A bias-free cognitive diagnostic method based on adaptive data augmentation

Abstract

This invention discloses a bias-reduction cognitive diagnosis method based on adaptive data augmentation, comprising: 1. Constructing heterogeneous data, including: student answer records for exercises and a relationship matrix between exercises and knowledge points; 2. Generating biased data through selective data augmentation, and inputting it along with student answer records into a bias-reduction cognitive diagnosis model, outputting bias prediction results, thereby constructing the symmetric cross-entropy loss of the bias-reduction cognitive diagnosis model; 3. Based on the prediction results output by the bias-reduction cognitive diagnosis model, adaptively generating unbiased data, and inputting it along with student answer records into the bias-reduction cognitive diagnosis model for processing, outputting bias-reduction prediction results, thereby constructing the weighted cross-entropy loss of the bias-reduction cognitive diagnosis model; 4. In the model inference stage, using the established bias-reduction cognitive diagnosis model to predict students' answering abilities. This invention can alleviate the bias problem in cognitive diagnosis.

Description

Technical Field

[0001] This invention belongs to the field of cognitive diagnosis in smart education, specifically a bias-free cognitive diagnosis method based on adaptive data augmentation. Background Technology

[0002] Cognitive diagnostics is a fundamental task in the field of smart education, aiming to infer students' mastery of knowledge points by utilizing their answers to exercises. It has been applied in various educational scenarios, such as exercise recommendation, knowledge tracking, and computer-adaptive testing.

[0003] Due to the sparsity of answer records, existing cognitive diagnostic methods fully utilize these records for student modeling. However, these methods inherit an inherent weakness of machine learning: a preference for learning simple features. The simple feature present in the answer records is the student's answer accuracy. For example, the model predicts that students with high answer accuracy will always answer questions correctly. This negatively impacts the accuracy and reliability of cognitive diagnosis, especially in out-of-distribution scenarios. For instance, in a province-wide joint exam, the question setters change, altering the question style and leading to a decrease in student accuracy. Over-reliance on student answer accuracy for modeling results in biased diagnosis. Therefore, addressing the bias in cognitive diagnostics for scenarios where modeling student representations using answer accuracy is biased is a worthwhile research topic. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention proposes a bias-free cognitive diagnosis method based on adaptive data augmentation. This method aims to alleviate the bias in cognitive diagnosis models that rely on answer accuracy for modeling, thereby more accurately predicting the weak knowledge points of students with high answer accuracy and the strong knowledge points of students with low answer accuracy, and accurately predicting students' answer performance.

[0005] To achieve the above-mentioned objectives, the present invention adopts the following technical solution:

[0006] The present invention provides a bias-free cognitive diagnostic method based on adaptive data augmentation, characterized by the following steps:

[0007] Step 1: Construct heterogeneous data, including: students' answer records for exercises. and the relationship matrix between exercises and knowledge points ;

[0008] Define the student set as The set of exercises is The knowledge point set is ;in, Indicates the first One student, Indicates the first One exercise, Indicates the first One knowledge point, Indicates the number of students. Indicates the number of exercises. Indicates the number of knowledge points. , , ;

[0009] Record students' answers to the exercises as follows: ,make Any triple in the set is denoted as ,in, Indicates the first students Answer number Exercises The corresponding answer tags, if Then it means the first students Answer number Exercises Correct; if Then it means the first students Answer number Exercises mistake;

[0010] Let the relationship matrix between exercises and knowledge points be denoted as... , among which, if Then it means the first Exercises With the One knowledge point Related; if Then it means the first Exercises With the One knowledge point Irrelevant;

[0011] The first students The set of questions that have been answered is denoted as ;

[0012] Based on students' correct answer rate, those whose correct answer rate exceeds a threshold will be... Students were divided into groups based on their accuracy in answering questions. The correct answer rate should not exceed the threshold. Students were divided into groups with lower accuracy rates on the questions. ;

