Depolarization cognitive diagnosis method based on deviation amplification and deviation elimination

Abstract

The invention discloses a depolarization cognitive diagnosis method based on deviation amplification and deviation elimination. The method comprises the following steps: 1, performing log division on student-exercise interaction logs; 2, constructing a cognitive diagnosis model; 3, designing a deviation amplification and deviation elimination data enhancement strategy to obtain an enhanced student-exercise interaction log; and 4, constructing a weighted optimization target, jointly training the deviation model and the depolarization model, and carrying out student modeling and cognitive diagnosis by using the depolarization model. According to the method, the accuracy and unbiasedness of the cognitive diagnosis model in processing student-exercise interaction logs can be remarkably improved, and the method is suitable for educational evaluation and personalized learning recommendation systems.

Description

Technical Field

[0001] The present invention relates to the field of online education, and in particular to a debiasing cognitive diagnosis method based on bias amplification and bias elimination, which aims to identify students' knowledge mastery level by analyzing logs of students' interactions with exercises on an online education platform and realize personalized student modeling. Background Art

[0002] With the rapid development of online education, how to extract valuable information from the interaction logs between students and exercises to evaluate their mastery of knowledge points and build personalized student models has become a hot topic in current research. Cognitive Diagnosis (CD), as a core technology in this field, is widely used in the recommendation of personalized learning paths. However, although many methods have been proposed to make full use of interaction log data to improve modeling effects, existing CD models often rely too much on the accuracy of students' historical answers. This dependence may cause the model to be unable to accurately reflect the students' mastery of knowledge concepts and ignore the progress or regression of students in the learning process.

[0003] For example, when students have a high accuracy rate in history, the model may overestimate their actual ability and ignore their weaknesses in specific concepts; on the contrary, for students with a low accuracy rate, the model may underestimate their ability and fail to capture their progress in certain knowledge points. This limitation not only affects the accurate diagnosis of students' abilities, but may also hinder the formulation of subsequent personalized teaching strategies and lead to biased evaluations of students' performance. Summary of the invention

[0004] In order to solve the above problems, the present invention proposes a debiased cognitive diagnosis method based on bias amplification and bias elimination, which aims to effectively reduce the phenomenon of overestimation of the abilities of high-accuracy students and underestimation of the abilities of low-accuracy students, and to achieve accurate and unbiased cognitive diagnosis, thereby more comprehensively capturing the students' real learning situation and providing more precise personalized educational support.

[0005] In order to achieve the above-mentioned purpose, the present invention adopts the following technical scheme:

[0006] The debiasing cognitive diagnosis method based on bias amplification and bias elimination of the present invention is characterized in that it is carried out according to the following steps:

[0007] Step 1: Construct the student-exercise interaction log and divide the log:

[0008] Step 1.1: Build a student-exercise interaction log:

[0009] Define the student set as , the exercise set is , the knowledge point set is ,in, represents the nth student, represents the mth exercise, represents the kth knowledge point, , , ; N represents the number of students, M represents the number of exercises, and K represents the number of knowledge points;

[0010] Collect student-exercise interaction logs Any interaction log in ,in, Indicates the nth student Answer the mth exercise The real label, Indicates the nth student Answer the mth exercise correctly ;like , then it means the nth student Wrong answer to question m ; The total number of interaction logs is recorded as ;

[0011] Let the exercise-knowledge point matrix be ,in, Represents the mth exercise Contains the kth knowledge point ; Represents the mth exercise Does not include the kth knowledge point ;

[0012] Step 1.2, log division:

[0013] The nth student The set of logs for answering all exercises is recorded as ,according to Calculate the nth student The correct answer rate ;

[0014] Based on the correct answer rate, the correct answer rate exceeds the threshold The students are divided into high accuracy student sets , the correct answer rate is lower than the threshold The students are divided into the low accuracy student set , the correct answer rate is between and The students between them are divided into a medium accuracy student set ;

[0015] Assemble the students with high accuracy The logs with the true label of 1 and the low-accuracy student set in all interaction logs The logs with the true label of 0 in all interaction logs constitute the deviation consistent log set ;

