Student cognitive diagnosis method based on data forgetting

Abstract

The invention discloses a student cognitive diagnosis method based on data forgetting, and the method comprises the steps: obtaining the cognitive state information of a student through model pre-training, carrying out the data partitioning of the student according to the similarity, and independently carrying out the model training on each partition. And during student score prediction, calculating the similarity between the student and other students in each partition, further adjusting the prediction weight of each partition model, and finally aggregating and outputting a score prediction result. According to the forgetting framework, after the sensitive data is deleted, the partition data can be quickly retrained, and the training time is remarkably shortened. According to the method, the forgetting framework is constructed according to the cognitive state of the students on the knowledge points, so that the model is prevented from being interfered by dirty data and invalid data, and cognitive diagnosis is completed more accurately. According to the invention, the current cognitive state of the student can be diagnosed more accurately by clearing the outdated data in time.

Description

Technical Field

[0001] The present invention relates to the field of educational big data, and in particular to a student cognitive diagnosis method based on data forgetting. Background Art

[0002] With the rapid development of information technology, the scale of educational data continues to expand, covering students' learning behavior data, test scores, feedback and other information. The generation and application of educational big data has brought unprecedented opportunities and challenges to personalized learning. How to effectively manage and utilize educational big data and safely and efficiently mine the information behind educational data has become the focus of model development in the education field. For the protection of student user privacy data, it is necessary not only to prevent data leakage, but also to consider the problem of forgetting privacy data to prevent the incomplete deletion of privacy data parameters in the training model.

[0003] As an advanced educational assessment tool, cognitive diagnostic technology can help educators accurately identify students' learning shortcomings and provide targeted teaching feedback by carefully analyzing students' knowledge mastery and cognitive abilities during the learning process. In contrast, traditional examinations and assessment methods are difficult to fully and carefully reflect students' cognitive status. Teachers often face the challenge of how to timely understand students' learning progress and effectively provide guidance for their weak links. Applying cognitive diagnostic technology to the assessment of students' learning can not only provide a deeper understanding of students' mastery of specific knowledge points, but also provide a scientific basis for the design of personalized teaching and learning paths.

[0004] As the number of student users continues to increase, cognitive diagnosis models are increasingly used in education systems, but they also bring significant information security risks. A large amount of students' personal information and learning behavior data are centrally stored and processed, which makes the system face a greater risk of data leakage. Traditional data protection measures are difficult to cope with the complex scenarios of large-scale data storage and model parameter updates, resulting in potential security vulnerabilities that may endanger students' privacy and data security while providing accurate education services. Existing cognitive diagnosis models fail to delete model parameters related to private data while updating and iterating user private data, facing the risk of privacy leakage. In addition, cognitive models face the following major challenges when forgetting private data: (1) The impact of data deletion on model accuracy: Cognitive diagnosis models rely on a large amount of historical data to establish and optimize the model structure. Once some data is deleted, the model's predictive ability may decline, resulting in reduced overall performance; (2) The efficiency of model retraining: After performing data forgetting, the model may need to be retrained or fine-tuned to adapt to the new state after data deletion. Especially when dealing with large-scale data, this process can be very time-consuming; (3) The difficulty of completely deleting data: In complex cognitive diagnostic models, data often exists in various forms, such as direct input, derived features, model parameters, etc. To completely delete a user's data, it may be necessary to operate at multiple levels of the model, which makes it difficult to ensure the integrity of data forgetting. How to ensure that all user-related privacy data is completely cleared is the core issue of achieving data forgetting; (4) Implementation of compliance requirements: In different regions and legal frameworks, the requirements for data forgetting may vary, and cognitive diagnostic models need to comply with the provisions of these laws and regulations. How to balance different compliance requirements in technical implementation and ensure the legality of data forgetting is an important challenge.

[0005] In this context, machine unlearning, as an emerging technical method, has gradually become an important means to solve privacy protection and data management problems. Machine unlearning aims to effectively delete specific data records from the model to prevent these records from affecting the training and prediction of the model. This technology is of great significance in protecting user privacy and complying with data protection regulations such as GDPR. For example, when students request to delete their personal data, how to effectively protect user privacy and meet increasingly stringent data regulations while efficiently forgetting data and ensuring that the model performance does not show a significant decline after forgetting has become a difficult problem that needs to be solved urgently. Summary of the invention

[0006] The purpose of the present invention is to provide a student cognitive diagnosis method based on data forgetting, which ensures the accuracy and stability of the prediction model while protecting privacy, and provides strong technical support for students' personalized learning.