[0013] Will The corresponding answer records with the answer tag "1" and The typical pattern consists of answer records with an answer tag of "0" in the corresponding answer records. ;

[0014] Will The corresponding answer records with an answer tag of "0" and The answer records with the answer tag "1" in the corresponding answer records constitute an atypical pattern. ;

[0015] Step 2: Define the current iteration number as... and initialize ;

[0016] Step 3, from typical patterns The middle generates the first through selective data augmentation Deviation data of round iteration and with The data are input into the cognitive bias diagnosis model for processing, and the first result is output. The deviation prediction results of the round of iterations are used to construct the deviation cognitive diagnosis model. Symmetric cross-entropy loss in round iterations ;

[0017] Step 4: Based on the output of the deviation cognitive diagnosis model... The prediction results from the first iteration are adaptively used to generate the second iteration. Unbiased data in round iteration and with The samples are input together into the biased cognitive diagnostic model for processing, and the first result is output. The biased prediction results from the rounds of iterations are used to construct the biased cognitive diagnostic model. Weighted cross-entropy loss in rounds of iteration ;

[0018] Step 5: Construct the first equation using equation (9). The overall objective function of the round iteration :

[0019] (9)

[0020] Step 6: Synchronous training of the cognitive diagnostic model:

[0021] Step 6.1: Use the Adam optimizer to perform the first step on the biased cognitive diagnostic model and the debiased cognitive diagnostic model. Train in rounds and minimize the first round. Loss function of round iteration To update the model parameter set and obtain the first... The biased cognitive diagnostic model after the first round of training and the first round of training A biased cognitive diagnostic model after rounds of training;

[0022] Step 6.2, Assign to Then, return to step 6.1 and execute sequentially until... The process continues until convergence, thus obtaining a well-established cognitive diagnostic model that corrects biases.

[0023] Step 6.3: In the model inference stage, the established bias-free cognitive diagnostic model is used to predict students' answering abilities and obtain the optimal answer prediction label corresponding to the students' answers to the exercises, so as to achieve cognitive diagnostic bias-free.

[0024] The characteristic of the bias-free cognitive diagnosis method based on adaptive data augmentation described in this invention is that step 3 is performed as follows:

[0025] Step 3.1, from typical patterns Randomly select the first Answer records of round iterations , will the The first iteration students One-hot encoding and the first The first iteration One exercise The one-hot encodings are respectively input into the deviation cognitive diagnosis model, and the corresponding outputs of the embedding layers in the deviation cognitive diagnosis model are then generated. The first iteration students The representation and the The first iteration One exercise The representation ;in, Indicates the first The first iteration students Answer number One exercise The corresponding answer tags;

[0026] Step 3.2, from the typical pattern Randomly select the answer records of the t-th iteration. And satisfy , will the The first iteration students One-hot encoding and the first The first iteration One exercise The one-hot encodings are respectively input into the deviation cognitive diagnosis model, and the corresponding outputs of the embedding layers in the deviation cognitive diagnosis model are then generated. The first iteration students The representation and the The first iteration One exercise The representation ;in, Indicates the first The first iteration students Answer number One exercise The corresponding answer tags;

[0027] Step 3.3: Use equation (1) to generate the first... Bias triplet of round iteration Thus, the first Deviation data of round iteration :

[0028] (1)

[0029] In equation (1), To show obedience The parameters of the distribution Indicates the first Deviation in the round of iterations The representation, Indicates the first Exercises on deviations in round iteration The representation, Indicating students with deviations For deviation exercises Answer records;

[0030] Step 3.4, and The data are input into the deviation cognitive diagnosis model for processing, and the corresponding output is the first... The first iteration students Answer number Exercises Corresponding deviation answer prediction label ;

[0031] Step 3.5: Construct a cognitive deviation diagnosis model using equation (2) in the first step. Symmetric cross-entropy loss in round iterations :

[0032] (2)

[0033] In equation (2), Indicates the first Cross-entropy loss in rounds of iteration, Indicates the first The inverse cross-entropy loss of the round of iterations, and we have:

[0034] (3)

[0035] (4).

