Mental health education auxiliary system based on big data

Abstract

The invention discloses a psychological health education auxiliary system based on big data, and relates to the technical field of psychological health education, the psychological health education auxiliary system uses a camera to collect eye movement data, combines CNN to extract static features and LSTM to capture time sequence changes, improves anxiety detection objectivity, combines embedded online learning, short video watching or classroom scenes, and improves anxiety detection accuracy. Daily data collection is carried out, the anxiety trend is continuously tracked by adopting a dynamic analysis method, and the change of the psychological state is monitored and evaluated in real time; the psychological intervention content can be dynamically adjusted based on anxiety scores and personalized feedback strategies, for example, relaxation training, psychological adjustment videos or emotion evaluation questionnaires are pushed, so that psychological health education is more personalized, and invalid or generalized psychological education information is reduced; in addition, a reinforcement learning algorithm is introduced, recommendation content is continuously optimized through interaction data of a tested object, the intervention effect is improved, and psychological education resources can adapt to dynamic changes of individual psychological states.

Description

Technical Field

[0001] The present invention relates to the technical field of mental health education, and particularly to a mental health education assistance system based on big data. Background Art

[0002] In mental health education in universities, hospitals, etc., methods such as psychological assessment questionnaires, psychological counseling, and group counseling are mostly used to guide students to improve their psychological cognition and adjustment abilities. However, psychological problems such as anxiety are latent, and many students will not actively seek help even if they have emotional distress; and questionnaire assessment highly depends on subjective feedback, and many students tend to choose normal answers, resulting in the questionnaire being unable to effectively reveal anxiety symptoms.

[0003] With the continuous development of short videos, gradually, the method of combining short videos with online mental health education is adopted to assist mental health education, which to a certain extent increases the effect of mental health education. However, its content is for a wide range of people and lacks an individualized identification and intervention mechanism; even if students watch the same mental health content, due to different personal mental states, their understandings and reactions will also vary. The existing mode cannot capture these individualized emotional fluctuations; the subconscious behaviors of anxious individuals during learning or watching short videos, such as eye movement patterns, attention distribution, etc., can largely reveal their mental states. Anxious people may show characteristics such as higher eye movement frequencies and shorter fixation durations when watching stress-related content, while traditional mental health education methods cannot effectively identify these subtle signals and cannot intervene in psychological problems in a timely manner.

[0004] It can be seen that how to break through the dependence on active help-seeking and subjective feedback and establish a non-intrusive dynamic detection mechanism based on objective data has become a key challenge in mental health education. Summary of the Invention

[0005] In view of the above existing problems, the present invention is proposed.

[0006] The present invention provides a mental health education assistance system based on big data to solve the problems that psychological assessment depends on subjective feedback, is difficult to dynamically track, short videos are popular but there is no individualized detection, and anxiety problems are easily overlooked.

[0007] To solve the above technical problems, the present invention provides the following technical solutions:

[0008] An embodiment of the present invention provides a mental health education assistance system based on big data, which includes,

[0009] An eye movement data acquisition module, configured to obtain eye movement data of a test subject when watching a video, an online course, or classroom learning during the mental health education process;

[0010] Anxiety detection module, which receives eye movement data and processes it through the anxiety detection model;

[0011] Anxiety score calculation module, used to receive the anxiety detection result of the anxiety detection module, calculate the anxiety score of the tested object based on the anxiety detection model, and classify the tested object according to different anxiety levels;

[0012] Personalized feedback generation module, which generates personalized feedback based on anxiety scores;

[0013] An interaction data recording module pushes personalized feedback to the tested object and records the interaction data of the tested object, wherein the interaction data includes viewing time, click behavior and active search behavior of the feedback content;

[0014] A dynamic adjustment module is used to receive the interaction data and input it into the anxiety detection model, dynamically analyze the anxiety trend of the subject, and adjust the personalized feedback content of the personalized feedback generation module according to the trend change;

[0015] Early warning module, when the anxiety trend of the dynamic adjustment module shows that the anxiety score of the subject continues to rise and reaches the preset threshold, the early warning mechanism is triggered and psychological intervention is carried out.

[0016] As a preferred solution of the mental health education auxiliary system based on big data described in the present invention, wherein: the eye movement data includes eye movement trajectory, pupil diameter change, blinking frequency and gaze dwell time;

[0017] The eye movement data acquisition module, based on camera equipment and combined with an eye movement tracking algorithm, performs real-time analysis of the eye movement characteristics of the subject; and can be imperceptibly embedded in online learning, short video viewing or classroom scenes.

[0018] As a preferred solution of the mental health education auxiliary system based on big data described in the present invention, the anxiety detection model is based on the fusion architecture of convolutional neural network CNN and long short-term memory network LSTM, wherein CNN is used to extract static features and LSTM is used to capture time series changes;

[0019] The anxiety detection module includes a data preprocessing unit for denoising and standardizing the eye movement data, and performing feature matching in combination with a preset anxiety behavior database.