[0007] The technical solution adopted by the present invention is: A student cognitive diagnosis method based on data forgetting includes the following steps: S1, modeling the student's answer information through the selected pre-training algorithm to obtain the student's current knowledge state, and processing the student's current cognitive state into a knowledge point vector that reflects the student's mastery of knowledge points; Knowledge point vector this The students’ proficiency in each knowledge point is recorded and then stored for use in subsequent steps; S2, select N The seed user is taken as the center of the sub-dataset by calculating the student knowledge point vector this The similarity of cognitive states between other users and the central user in the user data is divided into N sub-datasets.

[0008] S3, using the selected student cognitive diagnosis algorithm to perform model training, train an independent sub-model for each sub-dataset, and ensure that each sub-model independently learns the data in the corresponding sub-dataset; S4, aggregate the outputs of each sub-model using the preset weights to generate a preliminary score prediction model; S5, deleting the sub-model containing the private data and the related data in the corresponding sub-dataset according to the data forgetting request; S6, retraining the sub-model modified due to the deletion of privacy data using the selected student cognitive diagnosis algorithm, and then re-aggregating the sub-models according to the current sub-model weights; S7, when predicting the exercise score of a specified student, first use the pre-training algorithm to obtain the latest cognitive status of the specified student and generate the corresponding student knowledge point vector; S8, check whether there is a relevant answer record of the specified student in the training data set; if yes, execute S9; otherwise, directly execute S10; S9, find the sub-dataset where the specified student is located, and set the corresponding sub-model weight to 1, and set the weights of other sub-models to 0; according to the current sub-model weights, aggregate the prediction scores of each sub-model; S10, calculating the cognitive similarity between the designated student and the central user of each sub-model, and dynamically adjusting the weight of each sub-model according to the cognitive similarity; S11, each sub-model predicts the score of the designated student, and then dynamically adjusts the predicted score according to the sub-model weight to obtain the final predicted score of the designated student.

[0009] Furthermore, in step S5, when it is necessary to delete the private data, a data forgetting request is generated to determine the sub-model and sub-dataset location where the private data is located.

[0010] Further, in step S1, through the pre-training algorithm algorithm 1 Get the student's cognitive state and generate a length of K Pre-trained student knowledge point vector this ; in the knowledge point vector K The elements correspond to the students' K The value of each element represents the student's proficiency in the knowledge point, thus reflecting the student's overall cognitive status.

[0011] Furthermore, step S2 specifically includes the following steps: S2-1, first use the student knowledge point vector generated by pre-training this , calculate the Pearson correlation coefficient between each remaining user and each central user to measure the cognitive similarity between users; S2-2, based on the calculated cognitive similarity, the remaining users are sorted from large to small according to their similarity with each central user; S2-3, add user information to the corresponding sub-datasets in sequence until the amount of data in the sub-datasets reaches the preset maximum amount of data S max ; and so on until completion N In this way, each sub-dataset is ensured to contain the user data with the highest cognitive similarity to the central user, thereby improving the training effect and prediction accuracy of the sub-model.

[0012] Furthermore, in steps S1 and S7, a pre-training algorithm is used to train the training data set. S Perform model pre-training, the purpose of which is to generate pre-trained student knowledge point vectors this , so that the data set can be divided more accurately in the subsequent steps. Using cognitive diagnosis algorithm algorithm 2 Sub-model training is performed on each sub-dataset to generate a score prediction model. Using different cognitive diagnosis algorithms before and after can optimize the use of computing resources at different stages, especially on large-scale data sets, making the entire training process more efficient. In addition, it will reduce the possibility of model overfitting and the dependence on specific algorithms; Furthermore, step S5 specifically includes the following steps: S5-1, first determine the user data that needs to be forgotten, including the user's personal information, answer records or other sensitive data, and accurately locate the location of the data in the model through a unique identifier (such as user ID or record ID); S5-2, determining the sub-dataset and sub-model where the forgotten user data is located, and generating a corresponding forget request; Furthermore, the forget request should contain key information such as the data identifier to be deleted, the sub-dataset where the data is located, and the related sub-models.

[0013] Furthermore, to ensure the traceability of the data deletion process, the generated forgetting request is recorded in a log for later review and verification.

[0014] Furthermore, step S10 specifically includes the following steps: S10-1, firstly, based on the calculated similarity, a nonlinear function is used to map the similarity to a weight, so as to ensure that a sub-model with a higher similarity is assigned a larger weight; S10-2, normalizing the initially allocated weights to ensure that the sum of all weights is 1.

[0015] In addition, regularization terms are introduced to prevent certain sub-model weights from being too high or too low, thereby avoiding affecting the overall prediction performance. In each new prediction task, the weights are dynamically adjusted based on the latest student cognitive status and historical feedback data to continuously optimize the prediction effect.

[0016] Furthermore, in step S11, each sub-model performs a weighted average on the predicted scores of the designated students according to the adjusted weights to generate a final predicted score.

[0017] The present invention adopts the above technical scheme, so that the cognitive diagnosis model can quickly forget data or update the model while ensuring the privacy security of users, and better adapt to the high frequency of updating of educational big data. The method is based on pre-training and sub-model division strategies, and by dynamically adjusting the sub-model weights, it not only effectively alleviates the problem of model performance degradation caused by data deletion, but also greatly reduces the cost of model data update. With the changes in students' cognitive states and the continuous accumulation of historical data, the present invention can continuously optimize the prediction effect. In addition to the utility, the present invention ensures the thoroughness of privacy data deletion by generating forgetting requests and accurately locating the location of data to be deleted. In addition, the dynamic weight adjustment and retraining mechanism introduced ensures the model performance after the privacy data is deleted. Through the user data division and dynamic adjustment mechanism based on similarity, the present invention ensures the accuracy and stability of the prediction model while protecting privacy, and provides strong technical support for students' personalized learning. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments; Figure 1 A structural diagram of the data forgetting method for the cognitive diagnosis process of students; Figure 2 A flowchart of the data forgetting square score prediction process for the student cognitive diagnosis process; Figure 3 Schematic diagram of the forgetting process flow of the data forgetting method for the student cognitive diagnosis process. Implementation

[0019] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0020] like Figure 1 to Figure 3 As shown in one, the present invention discloses a student cognitive diagnosis method based on data forgetting, comprising the following steps: S1, modeling the student's answer information through the selected pre-training algorithm to obtain the student's current knowledge state, and processing the student's current cognitive state into a knowledge point vector that reflects the student's mastery of knowledge points; Knowledge point vector this The students’ proficiency in each knowledge point is recorded and then stored for use in subsequent steps; S2, select N The seed user is taken as the center of the sub-dataset by calculating the student knowledge point vector this The similarity of cognitive states between other users and the central user in the user data is divided into N sub-datasets.

[0021] S3, using the selected student cognitive diagnosis algorithm to perform model training, train an independent sub-model for each sub-dataset, and ensure that each sub-model independently learns the data in the corresponding sub-dataset; S4, aggregate the outputs of each sub-model using the preset weights to generate a preliminary score prediction model; S5, deleting the sub-model containing the private data and the related data in the corresponding sub-dataset according to the data forgetting request; S6, retraining the sub-model modified due to the deletion of privacy data using the selected student cognitive diagnosis algorithm, and then re-aggregating the sub-models according to the current sub-model weights; S7, when predicting the exercise score of a specified student, first use the pre-training algorithm to obtain the latest cognitive status of the specified student and generate the corresponding student knowledge point vector; S8, check whether there is a relevant answer record of the specified student in the training data set; if yes, execute S9; otherwise, directly execute S10; S9, find the sub-dataset where the specified student is located, and set the corresponding sub-model weight to 1, and set the weights of other sub-models to 0; according to the current sub-model weights, aggregate the prediction scores of each sub-model; S10, calculating the cognitive similarity between the designated student and the central user of each sub-model, and dynamically adjusting the weight of each sub-model according to the cognitive similarity; S11, each sub-model predicts the score of the designated student, and then dynamically adjusts the predicted score according to the sub-model weight to obtain the final predicted score of the designated student.

[0022] Furthermore, in step S5, when it is necessary to delete the private data, a data forgetting request is generated to determine the sub-model and sub-dataset location where the private data is located.

[0023] Further, in step S1, through the pre-training algorithm algorithm 1 Get the student's cognitive state and generate a length of K Pre-trained student knowledge point vector this ; in the knowledge point vector K The elements correspond to the students' K The value of each element represents the student's proficiency in the knowledge point, thus reflecting the student's overall cognitive status.

[0024] Furthermore, step S2 specifically includes the following steps: S2-1, first use the student knowledge point vector generated by pre-training this , calculate the Pearson correlation coefficient between each remaining user and each central user to measure the cognitive similarity between users; S2-2, based on the calculated cognitive similarity, the remaining users are sorted from large to small according to their similarity with each central user; S2-3, add user information to the corresponding sub-datasets in sequence until the amount of data in the sub-datasets reaches the preset maximum amount of data S max ; and so on until completion N In this way, each sub-dataset is ensured to contain the user data with the highest cognitive similarity to the central user, thereby improving the training effect and prediction accuracy of the sub-model.

[0025] Furthermore, in steps S1 and S7, a pre-training algorithm is used to train the training data set. SPerform model pre-training, the purpose of which is to generate pre-trained student knowledge point vectors this , so that the data set can be divided more accurately in the subsequent steps. Using cognitive diagnosis algorithm algorithm 2 Sub-model training is performed on each sub-dataset to generate a score prediction model. Using different cognitive diagnosis algorithms before and after can optimize the use of computing resources at different stages, especially on large-scale data sets, making the entire training process more efficient. In addition, it will reduce the possibility of model overfitting and the dependence on specific algorithms; Furthermore, step S5 specifically includes the following steps: S5-1, first determine the user data that needs to be forgotten, including the user's personal information, answer records or other sensitive data, and accurately locate the location of the data in the model through a unique identifier (such as user ID or record ID); S5-2, determining the sub-dataset and sub-model where the forgotten user data is located, and generating a corresponding forget request; Furthermore, the forget request should contain key information such as the data identifier to be deleted, the sub-dataset where the data is located, and the related sub-models.

[0026] Furthermore, to ensure the traceability of the data deletion process, the generated forgetting request is recorded in a log for later review and verification.

[0027] Furthermore, step S10 specifically includes the following steps: S10-1, firstly, based on the calculated similarity, a nonlinear function is used to map the similarity to a weight, so as to ensure that a sub-model with a higher similarity is assigned a larger weight; S10-2, normalizing the initially allocated weights to ensure that the sum of all weights is 1.

[0028] In addition, regularization terms are introduced to prevent the weights of some sub-models from being too high or too low, thereby avoiding affecting the overall prediction performance. In each new prediction task, the weights are dynamically adjusted according to the latest student cognitive status and historical feedback data to continuously optimize the prediction effect.

[0029] Furthermore, in step S11, each sub-model performs a weighted average on the predicted scores of the designated students according to the adjusted weights to generate a final predicted score.

[0030] The specific principle of the present invention is described in detail below: like Figure 1As shown in Figure 1, most student cognitive diagnosis models are divided into four main parts according to the process: pre-training part, data partitioning based on cognitive similarity, sub-model training part, and score prediction aggregation part. 1. Pre-training part. Goal: Generate a vector that reflects the student's knowledge mastery. this Process: Model the student’s answer information through the selected pre-training algorithm to obtain the student’s current cognitive state, and process it into a vector that reflects the student’s mastery of knowledge points. this , these vectors record the students' proficiency in each knowledge point, and then store them for use in subsequent steps; 2. Data division based on cognitive similarity. Goal: Divide user data into several sub-datasets. Process: Determine the number of sub-models N and the maximum capacity of each subdataset S max . Select N The seed user is taken as the center of the sub-dataset by calculating the student knowledge point vector this The similarity of cognitive states between other users and these central users in the user data is divided into N 3. Sub-model training part. Objective: Train an independent sub-model for each sub-dataset. Process: Use the student cognitive diagnosis algorithm algorithm 2 Perform model training to ensure that each sub-model independently learns the data in its corresponding sub-dataset; 4. Score prediction aggregation part. Goal: Generate a student's score prediction model. Process: After completing the sub-model training, use the preset weights to aggregate the outputs of each sub-model to generate a preliminary score prediction model. This model combines the prediction capabilities of all sub-models and improves the overall prediction effect by dynamically adjusting the weights.