[0036] Furthermore, step 4 is performed as follows:

[0037] Step 4.1: Calculate the answer record using formula (5). In the Exponential mean of cross-entropy in rounds of iteration :

[0038] (5)

[0039] In equation (5), It is momentum; Indicates the first The exponential mean of the cross-entropy in each round of iterations;

[0040] Step 4.2, use equation (6) to calculate the exponential mean. Normalization and truncation are used to obtain the answer record. In the Importance score of round iteration :

[0041] (6)

[0042] In equation (4), This represents the hyperparameter that controls the cutoff threshold;

[0043] Step 4.3, based on the first Importance score set of rounds of iteration and answer records From multiple distributions The middle sampling obtained the first Unbiased answer record of round iteration ;

[0044] Step 4.4, from the first Unbiased answer record of round iteration One answer record is randomly selected from the results. , will the The first iteration students One-hot encoding and the first The first iteration Exercises The one-hot encodings are respectively input into the debiased cognitive diagnostic model, and the corresponding embedding layers in the debiased cognitive diagnostic model output the first... The first iteration students The representation and the The first iteration One exercise The representation ;in, Indicates the first The first iteration students Answer number One exercise The corresponding answer tags;

[0045] Step 4.5, from the first Unbiased answer record of round iteration One answer record is randomly selected from the results. And satisfy , will the The first iteration students One-hot encoding and the first The first iteration One exercise The one-hot encodings are respectively input into the debiased cognitive diagnostic model, and the corresponding embedding layers in the debiased cognitive diagnostic model output the first... The first iteration students The representation and the The first iteration One exercise The representation ;in, Indicates the first The first iteration students answer One exercise The corresponding answer tags;

[0046] Step 4.6: Use equation (7) to generate the first... Unbiased triples in round-recursion Thus, the first Unbiased data in round iteration :

[0047] (7)

[0048] In equation (7), To show obedience The parameters of the distribution Indicates the first Unbiased students in rounds of iteration The representation, Indicates the first Unbiased exercises in round iteration The representation, Indicates an unbiased student For unbiased exercises Answer records;

[0049] Step 4.7, and The inputs are respectively fed into the biased cognitive diagnostic model for processing, and the corresponding outputs are... The first iteration students Answer number Exercises The corresponding biased answer prediction tags ;

[0050] Step 4.8: Construct a biased cognitive diagnostic model using equation (8) in the first step. Weighted cross-entropy loss in rounds of iteration :

[0051] (8).

[0052] The present invention provides an electronic device, including a memory and a processor, wherein the memory is used to store a program that supports the processor in executing the debiased cognitive diagnostic method, and the processor is configured to execute the program stored in the memory.

[0053] The present invention discloses a computer-readable storage medium on which a computer program is stored, wherein the computer program is executed by a processor to perform the steps of the bias-reduction cognitive diagnosis method.

[0054] Compared with existing technologies, the beneficial effects of this invention are reflected in:

[0055] 1. The cognitive diagnostic bias model introduced in this invention can effectively simulate the bias amplification phenomenon: The biased cognitive diagnostic model proposed in this invention can accurately simulate the bias in model prediction caused by students' answer accuracy. To address the problem of answer modeling over-reliance on answer accuracy, this invention employs a specific data augmentation strategy to generate biased data, thereby enhancing the model's sensitivity to answer accuracy bias, and introduces symmetric cross-entropy loss to amplify this bias characteristic. This bias amplification simulation helps educational applications identify model preferences that may lead to misdiagnosis, and helps optimize the cognitive diagnostic model, making it more robust.

[0056] 2. This invention proposes a training strategy based on a biased cognitive diagnostic model to assist in the debiasing cognitive diagnostic model, which improves the debiasing effect and enhances the accuracy of cognitive diagnosis. During the student modeling process, the cross-entropy of each training sample in the biased cognitive diagnostic model is used to identify unbiased samples to guide adaptive data augmentation and sample reweighting. This enables the model to more accurately diagnose the weak knowledge points of students with high answer accuracy and the proficient knowledge points of students with low answer accuracy. It is applicable to out-of-distribution scenarios, realizes cognitive diagnosis debiasing, and thus more comprehensively supports the field of smart education. Attached Figure Description

[0057] Figure 1 This is a flowchart of the bias removal method based on adaptive data augmentation for cognitive diagnosis according to the present invention. Detailed Implementation

[0058] In this embodiment, an adaptive data augmentation-based de-biased cognitive diagnosis method utilizes the idea of ​​a biased cognitive diagnosis model assisting a de-biased cognitive diagnosis model. It selects answer records for data augmentation in the de-biased cognitive diagnosis model based on the entropy of the sample within the biased cognitive diagnosis model, thereby mitigating bias issues in cognitive diagnosis. Specifically, as... Figure 1 As shown, the procedure is as follows:

[0059] Step 1: Construct heterogeneous data, including: students' answer records for exercises. and the relationship matrix between exercises and knowledge points ;

[0060] Define the student set as The set of exercises is The knowledge point set is ;in, Indicates the first One student, Indicates the first One exercise, Indicates the first One knowledge point, Indicates the number of students. Indicates the number of exercises. Indicates the number of knowledge points. , , ;

[0061] Record students' answers to the exercises as follows: ,make Any triple in the set is denoted as ,in, Indicates the first students Answer number Exercises The corresponding answer tags, if Then it means the first students Answer number Exercises Correct; if Then it means the first students Answer number Exercises mistake;

[0062] Let the relationship matrix between exercises and knowledge points be denoted as... , among which, if Then it means the first Exercises With the One knowledge point Related; if Then it means the first Exercises With the One knowledge point Irrelevant;

[0063] The first students The set of questions that have been answered is denoted as ;

[0064] Based on students' correct answer rate, those whose correct answer rate exceeds a threshold will be... Students were divided into groups based on their accuracy in answering questions. The correct answer rate should not exceed the threshold. Students were divided into groups with lower accuracy rates on the questions. ;

[0065] In practical implementation, threshold ; It can also be set to other values ​​depending on the application scenario to meet the requirements. That's all;

[0066] Will The corresponding answer records with the answer tag "1" and The typical pattern consists of answer records with an answer tag of "0" in the corresponding answer records. ;

[0067] Will The corresponding answer records with an answer tag of "0" and The answer records with the answer tag "1" in the corresponding answer records constitute an atypical pattern. ;

[0068] Cognitive diagnostic models rely too heavily on students' answer accuracy to predict performance, resulting in good performance in typical patterns but poor performance in atypical patterns.

[0069] Step 2: Define the current iteration number as... and initialize ;

[0070] Step 3, from typical patterns The middle generates the first through selective data augmentation Deviation data of round iteration and with The data are input into the cognitive bias diagnosis model for processing, and the first result is output. The deviation prediction results of the round of iterations are used to construct the deviation cognitive diagnosis model. Symmetric cross-entropy loss in round iterations :

[0071] Step 3.1, from typical patterns Randomly select the first Answer records of round iterations , will the The first iteration students One-hot encoding and the first The first iteration One exercise The one-hot encodings are respectively input into the deviation cognitive diagnosis model, and the corresponding outputs of the embedding layers in the deviation cognitive diagnosis model are then generated. The first iteration students The representation and the The first iteration One exercise The representation ;in, Indicates the first The first iteration students Answer number One exercise The corresponding answer tags;

[0072] Step 3.2, from the typical pattern Randomly select the answer records of the t-th iteration. And satisfy , will the The first iteration students One-hot encoding and the first The first iteration One exercise The one-hot encodings are respectively input into the deviation cognitive diagnosis model, and the corresponding outputs of the embedding layers in the deviation cognitive diagnosis model are then generated. The first iteration students The representation and the The first iteration One exercise The representation ;in, Indicates the first The first iteration students Answer number One exercise The corresponding answer labels; in practice, this method is model-independent, and student and question representations can be obtained from commonly used cognitive diagnostic models, such as IRT, MIRT, NCDM and KaNCD.

[0073] Step 3.3: Use equation (1) to generate the first... Bias triplet of round iteration ,like Figure 1 The selective data augmentation module is shown, thus obtaining the first... Deviation data of round iteration :

[0074] (1)

[0075] In equation (1), To show obedience The parameters of the distribution Indicates the first Deviation in the round of iterations The representation, Indicates the first Exercises on deviations in round iteration The representation, Indicating students with deviations For deviation exercises Answer records; in specific implementation, The two parameters of the distribution are set to 1.

[0076] Step 3.4, and The data are input into the deviation cognitive diagnosis model for processing, and the corresponding output is the first... The first iteration students Answer number Exercises Corresponding deviation answer prediction label ;

[0077] Step 3.5: To further amplify the bias of the biased cognitive diagnostic model, the biased cognitive diagnostic model is constructed using equation (2) in the first step. Symmetric cross-entropy loss in round iterations :

[0078] (2)

[0079] In equation (2), Indicates the first Cross-entropy loss in rounds of iteration, Indicates the first The inverse cross-entropy loss of the round of iterations, and we have:

[0080] (3)

[0081] (4)

[0082] Step 4: Based on the output of the deviation cognitive diagnosis model... The prediction results from the first iteration are adaptively used to generate the second iteration. Unbiased data in round iteration and with The samples are input together into the biased cognitive diagnostic model for processing, and the first result is output. The biased prediction results from the rounds of iterations are used to construct the biased cognitive diagnostic model. Weighted cross-entropy loss in rounds of iteration :

[0083] Step 4.1: To more stably correct the bias, use equation (5) to calculate the answer record. In the Exponential mean of cross-entropy in rounds of iteration :

[0084] (5)

[0085] In equation (5), It is momentum; Indicates the first The exponential mean of the cross-entropy in each iteration; in practice, momentum. ;momentum It can also be tuned as a hyperparameter.

[0086] Step 4.2, use equation (6) to calculate the exponential mean. Normalization and truncation are used to obtain the answer record. In the Importance score of round iteration :

[0087] (6)

[0088] In equation (4), This represents the hyperparameter that controls the cutoff threshold;

[0089] Step 4.3, based on the first Importance score set of rounds of iteration and answer records From multiple distributions The middle sampling obtained the first Unbiased answer record of round iteration In practice, when sampling from multiple distributions, importance scores are considered. Normalize so that the sum of the elements of the vector is .

[0090] Step 4.4, from the first Unbiased answer record of round iteration One answer record is randomly selected from the results. , will the The first iteration students One-hot encoding and the first The first iteration Exercises The one-hot encodings are respectively input into the debiased cognitive diagnostic model, and the corresponding embedding layers in the debiased cognitive diagnostic model output the first... The first iteration students The representation and the The first iteration One exercise The representation ;in, Indicates the first The first iteration students Answer number One exercise The corresponding answer tags;

[0091] Step 4.5, from the first Unbiased answer record of round iteration One answer record is randomly selected from the results. And satisfy , will the The first iteration students One-hot encoding and the first The first iteration One exercise The one-hot encodings are respectively input into the debiased cognitive diagnostic model, and the corresponding embedding layers in the debiased cognitive diagnostic model output the first... The first iteration students The representation and the The first iteration One exercise The representation ;in, Indicates the first The first iteration students answer One exercise The corresponding answer labels; in practice, this method is model-independent, and student and question representations can be obtained from commonly used cognitive diagnostic models, such as IRT, MIRT, NCDM and KaNCD.