[0016] Assemble the students with high accuracy The logs with the true label of 0 and the low-accuracy student set in all interaction logs The logs with the true label of 1 in all interaction logs constitute the deviation conflict log set ;

[0017] The medium accuracy students are grouped together All interaction logs constitute a set of deviation independent logs ;

[0018] Step 2: Construct a cognitive diagnosis model , including: Student feature extractor , Exercise Feature Extractor , interaction function and diagnostic layer, and interact with the log Process and get the nth student Answer the mth question correctly Probability , used to construct the loss function ;

[0019] Step 3: Based on the data enhancement strategy of deviation amplification and deviation elimination, construct an enhanced deviation consistent log set and deviation conflict log collection :

[0020] Step 3.1: According to the cognitive diagnosis model Initialize two deviation models of the same structure and debiasing model ;

[0021] Step 3.2: From the deviation consistent log collection Randomly extract 2 logs and input them into the deviation model The deviation is processed in the student feature extractor and the exercise feature extractor to obtain the deviation consistent log ; Thus constructing an enhanced deviation consistent log set ;in, Indicates students, Indicates Exercises, Indicates students Answer Exercises The synthetic label of

[0022] Step 3.3: According to a certain probability, set the deviation consistent log and from the deviation conflict log collection Extract a log from each and input it into the debiasing model The deviation conflict log is obtained by processing in the student feature extractor and the exercise feature extractor ; Thus constructing an enhanced deviation conflict log set ;in, Indicates students, Indicates Exercises, Indicates students Answer Exercises The synthetic label of

[0023] Step 4: Construct the total loss function , and use AdamW optimizer to optimize the total loss function Minimize the deviation model and debiasing model All parameters in are optimized and the deviation model after joint training is obtained and debiasing model , and use the jointly trained debiased model Achieve cognitive diagnosis of debiased students.

[0024] The debiasing cognitive diagnosis method based on bias amplification and bias elimination described in the present invention is also characterized in that step 2 is performed according to the following steps:

[0025] Step 2.1: The student feature extractor Extract the nth student Ability to master various knowledge points ;

[0026] Step 2.2: The exercise feature extractor Extract the mth exercise The difficulty information and discrimination information of The feature representation ;

[0027] Step 2.3: The interaction function Will , as well as After fusion, an interactive fusion feature representation is generated;

[0028] Step 2.4: The diagnostic layer Depend on The nth student is obtained by processing the fusion feature representation. Answer the mth question correctly The predicted label ;

[0029] Step 2.5: Use formula (1) to build a cognitive diagnosis model The loss function ;

[0030] (1)

[0031] In formula (1), represents cross entropy.

[0032] Further, the step 3.2 is performed as follows:

[0033] Step 3.2.1: From the deviation consistent log set Randomly select students Answer Exercises Log and students Answer Exercises Log ,and ,in, Indicates students Answer Exercises The real label, Indicates students Answer Exercises The true label of

[0034] Will Input to the bias model The student feature extractor and exercise feature extractor are processed and the corresponding students Ability to master and Exercises The feature representation ;

[0035] Will Input to the bias model The student feature extractor and exercise feature extractor are processed and the corresponding students Ability to master and Exercises The feature representation ;

[0036] Step 3.2.2: Use formula (2) to get students Ability to master , No. Exercises The feature representation , No. students Answer Exercises Synthetic tags , and form a consistent log of deviations ;

[0037] (2)

[0038] Mode middle, Represents a subject parameter is a uniformly distributed random variable.

[0039] Further, the step 3.3 is performed as follows:

[0040] Step 3.3.1: Collect the deviation consistent logs Input Bias Model Diagnose and obtain The probability set of correct diagnosis in the log ; Collect the deviation conflict logs Input Bias Model Diagnose and obtain The probability set of correct diagnosis in the log ;

[0041] Step 3.3.2, follow From the deviation consistent log collection Extract the students Answer Exercises Log , and input into the debiasing model The student feature extractor and exercise feature extractor are processed and the corresponding students Ability to master and Exercises The feature representation ;

[0042] according to From the deviation conflict log collection Extract from students Answer Exercises Log , and input into the debiasing model The student feature extractor and exercise feature extractor are processed and the corresponding students Ability to master and Exercises The feature representation ;

[0043] Step 3.3.3, use formula Get the first students Ability to master , No. Exercises The feature representation , No. students Answer Exercises Synthetic tags , and form a deviation conflict log ;

[0044] (3)

[0045] Mode middle, Represents a subject parameter is a uniformly distributed random variable.