[0031] For the score prediction process of the data forgetting method for the student cognitive diagnosis process, see Figure 2 The process diagram is shown.

[0032] 1. Student status update. Through pre-training algorithm algorithm 1 Get the specified student u The latest cognitive state of the student is converted into a knowledge point vector this u . The length of this vector is K , each element represents the student's mastery of a specific knowledge point. 2. Record checking and weight adjustment. The system checks the student's uIs there a historical answer record in the training data set? If so, directly find the sub-dataset where the student is located, set the weight of the corresponding sub-model to 1, and the weights of other sub-models to 0. This can maximize the use of the student's historical data for prediction and reduce the interference of irrelevant data. If there is no historical record, calculate the student's u The similarity of cognitive status with the central user of each sub-model is mapped to weight using a nonlinear mapping function for weight distribution. This ensures that students with high similarity can get higher weights, thus affecting the final prediction results. Dynamic weight update helps the model adjust the prediction strategy according to the latest cognitive status. 3. Score prediction and aggregation. Each sub-model independently predicts the student based on the current weight. u The student score is generated by taking a weighted average of the prediction scores of each sub-model (using regularization to prevent excessive influence of a single sub-model). u This process ensures the robustness and accuracy of the prediction results.

[0033] For the data forgetting process of the data forgetting method for the student cognitive diagnosis process, see Figure 3 The process diagram is shown.

[0034] 1. Generate a forget request. When it is necessary to delete a user's private data, a forget request is generated first. Use a unique identifier (such as a user ID or record ID) to accurately locate the data to be deleted. This data may include sensitive information such as personal information and answer records. The forget request should record in detail the data identifier to be deleted, the sub-dataset where the data is located, and the relevant sub-model information. This information will be used to accurately delete the data in subsequent steps. 2. Locate and delete data. According to the forget request, locate the sub-dataset and sub-model containing the private data. Subsequently, a specially designed deletion algorithm is used to ensure that the data is completely deleted to avoid any potential possibility of recovery. To ensure the traceability of the entire data forgetting process, the system records each forget request generated in the log. This helps to review the completeness and accuracy of data deletion in the future. 3. Sub-model retraining and re-aggregation. Since data deletion will result in incomplete training data sets for sub-models, the affected sub-models need to be retrained. During the retraining process, the same algorithm as the initial training is used. algorithm 2 , ensuring that the new model can adapt to data changes. The retrained sub-models will be aggregated again based on the current sub-model weights. In this way, the system can maintain the predictive power and performance stability of the overall model while deleting private data.

[0035] By integrating machine forgetting learning technology into the cognitive diagnosis model, the present invention can achieve the complete deletion of user data and model parameters, improve the level of privacy protection, and ensure that the system complies with relevant laws and regulations. In addition, the technology can also dynamically adjust the data distribution and remove outdated or invalid data, thereby further improving the accuracy and efficiency of the diagnostic model. More importantly, the forgetting mechanism can optimize the system's storage and computing resources so that it can still operate efficiently when processing large amounts of data. Using this technology, the diagnostic model can not only cope with the needs of scenarios where student data is frequently updated, but also track students' cognitive status in more real time, expanding its application scenarios in the field of educational technology.

[0036] Obviously, the described embodiments are part of the embodiments of the present application, rather than all of the embodiments. In the absence of conflict, the embodiments in the present application and the features in the embodiments can be combined with each other. The components of the embodiments of the present application generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the detailed description of the embodiments of the present application is not intended to limit the scope of the application claimed for protection, but merely represents the selected embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians of the art without making creative work are within the scope of protection of the present application.