[0092] Step 4.6: Use equation (7) to generate the first... Unbiased triples in round-recursion ,like Figure 1 The sample is shown in the mixup module, thus obtaining the first... Unbiased data in round iteration :

[0093] (7)

[0094] In equation (7), To show obedience The parameters of the distribution Indicates the first Unbiased students in rounds of iteration The representation, Indicates the first Unbiased exercises in round iteration The representation, Indicates an unbiased student For unbiased exercises Answer records;

[0095] Step 4.7, and The inputs are respectively fed into the biased cognitive diagnostic model for processing, and the corresponding outputs are... The first iteration students Answer number Exercises The corresponding biased answer prediction tags In practice, The two parameters of the distribution are set as follows: .

[0096] Step 4.8: Construct a biased cognitive diagnostic model using equation (8) in the first step. Weighted cross-entropy loss in rounds of iteration :

[0097] (8)

[0098] In specific implementation, the first Unbiased data in round iteration Corresponding importance score As shown in equation (7), not only will the answer records be mixed, but also the importance scores of the answer records will be mixed.

[0099] Step 5: Construct the first equation using equation (9). The overall objective function of the round iteration :

[0100] (9)

[0101] Step 6: Synchronous training of the cognitive diagnostic model:

[0102] Step 6.1: Calculate the symmetric cross-entropy loss of the biased cognitive diagnostic model according to equation (2); calculate the answer record at the [number]th ... The importance scores of each iteration are used to calculate the weighted cross-entropy loss of the biased cognitive diagnostic model according to equation (8); the Adam optimizer is then used to perform the first iteration on the biased cognitive diagnostic model and the biased cognitive diagnostic model. Train in rounds and minimize the first round. Loss function of round iteration To update the model parameter set and obtain the first... The biased cognitive diagnostic model after the first round of training and the first round of training A biased cognitive diagnostic model after rounds of training;

[0103] Step 6.2, Assign to Then, return to step 6.1 and execute sequentially until... The process continues until convergence, thus obtaining a well-established cognitive diagnostic model that corrects biases.

[0104] Step 6.3: In the model inference stage, the established bias-free cognitive diagnostic model is used to predict students' answering abilities and obtain the optimal answer prediction label corresponding to the students' answers to the exercises, so as to achieve cognitive diagnostic bias-free.

[0105] In this embodiment, an electronic device includes a memory and a processor. The memory stores a program that supports the processor in executing the above-described method, and the processor is configured to execute the program stored in the memory.

[0106] In this embodiment, a computer-readable storage medium stores a computer program, which is executed by a processor to perform the steps of the above method.

[0107] Example:

[0108] To evaluate the method's performance, experiments were conducted on three datasets: ASSIST2009, NIPS2020EC, and PISA2018. First, the datasets were randomly divided into... The dataset was divided into training, validation, and test sets. To verify the model's debiasing performance, some typical patterns from the test set were randomly removed, resulting in a ratio of typical to atypical patterns in the test set. If the model relies too heavily on predicting the accuracy of answers and fails to perform well on the test set, this will validate the model's bias removal performance. Three metrics were used in the test set: Area Under Curve (AUC), Accuracy (ACC), and Root Mean Square Error (RMSE).

[0109] This invention selected four cognitive diagnostic models as the backbone models: Item Response Theory (IRT), Multidimensional Item Response Theory (MIRT), Neural Cognitive Diagnosis Model (NCDM), and Knowledge-association based Neural Cognitive Diagnosis (KaNCD). Four bias removal methods were used for performance comparison: Reweight, Learning from Failure (LfF), Selected Mixup (SelecMix), and BiasEnsemble (BE). Specifically, Table 1 presents the experimental results. The best score is shown in bold.

[0110] Table 1. Comparison of diagnostic results between the method of the present invention and the comparative method.

[0111]

[0112] As can be seen from Table 1, on the three datasets, the MixCD method proposed in this invention significantly outperforms the comparative methods in terms of AUC, ACC, and RMSE.