[0046] Furthermore, the total loss function in step 4 is It is constructed in the following steps;

[0047] Step 4.1: Use formula (4) to construct the deviation model The loss function ;

[0048] (4)

[0049] Mode middle, The deviation model The cross entropy loss function is: express The number of logs in is the bias parameter that controls the blending;

[0050] Step 4.2: Use formula (5) to build a debiasing model The loss function ;

[0051] (5)

[0052] In formula (5), For the debiasing model The cross entropy loss function is: express The number of logs in is the debiasing parameter that controls the blend; is a set of weights, and , Indicates interaction log The weights are:

[0053] (6)

[0054] In formula (6), Representation bias model Output the nth student Answer the mth question correctly The predicted label of

[0055] Represents the debiased model Output the nth student Answer the mth question correctly The predicted label of

[0056] Step 4.3: Use formula (7) to construct the total loss function :

[0057] (7).

[0058] The electronic device of the present invention includes a memory and a processor, wherein the memory is used to store a program that supports the processor to execute the debiased cognitive diagnosis method, and the processor is configured to execute the program stored in the memory.

[0059] The present invention provides a computer-readable storage medium, wherein a computer program is stored on the computer-readable storage medium, and the computer program executes the steps of the debiasing cognitive diagnosis method when the computer program is executed by a processor.

[0060] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0061] 1. The present invention innovatively introduces bias amplification and bias elimination data enhancement strategies to assist the bias model to fully model the bias and the debiasing model to eliminate the bias. The bias amplification strategy uses a mixing operation to synthesize enhanced bias-consistent logs from the feature level. The bias elimination strategy uses the output of the biased model to probabilistically select bias alignment logs and bias conflict logs for mixing to enrich the diversity of bias conflict logs, thereby helping the cognitive diagnosis model to recognize those key logs that are inconsistent with the student's historical performance, thereby improving the accuracy and effectiveness of the cognitive diagnosis model in identifying the student's mastery of knowledge points.

[0062] 2. The present invention optimizes the model training process by jointly training the biased model and the debiased model and adopting a novel weighted optimization objective. This joint training method not only retains the deviation information in the deviation alignment log, but also reduces the negative impact of these deviations on the performance of the debiased model, thereby improving the diagnostic accuracy and unbiasedness of the students' mastery of knowledge points.

[0063] 3. The model independence of the present invention is one of its significant advantages, which means that the method can be easily integrated into various existing cognitive diagnosis models without modifying the model architecture. This model independence greatly expands the applicability of the present invention, enabling it to be widely used in different educational scenarios and research contexts. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 is an overall flow chart of the method of the present invention;

[0065] Figure 2 The proportion of each type of log and the accuracy of the original training method of the Multidimensional Item Response Theory (MIRT) cognitive diagnosis model for different log diagnoses;

[0066] Figure 3 It is the overall method diagram of the present invention;

[0067] Figure 4 This is a sample analysis diagram of the cognitive diagnostic model's overestimation of the abilities of students with high accuracy rates. DETAILED DESCRIPTION

[0068] In this embodiment, Figure 1 As shown, a debiased cognitive diagnostic method based on bias amplification and bias elimination is carried out in the following steps:

[0069] Step 1: Construct the student-exercise interaction log and divide the log:

[0070] Step 1.1: Build a student-exercise interaction log:

[0071] Define the student set as , the exercise set is , the knowledge point set is ,in, represents the nth student, represents the mth exercise, represents the kth knowledge point, , , ; N represents the number of students, M represents the number of exercises, and K represents the number of knowledge points;

[0072] Collect student-exercise interaction logs Any interaction log in ,in, Indicates the nth student Answer the mth exercise The real label, Indicates the nth student Answer the mth question correctly ;like , then it means the nth student Wrong answer to question m ; The total number of interaction logs is recorded as ;

[0073] Let the exercise-knowledge point matrix be ,in, Represents the mth exercise Contains the kth knowledge point ; Represents the mth exercise Does not include the kth knowledge point .

[0074] In this embodiment, the SLP-Math and SLP-English datasets are used. The SLP-Math dataset contains 4668 students, 222 exercises, 52 knowledge points, and 565,031 logs; the SLP-English dataset contains 4669 students, 339 exercises, 29 knowledge points, and 952,089 logs.

[0075] Step 1.2, log division:

[0076] The nth student The set of logs for answering all exercises is recorded as ,according to Calculate the nth student The correct answer rate ;

[0077] Based on the correct answer rate, the correct answer rate exceeds the threshold The students are divided into high accuracy student sets , the correct answer rate is lower than the threshold The students are divided into the low accuracy student set , the correct answer rate is between and The students between them are divided into a medium accuracy student set ;

[0078] Assemble the students with high accuracy The logs with the true label of 1 and the low-accuracy student set in all interaction logs The logs with the true label of 0 in all interaction logs constitute the deviation consistent log set ;

[0079] Assemble the students with high accuracy The logs with the true label of 0 and the low-accuracy student set in all interaction logs The logs with the true label of 1 in all interaction logs constitute the deviation conflict log set ;

[0080] The medium accuracy students are grouped together All interaction logs constitute a set of deviation independent logs .

[0081] In this embodiment, and are set to 0.8 and 0.2 respectively; Figure 2 As shown, Figure 2 Part (a) shows , and The proportion in the SLP-Math and SLP-English datasets; Figure 2 Part (b) shows the different performances when trained using the Multidimensional Item Response Theory (MIRT) cognitive diagnostic model.

[0082] Step 2: Construct a cognitive diagnosis model , including: Student feature extractor , Exercise Feature Extractor , interaction function and diagnostic layer, and interact with the log Process and get the nth student Answer the mth question correctly Probability , used to construct the loss function :

[0083] Step 2.1: Student feature extractor Extract the nth student Ability to master various knowledge points ;

[0084] Step 2.2: Exercise feature extractor Extract the mth exercise The difficulty information and discrimination information of The feature representation .

[0085] Step 2.3: Interaction function Will , as well as After fusion, an interactive fusion feature representation is generated;

[0086] Step 2.4, Diagnostic Layer Depend on The nth student Answer the mth exercise correctly The predicted label ;

[0087] Step 2.5: Use formula (1) to build a cognitive diagnosis model The loss function ;

[0088] (1)

[0089] In formula (1), represents cross entropy.

[0090] In this embodiment, since the present invention is independent of the model, the model Any existing cognitive diagnosis model can be used. Taking the neurocognitive diagnosis NCDM model as an example, the student feature extractor The generated student features are realized through embedding The shape is , represents the students’ mastery of each knowledge point; Exercise feature extractor Generate the features of the exercises through embedding Depend on and Composition, of which Represents the difficulty information of the exercise, and its shape is ; Represents the discrimination information of the exercise, which is a scalar value; the interaction function for , the features of students and exercises are combined with the knowledge point matrix through element-by-element multiplication and matrix multiplication to generate the interactive feature representation, which is input to the diagnosis layer ; Diagnostic layer It consists of 3 layers of fully connected MLPs and is responsible for making the final prediction of students' answering performance.

[0091] Step 3: Based on the data enhancement strategy of deviation amplification and deviation elimination, construct an enhanced deviation consistent log set and deviation conflict log collection :

[0092] Step 3.1: According to the cognitive diagnosis model Initialize two deviation models of the same structure and debiasing model ;

[0093] Step 3.2: From the deviation consistent log collection Randomly extract 2 logs and input them into the deviation model The deviation is processed in the student feature extractor and the exercise feature extractor to obtain the deviation consistent log ; Thus constructing an enhanced deviation consistent log set ;

[0094] Step 3.2.1: From the deviation consistent log set Randomly select students Answer Exercises Log and students Answer Exercises Log ,and ,in, Indicates students Answer Exercises The real label, Indicates students Answer Exercises The true label of

[0095] Will Input to the bias model The student feature extractor and exercise feature extractor are processed and the corresponding students Ability to master and Exercises The feature representation ;

[0096] Will Input to the bias model The student feature extractor and exercise feature extractor are processed and the corresponding students Ability to master and Exercises The feature representation .

[0097] Step 3.2.2: Use formula (2) to get students Ability to master , No. Exercises The feature representation , No. students Answer Exercises Synthetic tags , and form a consistent log of deviations ;

[0098] (2)

[0099] In formula (2), Represents a subject parameter is a uniformly distributed random variable.

[0100] Step 3.3: According to a certain probability, set the deviation consistent log and from the deviation conflict log collection Extract a log from each and input it into the debiasing model The deviation conflict log is obtained by processing in the student feature extractor and the exercise feature extractor ; Thus constructing an enhanced deviation conflict log set ;

[0101] Step 3.3.1: Collect the deviation consistent logs Input Bias Model Diagnose and obtain The probability set of correct diagnosis in the log ; Collect the deviation conflict logs Input Bias Model Diagnose and obtain The probability set of correct diagnosis in the log ;

[0102] Step 3.3.2, follow From the deviation consistent log collection Extract the students Answer Exercises Log , and input into the debiasing model The student feature extractor and exercise feature extractor are processed and the corresponding students Ability to master and Exercises The feature representation ;

[0103] according to From the deviation conflict log collection Extract from students Answer Exercises Log , and input into the debiasing model The student feature extractor and exercise feature extractor are processed and the corresponding students Ability to master and Exercises The feature representation .

[0104] Step 3.3.3: Use formula (3) to get students Ability to master , No. Exercises The feature representation , No. students Answer Exercises Synthetic tags , and form a deviation conflict log ;

[0105] (3)

[0106] In formula (3), Represents a subject parameter is a uniformly distributed random variable.

[0107] Step 4: Figure 3As shown, construct the total loss function , and use AdamW optimizer to optimize the total loss function Minimize the deviation model and debiasing model All parameters in are optimized and the deviation model after joint training is obtained and debiasing model , and use the jointly trained debiased model Achieve cognitive diagnosis of debiased students:

[0108] Step 4.1: Use formula (4) to construct the deviation model The loss function ;

[0109] (4)

[0110] In formula (4), The deviation model The cross entropy loss function is: express The number of logs in is the deviation parameter controlling the mixture; in this embodiment, the deviation model During training, the enhanced deviation consistent log is used for training. Set to 1.0.

[0111] Step 4.2: Use formula (5) to build a debiasing model The loss function ;

[0112] (5)

[0113] In formula (5), For the debiasing model The cross entropy loss function is: express The number of logs in is the debiasing parameter that controls the blend; is a set of weights, and , Indicates interaction log The weights are:

[0114] (6)

[0115] In formula (6), Representation bias model Output the nth student Answer the mth question correctly The predicted label of

[0116] Represents the debiased model Output the nth student Answer the mth exercise correctly The predicted label of During training, not only the original student-exercise interaction log Weighted, and combined with the enhanced deviation conflict log Conduct training, Set to 0.5.

[0117] Step 4.3: Use formula (7) to construct the total loss function :

[0118] (7).

[0119] In this embodiment, an electronic device includes a memory and a processor, wherein the memory is used to store a program that supports the processor to execute the above method, and the processor is configured to execute the program stored in the memory.

[0120] In this embodiment, a computer-readable storage medium stores a computer program on the computer-readable storage medium, and the computer program executes the steps of the above method when executed by a processor.

[0121] The test results of the present invention are further described in conjunction with the following charts:

[0122] In order to fully verify the unbiasedness and accuracy of the present invention in the field of cognitive diagnosis, the present invention conducted extensive out-of-distribution generalization (OOD) experiments on two educational datasets, SLP-Math and SLP-English. The experiments covered a variety of cognitive diagnosis models, including item response theory (IRT), multidimensional item response theory (MIRT), neurocognitive diagnosis (NCDM), and knowledge-aware cognitive diagnosis (KSCD) models.

[0123] In the experimental design, the present invention pays special attention to the composition of the test set, ensuring a balanced ratio between bias consistency, bias conflict, and bias independent logs to simulate the diversity in the real world. In addition, the effect of the present invention is compared with the native training method (Vanilla) and two commonly used debiasing baseline methods - learning from failure (LfF) and gradient alignment (GA).

[0124] