A student tablet-based adaptive tutoring method, storage medium and device

Abstract

The application discloses a student tablet-based adaptive tutoring method, a storage medium and equipment, through fusion of multi-source sensor data such as touch pressure, line of sight focus and equipment posture of the student tablet, second-level real-time evaluation and cause classification of cognitive load are realized, differentiated intervention strategies can be generated for different load types, and the accuracy and timeliness of tutoring are significantly improved. Combined with local edge computing and privacy encryption storage mechanism, the safety of student biological feature data is effectively guaranteed, and the concern of parents about privacy leakage is eliminated.

Description

Technical Field

[0001] This invention relates to the field of intelligent education technology, specifically to an adaptive tutoring method, storage medium, and device based on a student tablet. Background Technology

[0002] Existing student tutoring technologies primarily rely on cloud-based knowledge graphs and quiz result data for analysis. They typically infer a student's learning status solely based on quiz accuracy and time taken, lacking the ability to capture real-time physiological and behavioral data during the learning process. This lagging and one-dimensional assessment method cannot distinguish whether a student's learning difficulties stem from a lack of knowledge, psychological anxiety, or physical fatigue, leading to rigid tutoring strategies that often intervene only after the student has already experienced frustration. Summary of the Invention

[0003] In view of the shortcomings of the existing technology, the present invention aims to provide an adaptive tutoring method, storage medium and device based on student tablets.

[0004] To achieve the above objectives, the present invention adopts the following technical solution: An adaptive tutoring method based on student tablets includes the following steps: S1. Through the built-in touch screen sensor, front camera and posture sensor of the student tablet, the interactive behavior data of the student in the process of completing the pushed learning tasks are collected in real time. The interactive behavior data includes pen stroke pressure sequence, eye focus area coordinates and device tilt angle sequence. S2. Input the interactive behavior data obtained in step S1 into the preset cognitive load calculation model, calculate the cognitive load index at the current moment in real time, and dynamically adjust the display parameters of the student tablet interface and / or the push strategy of learning tasks according to the cognitive load index. S3. When the cognitive load index meets the preset conditions, the local generative assistance mechanism is triggered, which uses the large language model deployed on the student tablet to generate hierarchical prompts for the current learning task and / or adopts preset intervention strategies. S4. Record the students' subsequent interactive behavior data after triggering the local generative assistance mechanism, calculate the intervention effectiveness score, and use the intervention effectiveness score to update the weight parameters in the cognitive load calculation model in reverse, forming a closed-loop optimization.

[0005] Furthermore, in step S2, the calculation formula for the cognitive load calculation model is as follows: ; CLI t P represents the cognitive load index at time t. t Δt,tThe pen pressure sequence is represented within a time window Δt, σ() represents the standard deviation calculation function, and P norm This is the normalized pressure reference value; G i R represents the coordinates of the gaze focus area during the i-th sampling. task The valid area of ​​the current learning task on the student's tablet screen is defined by I(), which is an indicator function that takes the value 1 when the line of sight is within the valid area and 0 otherwise, and n is the number of samples. t Let θ be the tilt angle of the device at time t. opt θ is the preset optimal learning posture angle. max The maximum permissible deviation angle is denoted by w1, w2, and w3, which represent the pressure fluctuation weight, line-of-sight dispersion weight, and attitude deviation weight, respectively, and w1+w2+w3=1.

[0006] Furthermore, the dynamic adjustment of the display parameters of the student tablet interface and / or the push strategy for learning content described in step S2 specifically includes: When the cognitive load index CLI t When the cognitive load index (CLI) is in the range of [0.3, 0.6), the brightness of the student tablet interface remains unchanged, and only the text density displayed on the student tablet interface is adjusted; when the CLI is... t When the color temperature is in the range of [0.6, 0.8), the color temperature of the student tablet interface is reduced to below 4500K, and decorative elements unrelated to learning are hidden in the student tablet interface. At the same time, the local generation assistance mechanism in step S3 is triggered. When the cognitive load index CLT ≥ 0.8, the input function setting duration of the student tablet interface is forcibly locked, relaxation guidance audio is played, and the push of all learning tasks is paused.

