional coherence index based on the emotional fluctuation amplitude features and the context correlation features, and the emotional coherence index is used to represent the complexity of emotional evolution; The psychological state recognition and performance evaluation module: classifies and predicts the student psychological state based on the trained traditional machine learning model, and respectively calculates the classification accuracy in different emotional coherence index intervals; The accuracy detection and deficiency judgment module: judges whether the classification accuracy of the traditional machine learning model in the high emotional coherence index interval is lower than the preset accuracy threshold, and if it is lower, it is determined that the model has insufficient ability to capture the context features of the time series psychological data; The dynamic adjustment and iterative optimization module: when it is determined that the model accuracy is insufficient, the amount of data input reflecting the time series features is dynamically adjusted, and the adjusted traditional machine learning model is iteratively trained and evaluated.

Images