[0020] As a preferred solution of the big data-based mental health education auxiliary system described in the present invention, the personalized feedback includes psychological adjustment video recommendations, relaxation training push notifications, emotion adjustment prompts and psychological assessment questionnaires;

[0021] The personalized feedback generation module combines the historical anxiety scores and interaction data of the subjects and uses a reinforcement learning algorithm to optimize the feedback strategy.

[0022] As a preferred solution of the big data-based mental health education auxiliary system described in the present invention, the early warning module adopts a multi-level threshold judgment strategy to automatically match psychological intervention measures for different anxiety levels, including anonymous psychological support recommendations, psychological counseling appointment suggestions and emergency intervention notifications, to notify psychological counselors to intervene.

[0023] As a preferred solution of the big data-based mental health education auxiliary system described in the present invention, the auxiliary method of the system is:

[0024] Step S1, collecting eye movement data of the subject when watching videos, online courses or classroom learning during mental health education;

[0025] Step S2, inputting the eye movement data into an anxiety detection model for anxiety detection;

[0026] Step S3, scoring the anxiety of the subjects based on the anxiety detection results of step S2, and classifying them according to different anxiety levels;

[0027] Step S4, generating personalized feedback based on the anxiety score, pushing it to the subject, and recording the interaction data of the subject;

[0028] Step S5, inputting the interactive data of step S4 into the anxiety detection model, dynamically analyzing the anxiety trend of the tested subject, and adjusting the personalized feedback content according to the trend change;

[0029] Step S6: When the anxiety trend of step S5 indicates that the anxiety score of the subject continues to increase and reaches a preset threshold, an early warning mechanism is triggered.

[0030] As a preferred solution of the mental health education auxiliary system based on big data described in the present invention, in step S2, the step of constructing the anxiety detection model is:

[0031] Preprocess the eye movement data. Assume that the eye movement data contains N sets of samples, each of which consists of M-dimensional features: X = {x1, x2, …, x N},

[0032] Among them, X represents the complete eye movement dataset, which contains N groups of data, x n represents the eye movement feature vector of the nth sample, with a dimension of M, where N represents the total number of data samples and M represents the feature dimension of each sample, including eye movement trajectory, pupil diameter change, blink frequency, and gaze dwell time;

[0033] Normalized processing, the formula is:

[0034]

[0035] where x' n is the normalized eigenvector, μ is the eigen mean, and σ is the eigen standard deviation;

[0036] Taking the normalized data X' = {x'1, x'2, …, x' N} as the input, static feature extraction is performed using CNN. The CNN consists of L convolutional layers, and the convolutional operation for each layer is:

[0037] h l = f(W l * h l-1 + b l ),

[0038] where h l is the feature map of the l-th layer of the CNN, W l is the weight matrix of the convolutional kernel of the l-th layer, * represents the convolutional operation, b l is the bias term, and f(·) uses the ReLU activation function.

[0039] Finally, the static feature H extracted by the CNN is obtained:

[0040] H = {h1, h2, …, h N},

[0041] where H is the static feature sequence extracted by the CNN, corresponding to the number of input samples N;

[0042] Time series feature modeling is performed using LSTM. Taking the feature H extracted by the CNN as the input of the LSTM to capture the dynamic change patterns in the time series. The LSTM uses a single-layer structure, and the calculation method is:

[0043] s n = LSTM(h n , s n-1 ),

[0044] where s n is the LSTM hidden state vector of the n-th sample, h n is the corresponding CNN feature vector, and s n-1 is the hidden state at the previous moment.

[0045] Finally, the feature S processed by the LSTM is obtained:

[0046] S = {s1, s2, …, s N},

[0047] where S is the dynamic feature sequence generated by the LSTM;

[0048] Input the features S extracted by LSTM into the fully connected layer, and calculate the anxiety detection probability through Softmax. The formula is:

[0049] p = Softmax(W s s N + b s ),

[0050] where p is the probability distribution of the anxiety state, W s is the weight matrix of the fully connected layer, and b s is the bias term of the fully connected layer.

[0051] Finally, the anxiety detection model outputs the anxiety state category C of the subject:[[]]

[0052] C = argmax(p),

[0053] where C is the predicted anxiety state category.

[0054] As a preferred solution of the mental health education assistance system based on big data according to the present invention, wherein: the step of performing anxiety scoring on the subject based on the anxiety detection result in step S2 and classifying according to different anxiety levels is

[0055] Based on the anxiety detection result C in step S2, combined with the feature vector S, calculate the scoring and classify according to different anxiety levels, and calculate the anxiety score. The formula is:

[0056] A = w1p + w2|S|,

[0057] where A is the anxiety score, the range is set between [0,1], w1 and w2 are scoring weight coefficients, satisfying w1 + w2 = 1, p is the anxiety state probability calculated in step S2, and |S| is the norm of the dynamic feature vector extracted by LSTM:

[0058]

[0059] where |S| represents the average intensity of the time series features calculated by LSTM, measuring the degree of change in eye movement data, and s n is the nth group of sample feature vectors calculated by LSTM in step S2.