Claims

1. A bias-free cognitive diagnostic method based on adaptive data augmentation, characterized in that, The procedure is as follows: Step 1: Construct heterogeneous data, including: students' answer records for exercises. and the relationship matrix between exercises and knowledge points ; Define the student set as The set of exercises is The knowledge point set is ;in, Indicates the first One student, Indicates the first One exercise, Indicates the first One knowledge point, Indicates the number of students. Indicates the number of exercises. Indicates the number of knowledge points. , , ; Record students' answers to the exercises as follows: ,make Any triple in the set is denoted as ,in, Indicates the first students Answer number Exercises The corresponding answer tags, if Then it means the first students Answer number Exercises Correct; if Then it means the first students Answer number Exercises mistake; Let the relationship matrix between exercises and knowledge points be denoted as... , among which, if Then it means the first Exercises With the One knowledge point Related; if Then it means the first Exercises With the One knowledge point Irrelevant; The first students The set of questions that have been answered is denoted as ; Based on students' correct answer rate, those whose correct answer rate exceeds a threshold will be... Students were divided into groups based on their accuracy in answering questions. The correct answer rate should not exceed the threshold. Students were divided into groups with lower accuracy rates on the questions. ; Will The corresponding answer records with the answer tag "1" and The typical pattern consists of answer records with an answer tag of "0" in the corresponding answer records. ; Will The corresponding answer records with an answer tag of "0" and The answer records with the answer tag "1" in the corresponding answer records constitute an atypical pattern. ; Step 2: Define the current iteration number as... and initialize ; Step 3, from typical patterns The middle generates the first through selective data augmentation Deviation data of round iteration and with The data are input into the cognitive bias diagnosis model for processing, and the first result is output. The deviation prediction results of the round of iterations are used to construct the deviation cognitive diagnosis model. Symmetric cross-entropy loss in round iterations ; Step 4: Based on the output of the deviation cognitive diagnosis model... The prediction results from the first iteration are adaptively used to generate the second iteration. Unbiased data in round iteration and with The samples are input together into the biased cognitive diagnostic model for processing, and the first result is output. The biased prediction results from the rounds of iterations are used to construct the biased cognitive diagnostic model. Weighted cross-entropy loss in rounds of iteration ; Step 5: Construct the first equation using equation (9). The overall objective function of the round iteration : (9) Step 6: Synchronous training of the cognitive diagnostic model: Step 6.1: Use the Adam optimizer to perform the first step on the biased cognitive diagnostic model and the debiased cognitive diagnostic model. Train in rounds and minimize the first round. Loss function of round iteration To update the model parameter set and obtain the first... The biased cognitive diagnostic model after the first round of training and the first round of training A biased cognitive diagnostic model after rounds of training; Step 6.2, Assign to Then, return to step 6.1 and execute sequentially until... The process continues until convergence, thus obtaining a well-established cognitive diagnostic model that corrects biases. Step 6.3: In the model inference stage, the established bias-free cognitive diagnostic model is used to predict students' answering abilities and obtain the optimal answer prediction label corresponding to the students' answers to the exercises, so as to achieve cognitive diagnostic bias-free.

2. The method for bias-free cognitive diagnosis based on adaptive data augmentation according to claim 1, characterized in that, Step 3 is performed as follows: Step 3.1, from typical patterns Randomly select the first Answer records of round iterations , will the The first iteration students One-hot encoding and the first The first iteration One exercise The one-hot encodings are respectively input into the deviation cognitive diagnosis model, and the corresponding outputs of the embedding layers in the deviation cognitive diagnosis model are then generated. The first iteration students The representation and the The first iteration One exercise The representation ;in, Indicates the first The first iteration students Answer number One exercise The corresponding answer tags; Step 3.2, from the typical pattern Randomly select the answer records of the t-th iteration. And satisfy , will the The first iteration students One-hot encoding and the first The first iteration One exercise The one-hot encodings are respectively input into the deviation cognitive diagnosis model, and the corresponding outputs of the embedding layers in the deviation cognitive diagnosis model are then generated. The first iteration students The representation and the The first iteration One exercise The representation ;in, Indicates the first The first iteration students Answer number One exercise The corresponding answer tags; Step 3.3: Use equation (1) to generate the first... Bias triplet of round iteration Thus, the first Deviation data of round iteration : (1) In equation (1), To show obedience The parameters of the distribution Indicates the first Deviation in the round of iterations The representation, Indicates the first Exercises on deviations in round iteration The representation, Indicating students with deviations For deviation exercises Answer records; Step 3.4, and The data are input into the deviation cognitive diagnosis model for processing, and the corresponding output is the first... The first iteration students Answer number Exercises Corresponding deviation answer prediction tags ; Step 3.5: Construct a cognitive deviation diagnosis model using equation (2) in the first step. Symmetric cross-entropy loss in round iterations : (2) In equation (2), Indicates the first Cross-entropy loss in round iterations, Indicates the first The inverse cross-entropy loss of the round of iterations, and we have: (3) (4)。 3. The bias-free cognitive diagnosis method based on adaptive data augmentation according to claim 2, characterized in that, Step 4 is performed as follows: Step 4.1: Calculate the answer record using formula (5). In the Exponential mean of cross-entropy in rounds of iteration : (5) In equation (5), It is momentum; Indicates the first The exponential mean of the cross-entropy in each round of iterations; Step 4.2, use equation (6) to calculate the exponential mean. Normalization and truncation are used to obtain the answer record. In the Importance score of round iteration : (6) In equation (4), This represents the hyperparameter that controls the cutoff threshold; Step 4.3, based on the first Importance score set of rounds of iteration and answer records From multiple distributions The middle sampling obtained the first Unbiased answer record of round iteration ; Step 4.4, from the first Unbiased answer record of round iteration One answer record is randomly selected from the results. , will the The first iteration students One-hot encoding and the first The first iteration Exercises The one-hot encodings are respectively input into the debiased cognitive diagnostic model, and the corresponding embedding layers in the debiased cognitive diagnostic model output the first... The first iteration students The representation and the The first iteration One exercise The representation ;in, Indicates the first The first iteration students Answer number One exercise The corresponding answer tags; Step 4.5, from the first Unbiased answer record of round iteration One answer record is randomly selected from the results. And satisfy , will the The first iteration students One-hot encoding and the first The first iteration One exercise The one-hot encodings are respectively input into the debiased cognitive diagnostic model, and the corresponding embedding layers in the debiased cognitive diagnostic model output the first... The first iteration students The representation and the The first iteration One exercise The representation ;in, Indicates the first The first iteration students answer One exercise The corresponding answer tags; Step 4.6: Use equation (7) to generate the first... Unbiased triples in round-recursion Thus, the first Unbiased data in round iteration : (7) In equation (7), To show obedience The parameters of the distribution Indicates the first Unbiased students in rounds of iteration The representation, Indicates the first Unbiased exercises in round iteration The representation, Indicates an unbiased student For unbiased exercises Answer records; Step 4.7, and The inputs are respectively fed into the biased cognitive diagnostic model for processing, and the corresponding outputs are... The first iteration students Answer number Exercises The corresponding biased answer prediction tags ; Step 4.8: Construct a biased cognitive diagnostic model using equation (8) in the first step. Weighted cross-entropy loss in rounds of iteration : (8)。 4. An electronic device, comprising a memory and a processor, characterized in that, The memory is used to store a program that supports the processor in executing the debiased cognitive diagnostic method according to any one of claims 1-3, and the processor is configured to execute the program stored in the memory.

5. A computer-readable storage medium storing a computer program thereon, characterized in that, The computer program, when run by a processor, performs the steps of the debiased cognitive diagnostic method according to any one of claims 1-3.

Images