[0007] Further, in step S1, the acquisition of the pen pressure sequence specifically involves: acquiring the pen pressure sequence by collecting the pressure value when the student writes on the student tablet screen using an electromagnetic pen, with a sampling frequency of not less than 100Hz; the acquisition of the gaze focus area coordinates specifically involves: capturing the student's pupil position through the front-facing camera, and calculating the intersection of the student's gaze and the student tablet screen by combining the data from the posture sensor built into the student tablet, thereby obtaining the gaze focus area coordinates; the acquisition of the device tilt angle sequence specifically involves: acquiring the tilt angle of the student tablet relative to the horizontal plane through the posture sensor built into the student tablet.

[0008] Furthermore, in step S3, when the local generative assistance mechanism is triggered, the cognitive load index CLI is first analyzed. t The dominant component is elevated, and corresponding hierarchical prompts and / or appropriate intervention strategies are generated based on the type of dominant component: If it is a pressure fluctuation weighting item In CLI tIf the contribution rate exceeds 50%, it is judged as anxiety-related load. The hierarchical prompts generated by the large language model include emotional reassurance statements and more granular guidance content that breaks down the steps for the current learning task; if the focus is distracted, the weighting factor... In CLI t If the contribution of the input exceeds 50%, it is judged as a distraction load. The hierarchical prompts generated by the large language model include interactive questions or highlighted / blinking prompts related to the current learning content to recapture attention; if the posture deviation weight term... In CLI t If the contribution of the content on the screen exceeds 50%, it is determined to be a fatigue-type load. The content currently displayed on the screen will be forcibly switched to voice broadcast mode to reduce the need for screen reading. When the pressure fluctuation weight term, the line of sight distraction weight term, and the posture deviation weight term are in CLI t When the contribution of each factor does not exceed 50%, it indicates that the student's cognitive load is caused by a combination of factors, without a single dominant factor. This is classified as a comprehensive load. A comprehensive, layered guidance prompt is generated using a large language model, including mild emotional reassurance statements, concise summaries of the core knowledge points of the current learning task, audio explanations, and highlighting key steps. This avoids the secondary cognitive pressure caused by reading long passages of text. Simultaneously, a combined intervention strategy of information degradation and micro-rests is employed. Information degradation: Reduce the content display density on the student tablet interface and simplify the interaction to a single-step click confirmation mode to lower the operation threshold; Micro-break intervention: Forcefully pause the current learning task for 5 to 10 seconds, play soothing sound effects or simple breathing guidance animations on the interface to help students relax and reset from the overall tense learning state, and resume after the countdown ends.

[0009] Further, in step S4, the calculation logic for the intervention effectiveness score is as follows: if, after receiving generative assistance, the average cognitive load index of the student decreases by more than 20% compared to before the intervention within a set continuous period, and the correct answer rate remains unchanged or increases, then the intervention effectiveness score is 1; if the cognitive load index does not decrease or the correct answer rate decreases, then the intervention effectiveness score is 0; if the cognitive load index decreases but the decrease does not exceed 20%, and the correct answer rate remains unchanged or increases, then the intervention effectiveness score is defined as 0.5, representing "partially effective"; based on the intervention effectiveness score, the gradient descent method is used to update the weight parameters w1, w2, and w3 in the cognitive load calculation model, so that the cognitive load index can better reflect the student's true state.

[0010] Further, in step S4, the specific process of updating the weight parameters w1, w2, and w3 in the cognitive load calculation model using the gradient descent method based on the intervention effectiveness score is as follows: A1. Define the loss function: First, convert the intervention effectiveness score into an error value that needs to be optimized, and set the error = 1 - intervention effectiveness score; if the intervention effectiveness score is 0, the error is the largest, indicating that the weight allocation of the current cognitive load calculation model is unreasonable and needs to be adjusted significantly; if the score is 1, the error is 0, and the weights of the cognitive load calculation model remain unchanged. A2. Determine the direction of adjustment: Review the pre-intervention period. , , The specific value; if the error is large, it means that the cognitive load calculation model misdiagnosed the student's true state, that is, it assigned too high a weight to the feature with the largest value at that time; therefore, the adjustment direction is set as follows: reduce the weight of the feature with the largest value before the intervention, and correspondingly increase the weight of the other two feature items. A3. Calculate the initial new weights: Introduce a preset learning rate, and calculate the initial update values ​​of w1, w2, and w3 respectively using the formula Initial New Weights = Old Weights - Learning Rate × Gradient; A4. Normalization: Since w1 + w2 + w3 = 1 is strictly required, after obtaining the initial new weights, these three values ​​need to be divided by their sum respectively.

[0011] Furthermore, when storing data, the student tablet only saves statistical indicators after feature extraction, including the daily average cognitive load index, the total number of triggers of the generative auxiliary mechanism, and the final convergence value of each weight parameter of the cognitive load calculation model. The original biometric data, including the original value of the pen pressure sequence and the gaze focus image frame, are destroyed in memory immediately. Before being stored in the storage medium, the statistical indicators are encrypted using the AES-256 standard and can only be decrypted and exported after obtaining the authorization key of the parent bound to the student tablet, ensuring that the student's biometric privacy data is not leaked.

[0012] The present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0013] The present invention also provides a student tablet, including a processor and a memory, wherein the memory is used to store a computer program; and the processor is used to execute the computer program to implement the above-described method.

[0014] The beneficial effects of this invention are as follows: This invention integrates multi-source sensor data, including touch pressure, gaze focus, and device posture, from student tablets to achieve real-time assessment and causal classification of cognitive load within seconds. It can generate differentiated intervention strategies for different load types, significantly improving the accuracy and timeliness of tutoring and intervention. Combined with local edge computing and privacy-encrypted storage mechanisms, it effectively ensures the security of students' biometric data, eliminating parents' concerns about privacy leaks. Detailed Implementation

[0015] The present invention will be further described below. It should be noted that this embodiment is based on the present technical solution and provides detailed implementation methods and specific operation processes, but the protection scope of the present invention is not limited to this embodiment.

[0016] This embodiment provides an adaptive tutoring method based on student tablets, including the following steps: S1. Through the built-in touch screen sensor, front camera and posture sensor of the student tablet, the interactive behavior data of the student in the process of completing the pushed learning tasks are collected in real time. The interactive behavior data includes pen touch pressure sequence, eye focus area coordinates and device tilt angle sequence; the device tilt angle is the angle between the student tablet and the horizontal plane formed by tilting the student tablet away from the student's body (tilting backward).

[0017] S2. Input the interactive behavior data obtained in step S1 into the preset cognitive load calculation model, calculate the cognitive load index at the current moment in real time, and dynamically adjust the display parameters of the student tablet interface and / or the push strategy of learning tasks according to the cognitive load index. S3. When the cognitive load index meets the preset conditions, the local generative assistance mechanism is triggered, which uses the large language model deployed on the student tablet to generate hierarchical prompts for the current learning task and / or adopts preset intervention strategies. S4. Record the students' subsequent interactive behavior data after triggering the local generative assistance mechanism, calculate the intervention effectiveness score, and use the intervention effectiveness score to update the weight parameters in the cognitive load calculation model in reverse, forming a closed-loop optimization.

[0018] In this embodiment, the calculation formula for the cognitive load calculation model in step S2 is as follows: ; CLI t P represents the cognitive load index at time t. t Δt,t The pen pressure sequence is represented within a time window Δt, σ() represents the standard deviation calculation function, and P norm This is the normalized pressure reference value; G iR represents the coordinates of the gaze focus area during the i-th sampling. task The valid area of ​​the current learning task on the student's tablet screen is defined by I(), which is an indicator function that takes the value 1 when the line of sight is within the valid area and 0 otherwise, and n is the number of samples. t Let θ be the tilt angle of the device at time t. opt θ is the preset optimal learning posture angle (i.e., the optimal tilt angle of the student tablet). max The maximum permissible deviation angle is denoted by w1, w2, and w3, which represent the pressure fluctuation weight, line-of-sight dispersion weight, and attitude deviation weight, respectively, and w1+w2+w3=1.

[0019] In this embodiment, the dynamic adjustment of the display parameters of the student tablet interface and / or the push strategy of learning content in step S2 specifically includes: When the cognitive load index CLI t When the cognitive load index (CLI) is in the range of [0.3, 0.6), the brightness of the student tablet interface remains unchanged, and only the text density displayed on the student tablet interface is adjusted; when the CLI is... t When the color temperature is in the range of [0.6, 0.8), the color temperature of the student tablet interface is reduced to below 4500K, and decorative elements unrelated to learning are hidden in the student tablet interface. At the same time, the local generation assistance mechanism in step S3 is triggered. When the cognitive load index CLT ≥ 0.8, the input function setting of the student tablet interface is forcibly locked for a duration (e.g., 5 minutes), and relaxation guidance audio is played, and all learning task pushes are paused.

[0020] Through the above process, the cognitive load index obtained by fusion calculation of multi-source sensor data can be used to identify whether there is an overlearning problem, and corresponding measures can be taken accordingly.

[0021] In this embodiment, in step S1, the acquisition of the pen pressure sequence specifically involves: acquiring the pen pressure sequence by collecting the pressure value when the student writes on the student tablet screen using an electromagnetic pen, with a sampling frequency of not less than 100Hz; the acquisition of the gaze focus area coordinates specifically involves: capturing the student's pupil position through the front-facing camera, and calculating the intersection of the student's gaze and the student tablet screen by combining the data from the posture sensor (such as a six-axis gyroscope) built into the student tablet, thereby obtaining the gaze focus area coordinates; the acquisition of the device tilt angle sequence specifically involves: acquiring the tilt angle of the student tablet relative to the horizontal plane through the posture sensor built into the student tablet.

[0022] In this embodiment, in step S3, when the local generative assistance mechanism is triggered, the cognitive load index CLI is first analyzed. t The dominant component is elevated, and corresponding hierarchical prompts and / or appropriate intervention strategies are generated based on the type of dominant component: If it is a pressure fluctuation weighting item In CLI t The contribution percentage of the pressure fluctuation weighting term in CLI (i.e., the weighting percentage of the pressure fluctuation weighting term) t If the percentage of anxiety load exceeds 50%, it is judged as anxiety-related load. The hierarchical prompts generated by the large language model include emotional reassurance statements and more granular guidance content that breaks down the steps for the current learning task. If the focus is distracted, the weighting factor... In CLI t The contribution percentage in CLI (i.e., the percentage of the distraction weighting term) t If the percentage of [something] exceeds 50%, it is considered a distraction load. The hierarchical prompts generated by the large language model include interactive questions or highlighted / blinking prompts related to the current learning content to recapture attention. If the posture deviation weighting term [is affected]... In CLI t The contribution percentage in CLI (i.e., the weighting of the state bias term) t If the percentage of screen time exceeds 50%, it is considered fatigue-related overload, and the content currently displayed on the screen will be forcibly switched to voice broadcast mode to reduce the need for screen reading.

[0023] When the pressure fluctuation weight term, the line of sight distraction weight term, and the posture deviation weight term are in CLI t When the contribution of each factor in the learning process does not exceed 50%, it indicates that the student's cognitive load is caused by a combination of factors, without a single dominant factor. In this case, it is classified as a comprehensive load. A comprehensive, layered guidance prompt is generated using a large language model, including mild emotional reassurance statements, concise summaries of the core knowledge points of the current learning task, audio explanations, and highlighting of key steps, avoiding secondary cognitive pressure from reading long passages of text. Simultaneously, a combined intervention strategy of information degradation and micro-rests is employed. Information degradation: Reduce the content display density on the student tablet interface (e.g., reduce the amount of text displayed on the same screen and increase the white space on the page), and simplify the interactive operation to a single-step click confirmation mode to lower the operation threshold.

[0024] Micro-break intervention: Forcefully pause the current learning task for 5 to 10 seconds, play soothing sound effects or simple breathing guidance animations on the interface to help students relax and reset from the overall tense learning state, and resume after the countdown ends.

[0025] In this embodiment, in step S4, the calculation logic for the intervention effectiveness score is as follows: if, after receiving generative assistance, the student's average cognitive load index decreases by more than 20% within a set continuous time period (e.g., 5 consecutive minutes) compared to before the intervention, and the correct answer rate remains unchanged or increases, then the intervention effectiveness score is 1; if the cognitive load index does not decrease (remains unchanged or increases) or the correct answer rate decreases, then the intervention effectiveness score is 0; if the cognitive load index decreases but the decrease does not exceed 20%, and the correct answer rate remains unchanged or increases, then the intervention effectiveness score is defined as 0.5, representing "partially effective"; based on the intervention effectiveness score, the gradient descent method is used to update the weight parameters w1, w2, and w3 in the cognitive load calculation model, so that the cognitive load index can better reflect the student's true state.

[0026] In this embodiment, the specific process of updating the weight parameters w1, w2, and w3 in the cognitive load calculation model using the gradient descent method in step S4, based on the intervention effectiveness score, is as follows: A1. Define the loss function: First, convert the intervention effectiveness score into an error value that needs to be optimized, and set the error = 1 - intervention effectiveness score; if the intervention effectiveness score is 0 (intervention is ineffective), the error is the largest, indicating that the weight allocation of the current cognitive load calculation model is unreasonable and needs to be adjusted significantly; if the score is 1 (intervention is effective), the error is 0, and the weight of the cognitive load calculation model remains unchanged.

[0027] A2. Determine the direction of adjustment (calculate the gradient): Review the pre-intervention period. , , The specific numerical value is needed. If the error is large, it indicates that the cognitive load calculation model misdiagnosed the student's true state, that is, it assigned too high a weight to the feature with the largest value at that time. Therefore, the adjustment direction (gradient) is set as follows: reduce the weight of the feature with the largest value before the intervention, and correspondingly increase the weight of the other two feature items.

[0028] A3. Calculate the initial new weights (parameter update): Introduce a preset learning rate to control the magnitude of each adjustment and prevent over-adjustment. Following the core logic of gradient descent, calculate the initial update values ​​of w1, w2, and w3 using the formula: Initial New Weights = Old Weights - Learning Rate × Gradient.

[0029] A4. Normalization (Saving Constraints): Since w1 + w2 + w3 = 1 is strictly required, after obtaining the initial new weights, these three values ​​need to be divided by their sum (i.e., final weight = initial new weight / sum of the three initial new weights). This normalization step ensures that the updated weight sum is still 1, completing the entire closed-loop optimization process.

[0030] More specifically, in step S2, the normalized pressure reference value P norm and the preset optimal learning posture angle θ opt It is not a fixed constant, but a variable that is dynamically adjusted based on the subject type; the student tablet has a built-in subject feature configuration file, and when it detects that the current learning task is mathematics or physics, P... norm Set to 1.2 Newtons, θ opt Set to 35 degrees to suit high-frequency writing scenarios, this angle provides comfortable support for the arm and wrist, facilitating writing force; When the current learning task is detected to be English or Chinese, P norm Set to 0.6 Newtons, θ opt The angle is set to 25 degrees to suit low-frequency writing and high-frequency reading scenarios. This angle better matches the natural downward angle of the eye, effectively reducing screen glare and relieving neck strain. Thus, before each learning task begins, the corresponding normalized stress baseline value P is automatically loaded based on the subject selected by the user. norm and the preset optimal learning posture angle θ opt This eliminates cognitive load assessment bias caused by subject-specific differences.

[0031] In this embodiment, when the student tablet performs data storage, it only saves statistical indicators after feature extraction, including the daily average cognitive load index, the total number of triggers of the generative auxiliary mechanism, and the final convergence value of each weight parameter of the cognitive load calculation model. The original biometric data, including the original value of the pen pressure sequence and the gaze focus image frame, are destroyed in memory immediately. Before being stored in the storage medium, the statistical indicators are encrypted using the AES-256 standard, and can only be decrypted and exported after obtaining the parent's authorization key bound to the student tablet, ensuring that the student's biometric privacy data is not leaked.

[0032] For those skilled in the art, various corresponding changes and modifications can be made based on the above technical solutions and concepts, and all such changes and modifications should be included within the protection scope of the claims of this invention.

Claims

1. An adaptive tutoring method based on student tablets, characterized in that, Includes the following steps: S1. Through the touch screen sensor, front camera and posture sensor built into the student tablet, the interactive behavior data of the student in the process of completing the pushed learning tasks are collected in real time. The interactive behavior data includes pen stroke pressure sequence, eye focus area coordinates and device tilt angle sequence. S2. Input the interactive behavior data obtained in step S1 into the preset cognitive load calculation model, calculate the cognitive load index at the current moment in real time, and dynamically adjust the display parameters of the student tablet interface and / or the push strategy of learning tasks according to the cognitive load index. S3. When the cognitive load index meets the preset conditions, the local generative assistance mechanism is triggered, which uses the large language model deployed on the student tablet to generate hierarchical prompts for the current learning task and / or adopts preset intervention strategies. S4. Record the students' subsequent interactive behavior data after triggering the local generative assistance mechanism, calculate the intervention effectiveness score, and use the intervention effectiveness score to update the weight parameters in the cognitive load calculation model in reverse, forming a closed-loop optimization.

2. The method according to claim 1, characterized in that, In step S2, the calculation formula for the cognitive load calculation model is as follows: ； CLI t P represents the cognitive load index at time t. t Δt,t The pen pressure sequence is represented within a time window Δt, σ() represents the standard deviation calculation function, and P norm This is the normalized pressure reference value; G i R represents the coordinates of the gaze focus area during the i-th sampling. task The valid area of ​​the current learning task on the student's tablet screen is defined by I(), which is an indicator function that takes the value 1 when the line of sight is within the valid area and 0 otherwise, and n is the number of samples. t Let θ be the tilt angle of the device at time t. opt θ is the preset optimal learning posture angle. max The maximum permissible deviation angle is denoted by w1, w2, and w3, which represent the pressure fluctuation weight, line-of-sight dispersion weight, and attitude deviation weight, respectively, and w1+w2+w3=1.

3. The method according to claim 2, characterized in that, The dynamic adjustment of display parameters of the student tablet interface and / or the push strategy for learning content mentioned in step S2 specifically includes: When the cognitive load index CLI t When the cognitive load index (CLI) is in the range of [0.3, 0.6), the brightness of the student tablet interface remains unchanged, and only the text density displayed on the student tablet interface is adjusted; when the CLI is... t When the color temperature is in the range of [0.6, 0.8), the color temperature of the student tablet interface is reduced to below 4500K, and decorative elements unrelated to learning are hidden in the student tablet interface. At the same time, the local generation assistance mechanism in step S3 is triggered. When the cognitive load index CLT ≥ 0.8, the input function setting duration of the student tablet interface is forcibly locked, relaxation guidance audio is played, and the push of all learning tasks is paused.

4. The method according to claim 1, characterized in that, In step S1, the acquisition of the pen pressure sequence specifically involves: acquiring the pen pressure sequence by collecting the pressure value when the student writes on the student tablet screen using an electromagnetic pen, with a sampling frequency of not less than 100Hz; the acquisition of the gaze focus area coordinates specifically involves: capturing the student's pupil position through the front-facing camera, and calculating the intersection of the student's gaze and the student tablet screen by combining the data from the posture sensor built into the student tablet, thus obtaining the gaze focus area coordinates; the acquisition of the device tilt angle sequence specifically involves: acquiring the tilt angle of the student tablet relative to the horizontal plane through the posture sensor built into the student tablet.

5. The method according to claim 2, characterized in that, In step S3, when the local generative assistance mechanism is triggered, the cognitive load index CLI is first analyzed. t The dominant component is elevated, and corresponding hierarchical prompts and / or appropriate intervention strategies are generated based on the type of dominant component: If it is a pressure fluctuation weighting item In CLI t If the contribution rate exceeds 50%, it is judged as anxiety-type load. The hierarchical prompts generated by the large language model include emotional reassurance statements and more granular guidance content with step breakdown generated for the current learning task; if the focus is distracted, the weighting factor is... In CLI t If the contribution of the input exceeds 50%, it is judged as a distraction load. The hierarchical prompts generated by the large language model include interactive questions or highlighted / blinking prompts related to the current learning content to recapture attention; if the posture deviation weight term... In CLI t If the contribution of the content on the screen exceeds 50%, it is determined to be a fatigue-type load. The content currently displayed on the screen will be forcibly switched to voice broadcast mode to reduce the need for screen reading. When the pressure fluctuation weight term, the line of sight distraction weight term, and the posture deviation weight term are in CLI t When the contribution of each factor does not exceed 50%, it indicates that the student's cognitive load is caused by a combination of factors, without a single dominant factor. This is classified as a comprehensive load. A comprehensive, layered guidance prompt is generated using a large language model, including mild emotional reassurance statements, concise summaries of the core knowledge points of the current learning task, audio explanations, and highlighting key steps. This avoids the secondary cognitive pressure caused by reading long passages of text. Simultaneously, a combined intervention strategy of information degradation and micro-rests is employed. Information degradation: Reduce the content display density on the student tablet interface and simplify the interactive operation to a single-step click confirmation mode to lower the operation threshold; Micro-break intervention: Forcefully pause the current learning task for 5 to 10 seconds, play soothing sound effects or simple breathing guidance animations on the interface to help students relax and reset from the overall tense learning state, and resume after the countdown ends.

6. The method according to claim 2, characterized in that, In step S4, the calculation logic for the intervention effectiveness score is as follows: if, after receiving generative assistance, the average cognitive load index of the student decreases by more than 20% compared to before the intervention within a set continuous period, and the correct answer rate remains unchanged or increases, then the intervention effectiveness score is 1; if the cognitive load index does not decrease or the correct answer rate decreases, then the intervention effectiveness score is 0; if the cognitive load index decreases but the decrease does not exceed 20%, and the correct answer rate remains unchanged or increases, then the intervention effectiveness score is defined as 0.5, representing "partially effective"; based on the intervention effectiveness score, the gradient descent method is used to update the weight parameters w1, w2, and w3 in the cognitive load calculation model, so that the cognitive load index can better reflect the student's true state.

7. The method according to claim 2, characterized in that, In step S4, the specific process of updating the weight parameters w1, w2, and w3 in the cognitive load calculation model using the gradient descent method based on the intervention effectiveness score is as follows: A1. Define the loss function: First, convert the intervention effectiveness score into an error value that needs to be optimized, and set the error = 1 - intervention effectiveness score; if the intervention effectiveness score is 0, the error is the largest, which indicates that the weight allocation of the current cognitive load calculation model is unreasonable and needs to be adjusted significantly. If the score is 1 and the error is 0, the weights of the cognitive load calculation model remain unchanged; A2. Determine the direction of adjustment: Review the pre-intervention period. , , The specific value; if the error is large, it means that the cognitive load calculation model misdiagnosed the student's true state, that is, it assigned too high a weight to the feature with the largest value at that time; therefore, the adjustment direction is set as follows: reduce the weight of the feature with the largest value before the intervention, and correspondingly increase the weight of the other two feature items. A3. Calculate the initial new weights: Introduce a preset learning rate, and calculate the initial update values ​​of w1, w2, and w3 respectively using the formula Initial New Weights = Old Weights - Learning Rate × Gradient; A4. Normalization: Since w1 + w2 + w3 = 1 is strictly required, after obtaining the initial new weights, these three values ​​need to be divided by their sum respectively.

8. The method according to claim 1, characterized in that, When storing data, the student tablet only saves statistical indicators after feature extraction, including the daily average cognitive load index, the total number of triggers of the generative auxiliary mechanism, and the final convergence value of each weight parameter of the cognitive load calculation model. The original biometric data, including the original value of the pen pressure sequence and the gaze focus image frame, are destroyed in memory immediately. Before being stored in the storage medium, the statistical indicators are encrypted using the AES-256 standard and can only be decrypted and exported after obtaining the authorization key of the parent bound to the student tablet, ensuring that the student's biometric privacy data is not leaked.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method described in any one of claims 1-8.

10. A student tablet, characterized in that, It includes a processor and a memory, the memory being used to store a computer program; the processor being used to execute the computer program to implement the method of any one of claims 1-8.

Images