[0060] Normalize the anxiety score. The formula is:

[0061]

[0062] where A' is the normalized anxiety score, the range is [0,1], A min , A maxThe minimum and maximum values of the anxiety scores in the data;

[0063] Classify by anxiety level. Based on the normalized anxiety score A', set the anxiety level thresholds:

[0064] Low anxiety: A' < τ1,

[0065] Medium anxiety: τ1 ≤ A' < τ2,

[0066] High anxiety: A' ≥ τ2,

[0067] Among them, τ1 and τ2 are empirical thresholds, set according to psychological assessment data and determined by the quantile method;

[0068] Finally, output the anxiety level L:

[0069] If A' < τ1, then L = Low anxiety,

[0070] If τ1 ≤ A' < τ2, then L = Medium anxiety,

[0071] If A' ≥ τ2, then L = High anxiety.

[0072] As a preferred solution of the mental health education assistance system based on big data described in the present invention, wherein: the step of inputting the interaction data in step S4 into the anxiety detection model, dynamically analyzing the anxiety trend of the tested object, and adjusting the personalized feedback content according to the trend change is,

[0073] Preprocess the interaction data, and define the interaction data set as D:

[0074] D = {d1, d2, …, d Q},

[0075] Among them, D represents the interaction data set of the tested object, d q represents the feature vector of the q-th interaction, including the viewing duration, click behavior, and active search behavior, and Q represents the time step of the interaction data,

[0076] Normalize the interaction data, and the formula is:

[0077]

[0078] Among them, d' q is the normalized interaction data, μ D is the mean of the interaction data, and σ D is the standard deviation;

[0079] Define the anxiety trend as a sequence of anxiety scores changing over time, denoted as T A :

[0080] TA = {A1, A2, …, A Q},

[0081] where T A is the sequence of anxiety scores changing with interaction data, and A q is the anxiety score corresponding to the q-th interaction, calculated using exponential smoothing:

[0082] A' q = αA q + (1 - α)A' q-1 ,

[0083] where A' q is the smoothed anxiety score trend value, α is the smoothing coefficient, with a range of 0 < α < 1, and A' q-1 is the smoothed anxiety score from the previous calculation.

[0084] Calculate the anxiety trend change rate, with the formula:

[0085] ΔA q = A' q - A' q-1 ,

[0086] where ΔA q represents the change amplitude of the anxiety trend;

[0087] Based on the anxiety trend change rate ΔA q perform personalized feedback adjustment, with the adjustment formula:

[0088] F q = F q-1 + βΔA q ,

[0089] where F q is the adjusted personalized feedback intensity, F q-1 is the previous personalized feedback content, and β is the feedback adjustment coefficient.

[0090] If the anxiety trend is rising, ΔA q > τ F , then increase the content of psychological regulation intervention.

[0091] If the anxiety trend is falling, ΔA q < -τ F , then reduce the intervention intensity.

[0092] As a preferred solution of the mental health education assistance system based on big data described in the present invention, wherein: the step of triggering the warning mechanism when the anxiety trend in step S5 indicates that the anxiety score of the subject under test continues to rise and reaches the preset threshold is

[0093] Calculate the change trend of the anxiety score, with the formula:

[0094]

[0095] where G q is the change trend of the anxiety score, K is the size of the observation window, that is, the average value of the score change rate is calculated within K interaction data, and ΔA q-k is the anxiety trend change rate defined in step S5.

[0096] If G q exceeds the preset anxiety growth threshold τ G , it is determined that the anxiety score is continuously rising:

[0097] G q > τ G ,

[0098] where τ G is the threshold for the change of the anxiety score, set based on statistical analysis.

[0099] Set the trigger warning mechanism, and the trigger condition of the warning mechanism is:

[0100] A' q ≥ τ W and G q > τ G ,

[0101] where A' q is the smoothed anxiety score, τ W is the anxiety score warning threshold, set as the quantile of high-anxiety users or specific clinical criteria, and τ G is the change threshold for the continuous rise of the anxiety score.

[0102] If this condition is met, the warning is triggered: W q = 1,

[0103] where W q indicates whether the warning is triggered currently. If it is equal to 1, the warning is triggered; if it is equal to 0, the warning is not triggered.

[0104] According to the anxiety score A' q and the anxiety trend G q of the test subject, determine the warning level:

[0105] If τ W ≤ A' q < τ W + δ1 and G q > τ G , then L W = Low-level warning,

[0106] If τ W+δ1 ≤ A' q <τ W +δ2 and G q >τ G , then L W = Intermediate warning,

[0107] If A' q ≥ τ W +δ2 and G q >τ G , then L W = High - level warning,

[0108] Among them, L W is the warning level, and δ1, δ2 are the boundary values of the warning levels of the anxiety score,

[0109] When W q = 1, different intervention strategies are adopted according to L W Take different intervention strategies.

[0110] The beneficial effects of the present invention are as follows: The present invention uses a camera to collect eye movement data, combines CNN to extract static features, and LSTM captures time - series changes to improve the objectivity of anxiety detection. It combines embedded online learning, short - video viewing or classroom scenarios for daily data collection, and uses a dynamic analysis method to continuously track the anxiety trend, and real - time monitors and evaluates the changes in mental state.

[0111] The present invention can dynamically adjust the content of psychological intervention based on anxiety scores and personalized feedback strategies, such as pushing relaxation training, psychological adjustment videos or emotion assessment questionnaires, making mental health education more personalized and reducing ineffective or generalized psychological education information; in addition, a reinforcement learning algorithm is introduced, and the recommended content is continuously optimized through the interaction data of the tested object to improve the intervention effect, so that mental education resources can adapt to the dynamic changes of individual mental states.

[0112] The present invention designs a hierarchical warning mechanism for individuals at high risk of anxiety, judges the severity of psychological problems through anxiety trend analysis, and automatically triggers different levels of psychological intervention strategies when the anxiety score continues to rise to a preset threshold, effectively reducing the cumulative risk of mental health problems.

[0113] In summary, the present invention can effectively make up for the problem of lag in identifying psychological problems in traditional solutions, improve the dynamic tracking ability and personalized intervention ability, and make the detection and intervention of anxiety problems more intelligent and efficient. Brief Description of the Drawings

[0114] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0115] Figure 1 It is a schematic framework diagram of the mental health education assistance system based on big data of the present invention. Specific embodiments

[0116] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following will make a detailed description of the specific embodiments of the present invention with reference to the accompanying drawings of the specification.

[0117] In the following description, many specific details are set forth to fully understand the present invention. However, the present invention can also be implemented in other ways different from those described herein. Those skilled in the art can make similar generalizations without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0118] Secondly, the so-called "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that can be included in at least one implementation manner of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that mutually excludes other embodiments.

[0119] Embodiment 1, referring to Figure 1 , this embodiment provides a mental health education assistance system based on big data, including:

[0120] An eye movement data acquisition module, which is used to obtain the eye movement data of the test subject when watching videos, online courses, or classroom learning during the mental health education process;

[0121] The eye movement data includes eye movement trajectories, pupil diameter changes, blink frequencies, and fixation times;

[0122] The eye movement data acquisition module, based on camera devices, combines eye movement tracking algorithms to perform real-time analysis on the eye movement characteristics of the test subject; and can be imperceptibly embedded in online learning, short video viewing, or classroom scenarios;

[0123] An anxiety detection module, which receives the eye movement data and processes it through an anxiety detection model;

[0124] The anxiety detection model is based on a fusion architecture of a convolutional neural network CNN and a long short-term memory network LSTM. Among them, CNN is used to extract static features, and LSTM is used to capture time series changes;

[0125] The anxiety detection module includes a data preprocessing unit for denoising and normalizing the eye movement data and performing feature matching in combination with a preset anxiety behavior database.

[0126] The anxiety score calculation module is configured to receive the anxiety detection result of the anxiety detection module, calculate the anxiety score of the subject to be tested based on the anxiety detection model, and classify it according to different anxiety levels.

[0127] The personalized feedback generation module generates personalized feedback according to the anxiety score.

[0128] The personalized feedback includes recommendations for psychological adjustment videos, relaxation training pushes, emotion regulation prompts, and psychological assessment questionnaires.

[0129] The personalized feedback generation module combines the historical anxiety scores and interaction data of the subject to be tested and uses a reinforcement learning algorithm to optimize the feedback strategy.

[0130] The interaction data recording module pushes personalized feedback to the subject to be tested and records the interaction data of the subject to be tested. The interaction data includes the viewing duration, click behavior, and active search behavior of the feedback content.

[0131] The dynamic adjustment module is configured to receive the interaction data, input it into the anxiety detection model, dynamically analyze the anxiety trend of the subject to be tested, and adjust the personalized feedback content of the personalized feedback generation module according to the trend change.

[0132] The early warning module triggers an early warning mechanism for psychological intervention when the anxiety trend of the dynamic adjustment module indicates that the anxiety score of the subject to be tested continues to rise and reaches a preset threshold.

[0133] The early warning module adopts a multi-level threshold judgment strategy and automatically matches psychological intervention measures for different anxiety levels, including anonymous psychological support recommendations, psychological counseling appointment suggestions, and emergency intervention notifications, and notifies psychological counselors to intervene.

[0134] This embodiment also provides an auxiliary method for the above-mentioned mental health education assistance system based on big data, including:

[0135] Step S1, collecting the eye movement data of the subject to be tested during the process of watching videos, online courses, or classroom learning in mental health education.

[0136] Step S2, inputting the eye movement data into the anxiety detection model for anxiety detection.

[0137] In step S2, the steps for constructing the anxiety detection model are as follows:

[0138] Preprocess the eye movement data. Assume the eye movement data contains N groups of samples, and each sample consists of M-dimensional features: X = {x1, x2, …, x N},

[0139] where X represents the complete eye movement data set, containing N groups of data, and x n represents the eye movement feature vector of the nth sample, with a dimension of M. N represents the total number of data samples, and M represents the feature dimension of each sample, including eye movement trajectory, pupil diameter change, blink frequency, and gaze fixation time;

[0140] Perform normalization. The formula is:

[0141]

[0142] where x' n is the normalized feature vector, μ is the feature mean, and σ is the feature standard deviation;

[0143] Use the normalized data X' = {x'1, x'2, …, x' N} as the input, and adopt CNN for static feature extraction. CNN consists of L convolutional layers, and the convolutional operation of each layer is:

[0144] h l = f(W l * h l-1 + b l ),

[0145] where h l is the feature map of the lth layer of CNN, W l is the weight matrix of the convolutional kernel of the lth layer, * represents the convolutional operation, b l is the bias term, and f(·) adopts the ReLU activation function,

[0146] Finally, obtain the static feature H extracted by CNN:

[0147] H = {h1, h2, …, h N},

[0148] where H is the static feature sequence extracted by CNN, corresponding to the number of input samples N;

[0149] Adopt LSTM for time series feature modeling. Use the feature H extracted by CNN as the input of LSTM to capture the dynamic change patterns in the time series. LSTM adopts a single-layer structure, and the calculation method is:

[0150] s n = LSTM(h n , s n-1 ),

[0151] Among them, s n is the LSTM hidden state vector of the nth sample, h n is the corresponding CNN feature vector, s n-1 is the hidden state at the previous moment,

[0152] Finally, the feature S processed by LSTM is obtained:

[0153] S = {s1, s2, …, s N},

[0154] where S is the dynamic feature sequence generated by LSTM;

[0155] The feature S extracted by LSTM is input into the fully connected layer, and the anxiety detection probability is calculated through Softmax. The formula is:

[0156] p = Softmax(W s s N + b s ),

[0157] where p is the probability distribution of the anxiety state, W s is the weight matrix of the fully connected layer, b s is the bias term of the fully connected layer,

[0158] Finally, the anxiety detection model outputs the anxiety state category C of the subject to be tested:

[0159] C = arg max(p),

[0160] where C is the predicted anxiety state category;

[0161] Specifically, the anxiety detection model combines CNN and LSTM to extract static and dynamic features;

[0162] The convolutional layer is used to capture local features and extract effective eye movement data patterns,

[0163] LSTM processes time series data. Through the recursive structure, it captures long-term dependencies, realizes the dynamic modeling of anxiety changes, enables the fully connected layer and Softmax to calculate the anxiety detection probability, and determines the anxiety category;

[0164] Step S3: Based on the anxiety detection result in step S2, the subject to be tested is given an anxiety score and classified according to different anxiety levels;

[0165] The steps of giving the subject to be tested an anxiety score based on the anxiety detection result in step S2 and classifying according to different anxiety levels are as follows:

[0166] Based on the anxiety detection result C in step S2, combined with the feature vector S, calculate the score, classify it according to different anxiety levels, and calculate the anxiety score. The formula is:

[0167] A = w1p + w2|S|,

[0168] where A is the anxiety score, the range is set between [0, 1], w1 and w2 are score weight coefficients, satisfying w1 + w2 = 1, p is the anxiety state probability calculated in step S2, and |S| is the norm of the dynamic feature vector extracted by LSTM:

[0169]

[0170] where |S| represents the average intensity of the time series features calculated by LSTM, measuring the degree of change in eye movement data, and s n is the nth group of sample feature vectors calculated by LSTM in step S2.

[0171] The normalized anxiety score, the formula is:

[0172]

[0173] where A' is the normalized anxiety score, the range is [0, 1], A min , A max the minimum and maximum values of the anxiety score in the data;

[0174] Classify by anxiety level. Based on the normalized anxiety score A', set the anxiety level threshold:

[0175] Low anxiety: A' < τ1,

[0176] Medium anxiety: τ1 ≤ A' < τ2,

[0177] High anxiety: A' ≥ τ2,

[0178] where τ1 and τ2 are empirical thresholds, set according to psychological assessment data, and determined by the quantile method;

[0179] Finally, output the anxiety level L:

[0180] If A' < τ1, then L = Low anxiety,

[0181] If τ1 ≤ A' < τ2, then L = Medium anxiety,

[0182] If A' ≥ τ2, then L = High anxiety;

[0183] Specifically, based on the features extracted by CNN-LSTM and the anxiety detection results here, an anxiety score is calculated and the anxiety level is classified; by combining the anxiety detection probability and the norm of the time series features, a weighted anxiety score is calculated and normalized to adapt to the data distributions of different subjects; the quantile method is used to set the anxiety score threshold for classifying low, medium, and high anxiety.

[0184] Step S4: Generate personalized feedback according to the anxiety score, push it to the subject, and record the interaction data of the subject.

[0185] Step S5: Input the interaction data in Step S4 into the anxiety detection model to dynamically analyze the anxiety trend of the subject, and adjust the personalized feedback content according to the trend change.

[0186] The step of inputting the interaction data in Step S4 into the anxiety detection model to dynamically analyze the anxiety trend of the subject and adjusting the personalized feedback content according to the trend change is as follows:

[0187] Preprocess the interaction data and define the interaction data set as D:

[0188] D = {d1, d2, …, d Q},

[0189] where D represents the interaction data set of the subject, d q represents the feature vector of the q-th interaction, including the viewing duration, click behavior, and active search behavior, and Q represents the time step of the interaction data.

[0190] Normalize the interaction data. The formula is:

[0191]

[0192] where d' q is the normalized interaction data, μ D is the mean of the interaction data, and σ D is the standard deviation.

[0193] Define the anxiety trend as a sequence of anxiety scores changing over time, denoted as T A :

[0194] T A = {A1, A2, …, A Q},

[0195] where T A is the sequence of anxiety scores changing with the interaction data, A q is the anxiety score corresponding to the q-th interaction, and is calculated using exponential smoothing:

[0196] A′q = αA q + (1 - α)A' q-1 ,

[0197] where A' q is the smoothed anxiety score trend value, α is the smoothing coefficient with a range of 0 < α < 1, and A' q-1 is the smoothed anxiety score calculated in the previous calculation.

[0198] Calculate the anxiety trend change rate, and the formula is:

[0199] ΔA q = A' q - A' q-1 ,

[0200] where ΔA q represents the change amplitude of the anxiety trend;

[0201] Based on the anxiety trend change rate ΔA q perform personalized feedback adjustment, and the adjustment formula is:

[0202] F q = F q-1 + βΔA q ,

[0203] where F q is the adjusted personalized feedback intensity, F q-1 is the previous personalized feedback content, and β is the feedback adjustment coefficient.

[0204] If the anxiety trend rises and ΔA q > τ F , then increase the content of psychological regulation intervention.

[0205] If the anxiety trend drops and ΔA q < -τ F , then reduce the intervention intensity;

[0206] Specifically, in this step, use the interaction data of the test subject to dynamically analyze the anxiety trend and adjust the personalized feedback content; adopt mean-variance normalization to reduce the impact of data fluctuations; then calculate the change trend of the anxiety score through exponential smoothing to improve the stability of trend analysis; and then adaptively adjust the personalized feedback intensity according to the change rate of the anxiety trend.

[0207] Step S6, when the anxiety trend in step S5 indicates that the anxiety score of the test subject continues to rise and reaches the preset threshold, trigger the warning mechanism;

[0208] When the anxiety trend in step S5 indicates that the anxiety score of the test subject continues to rise and reaches the preset threshold, the steps to trigger the warning mechanism are as follows:

[0209] Calculate the change trend of the anxiety score. The formula is:

[0210]

[0211] Among them, Gq is the change trend of the anxiety score, K is the size of the observation window, that is, the average value of the score change rate is calculated within K interaction data, and ΔA q-k is the change rate of the anxiety trend defined in step S5.

[0212] If G q exceeds the preset anxiety growth threshold τ G , it is determined that the anxiety score is continuously rising:

[0213] G q > τ G ,

[0214] Among them, τ G is the threshold value of the anxiety score change, set based on statistical analysis;

[0215] Set the trigger warning mechanism. The trigger condition of the warning mechanism is:

[0216] A' q ≥ τ W and G q > τ G ,

[0217] Among them, A' q is the smoothed anxiety score, τ W is the anxiety score warning threshold, set as the quantile of high-anxiety users or specific clinical criteria, and τ G is the change threshold for the continuous rise of the anxiety score.

[0218] If this condition is met, trigger the warning: W q = 1,

[0219] Among them, W q indicates whether the warning is triggered currently. If it is equal to 1, the warning is triggered; if it is equal to 0, the warning is not triggered;

[0220] According to the anxiety score A' q and the anxiety trend G q of the test subject, determine the warning level:

[0221] If τ W ≤ A' q < τ W + δ1 and G q > τ G , then L W = low-level warning,

[0222] If τW +δ1 ≤ A' q < τ W +δ2 and G q > τ G , then L W = Intermediate warning,

[0223] If A' q ≥ τ W +δ2 and G q > τ G , then L W = High - level warning,

[0224] where L W is the warning level, and δ1, δ2 are the boundary values of the warning levels of the anxiety score,

[0225] When W q = 1, different intervention strategies are adopted according to L W ; Specifically, this step determines when to trigger the warning mechanism and conducts hierarchical intervention based on the changes in the anxiety score and trend:

[0226] Specifically, this step determines when to trigger the warning mechanism and conducts hierarchical intervention based on the changes in the anxiety score and trend:

[0227] Conduct trend analysis, calculate the growth trend of the anxiety score over a period of time, and combine the score level to determine whether to trigger the warning. Only when the anxiety score reaches the set threshold and the anxiety score continues to rise, the warning mechanism is activated; to avoid the deterioration of the anxiety level.

[0228] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.

Claims

1. A mental health education assistance system based on big data, characterized in that: include, The eye movement data collection module is used to obtain the eye movement data of the subjects when they are watching videos, online courses or studying in the classroom during the mental health education process; Anxiety detection module, which receives eye movement data and processes it through the anxiety detection model; Anxiety score calculation module, used to receive the anxiety detection result of the anxiety detection module, calculate the anxiety score of the tested object based on the anxiety detection model, and classify the tested object according to different anxiety levels; Personalized feedback generation module, which generates personalized feedback based on anxiety scores; An interaction data recording module pushes personalized feedback to the tested object and records the interaction data of the tested object, wherein the interaction data includes viewing time, click behavior and active search behavior of the feedback content; A dynamic adjustment module is used to receive the interaction data and input it into the anxiety detection model, dynamically analyze the anxiety trend of the subject, and adjust the personalized feedback content of the personalized feedback generation module according to the trend change; Early warning module, when the anxiety trend of the dynamic adjustment module shows that the anxiety score of the subject continues to rise and reaches the preset threshold, the early warning mechanism is triggered and psychological intervention is carried out.

2. The mental health education assistance system based on big data according to claim 1, characterized in that: The eye movement data includes eye movement trajectory, pupil diameter change, blink frequency and gaze dwell time; The eye movement data acquisition module is based on a camera device and combines an eye movement tracking algorithm to perform real-time analysis on the eye movement characteristics of the subject.

3. The mental health education assistance system based on big data according to claim 2, wherein: The anxiety detection model is based on the fusion architecture of convolutional neural network (CNN) and long short-term memory (LSTM), where CNN is used to extract static features and LSTM is used to capture time series changes. The anxiety detection module includes a data preprocessing unit for denoising and standardizing the eye movement data, and performing feature matching in combination with a preset anxiety behavior database.

4. The mental health education assistance system based on big data according to claim 3, characterized in that: The personalized feedback includes psychological adjustment video recommendations, relaxation training push notifications, emotion adjustment tips, and psychological assessment questionnaires; The personalized feedback generation module combines the historical anxiety scores and interaction data of the subjects and uses a reinforcement learning algorithm to optimize the feedback strategy.

5. The mental health education assistance system based on big data according to claim 4, characterized in that: The early warning module adopts a multi-level threshold judgment strategy to automatically match psychological intervention measures for different anxiety levels, including anonymous psychological support recommendations, psychological counseling appointment suggestions and emergency intervention notifications, to notify psychological counselors to intervene.

6. The mental health education assistance system based on big data according to claim 5, characterized in that, The auxiliary methods of this system are: Step S1, collecting eye movement data of the subject when watching videos, online courses or classroom learning during mental health education; Step S2, inputting the eye movement data into an anxiety detection model for anxiety detection; Step S3, scoring the anxiety of the subjects based on the anxiety detection results of step S2, and classifying them according to different anxiety levels; Step S4, generating personalized feedback based on the anxiety score, pushing it to the subject, and recording the interaction data of the subject; Step S5, inputting the interactive data of step S4 into the anxiety detection model, dynamically analyzing the anxiety trend of the tested subject, and adjusting the personalized feedback content according to the trend change; Step S6: When the anxiety trend of step S5 indicates that the anxiety score of the subject continues to increase and reaches a preset threshold, an early warning mechanism is triggered.

7. The mental health education assistance system based on big data according to claim 6, characterized in that: In step S2, the steps of constructing the anxiety detection model are: Preprocess the eye movement data. Assume the eye movement data contains N groups of samples, and each sample consists of M-dimensional features Composition: X = {x1, x2, …, x N} Among them, X represents the complete eye movement data set, which contains N groups of data, and x n represents the eye movement feature vector of the nth sample, with a dimension of M. N represents the total number of data samples, and M represents the feature dimension of each sample, including eye movement trajectory, pupil diameter change, blink frequency, and gaze fixation time; Normalize the data. The formula is: where x' n is the normalized eigenvector, μ is the eigen mean, and σ is the eigen standard deviation; Taking the normalized data X' = {x'1, x'2, …, x' N} as the input, static feature extraction is performed using CNN. The CNN consists of L convolutional layers, and the convolutional operation for each layer is as follows: h l = f(W l * h l-1 + b l ), where h l is the feature map of the l-th layer of CNN, W l is the convolutional kernel weight matrix of the l-th layer, * represents the convolution operation, b l is the bias term, and f(·) uses the ReLU activation function Finally, obtain the static features H extracted by the CNN: H = {h1, h2, …, h N}, Among them, H is the sequence of static features extracted by the CNN, corresponding to the number of input samples N; Use LSTM for time series feature modeling. Take the features H extracted by the CNN as the input of the LSTM to capture the dynamic change patterns in the time series. The LSTM adopts a single-layer structure, and the calculation method is: s n = LSTM(h n , s n-1 ), where s n is the LSTM hidden state vector of the nth sample, h n is the corresponding CNN feature vector, and s n-1 is the hidden state at the previous moment. Finally, obtain the features S processed by the LSTM: S = {s1, s2, …, s N}, Among them, S is the sequence of dynamic features generated by the LSTM; Input the features S extracted by the LSTM into the fully connected layer, and calculate the anxiety detection probability through Soft max. The formula is: p = Softmax(W s s N + b s ), Among them, p is the probability distribution of the anxiety state, and W s is the weight matrix of the fully connected layer, and b s is the bias term of the fully connected layer. Finally, the anxiety detection model outputs the anxiety state category C of the subject being tested: C = arg max(p), where C is the predicted anxiety state category.

8. The mental health education assistance system based on big data according to claim 7, characterized in that: The step of performing an anxiety score on the subject being tested based on the anxiety detection result in step S2 and classifying them according to different anxiety levels is as follows Based on the anxiety detection result C in step S2, combine the feature vector S to calculate the score and classify according to different anxiety levels. Calculate the anxiety score. The formula is: A = w1p + w2|S|, where A is the anxiety score, with the range set between [0,1], w1 and w2 are score weight coefficients, satisfying w1 + w2 = 1, p is the anxiety state probability calculated in step S2, and |S| is the norm of the dynamic feature vector extracted by the LSTM: Among them, |S| represents the average intensity of the time series features calculated by the LSTM, which measures the degree of change in the eye movement data, and s n is the nth group of sample feature vectors calculated by the LSTM in step S2. Normalize the anxiety score. The formula is: where A' is the normalized anxiety score, ranging from [0, 1], A min , A max is the minimum and maximum values of the anxiety scores in the data; Classify according to anxiety levels. Based on the normalized anxiety score A', set the anxiety level thresholds: Low anxiety: A' < τ1, Moderate anxiety: τ1 ≤ A' < τ2, High anxiety: A' ≥ τ2, where τ1 and τ2 are empirical thresholds, set according to psychological assessment data, and determined using the quantile method; Finally, output the anxiety level L: If A' < τ1, then L = low anxiety, If τ1 ≤ A' < τ2, then L = moderate anxiety, If A' ≥ τ2, then L = high anxiety.

9. The mental health education assistance system based on big data according to claim 8, wherein: The step of inputting the interaction data in step S4 into the anxiety detection model, dynamically analyzing the anxiety trend of the subject being tested, and adjusting the personalized feedback content according to the trend change is as follows Preprocess the interaction data. Define the interaction data set as D: D = {d1, d2, …, d Q}, Among them, D represents the interaction dataset of the object under test, and d q represents the feature vector of the q-th interaction, including viewing duration, click behavior, and active search behavior. Q represents the time step of the interaction data, Normalize the interaction data. The formula is: where d' q is the normalized interaction data, μ D is the mean of the interaction data, and σ D is the standard deviation; Define the anxiety trend as a sequence of anxiety scores over time, denoted as T A : T A = {A1, A2, …, A Q}, where, T A is the sequence of the anxiety score varying with the interaction data, and A q is the anxiety score corresponding to the q-th interaction, which is calculated by exponential smoothing: A' q = αA q + (1 - α)A' q-1 , Among them, A' q is the smoothed anxiety score trend value, α is the smoothing coefficient, with the range 0 < α < 1, and A' q-1 is the smoothed anxiety score calculated in the previous time. Calculate the anxiety trend change rate. The formula is: ΔA q = A' q - A' q-1 , where ΔA q represents the change range of the anxiety trend; According to the anxiety trend change rate ΔA q Perform personalized feedback adjustment, and the adjustment formula is: F q = F q-1 + βΔA q , Among them, F q is the adjusted personalized feedback intensity, F q-1 is the previous personalized feedback content, and β is the feedback adjustment coefficient. If the anxiety trend is rising, ΔA q > τ F , then increase the content of psychological adjustment intervention If the anxiety trend decreases, ΔA q <-τ F , then reduce the intervention intensity.

10. The mental health education assistance system based on big data according to claim 9, characterized in that: The step of triggering the warning mechanism when the anxiety trend in step S5 indicates that the anxiety score of the subject being tested continues to rise and reaches the preset threshold is as follows Calculate the anxiety score change trend. The formula is: Among them, G q is the changing trend of the anxiety score, K is the size of the observation window, that is, the average value of the score change rate is calculated within the K interaction data, and ΔA q-k is the anxiety trend change rate defined in step S5. If G q exceeds the preset anxiety growth threshold τ G , it is determined that the anxiety score is continuously rising: G q >τ G , Among them, τ G is the threshold for the change in anxiety score, which is set based on statistical analysis; Set the trigger for the warning mechanism. The trigger condition for the warning mechanism is: A' q ≥τ W and G q >τ G , Among them, A' q is the smoothed anxiety score, and τ W is the anxiety score warning threshold, which is set as the quantile of high-anxiety users or specific clinical criteria, and τ G is the change threshold for the continuous increase of the anxiety score. If this condition is met, a warning is triggered: W q = 1, Among them, W q indicates whether an early warning is currently triggered. If it is equal to 1, the early warning is triggered; if it is equal to 0, the early warning is not triggered. According to the anxiety score A' of the subject under test q and the anxiety trend G q , determine the early warning level: If τ W ≤ A' q < τ W + δ1 and G q > τ G , then L W = Low-level warning, If τ W + δ1 ≤ A' q < τ W + δ2 and G q > τ G , then L W = Medium-level warning, If A' q ≥ τ W + δ2 and G q > τ G , then L W = High - level warning, Among them, L W is the warning level, and δ1 and δ2 are the boundary values of the warning levels of the anxiety score. When W q = 1, different intervention strategies are adopted according to L W ​

Images