Claims

1. A student cognitive diagnosis method based on data forgetting, characterized by: It includes the following steps: S1, modeling the student's answer information through the selected pre-training algorithm to obtain the student's current knowledge state, and processing the student's current cognitive state into a knowledge point vector that reflects the student's mastery of knowledge points; Knowledge point vector Stu The students’ proficiency in each knowledge point is recorded and then stored for use in subsequent steps; S2, select N The seed user is taken as the center of the sub-dataset by calculating the student knowledge point vector Stu The similarity of cognitive states between other users and the central user in the user data is divided into N sub-datasets; S3, using the selected student cognitive diagnosis algorithm to perform model training, train an independent sub-model for each sub-dataset, and ensure that each sub-model independently learns the data in the corresponding sub-dataset; S4, aggregate the outputs of each sub-model using the preset weights to generate a preliminary score prediction model; S5, deleting the sub-model containing the private data and the related data in the corresponding sub-dataset according to the data forgetting request; S6, retraining the sub-model modified due to the deletion of privacy data using the selected student cognitive diagnosis algorithm, and then re-aggregating the sub-models according to the current sub-model weights; S7, when predicting the exercise score of a specified student, first use the pre-training algorithm to obtain the latest cognitive status of the specified student and generate the corresponding student knowledge point vector; S8, check whether there is a relevant answer record of the specified student in the training data set; if yes, execute S9; otherwise, directly execute S10; S9, find the sub-dataset where the specified student is located, and set the corresponding sub-model weight to 1, and set the weights of other sub-models to 0; according to the current sub-model weights, aggregate the prediction scores of each sub-model; S10, calculating the cognitive similarity between the designated student and the central user of each sub-model, and dynamically adjusting the weight of each sub-model according to the cognitive similarity; S11, each sub-model predicts the score of the designated student, and then dynamically adjusts the predicted score according to the sub-model weight to obtain the final predicted score of the designated student.

2. The method for student cognitive diagnosis based on data forgetting according to claim 1, characterized in that: In step S1, the pre-training algorithm algorithm 1 Get the student's cognitive state and generate a length of K Pre-trained student knowledge point vector Stu ; in the knowledge point vector K The elements correspond to the students' K The value of each element represents the student's proficiency in the knowledge point, thus reflecting the student's overall cognitive status.

3. The method for student cognitive diagnosis based on data forgetting according to claim 1, characterized in that: Step S2 specifically includes the following steps: S2-1, first use the student knowledge point vector generated by pre-training Stu , calculate the Pearson correlation coefficient between each remaining user and each central user to measure the cognitive similarity between users; S2-2, based on the calculated cognitive similarity, the remaining users are sorted from large to small according to their similarity with each central user; S2-3, add user information to the corresponding sub-datasets in sequence until the amount of data in the sub-datasets reaches the preset maximum amount of data S max ; and so on until completion N The division of the sub-datasets.

4. The method for student cognitive diagnosis based on data forgetting according to claim 1, characterized in that: In step S5, when the private data needs to be deleted, a data forgetting request is generated to determine the sub-model and sub-dataset location where the private data is located.

5. The method for student cognitive diagnosis based on data forgetting according to claim 4 is characterized by: Step S5 specifically includes the following steps: S5-1, first determine the user data that needs to be forgotten, including the user's personal information, answer records or other sensitive data, and accurately locate the location of the data in the model through a unique identifier; S5-2, determine the sub-dataset and sub-model where the forgotten user data is located, and generate a corresponding forget request.

6. The method for student cognitive diagnosis based on data forgetting according to claim 4, characterized in that: The forget request includes key information such as the data identifier to be deleted, the sub-dataset where the data is located, and the related sub-models.

7. The method for student cognitive diagnosis based on data forgetting according to claim 4 is characterized by: Record the generated forget requests in a log for later review and verification.

8. The method for student cognitive diagnosis based on data forgetting according to claim 1, characterized in that: Step S10 specifically includes the following steps: S10-1, firstly, based on the calculated similarity, a nonlinear function is used to map the similarity to a weight, so as to ensure that a sub-model with a higher similarity is assigned a larger weight; S10-2, normalizing the initially allocated weights to ensure that the sum of all weights is 1.

9. The method for student cognitive diagnosis based on data forgetting according to claim 1, characterized in that: In step S11, each sub-model performs a weighted average of the predicted scores of the designated students according to the adjusted weights to generate a final predicted score.

Images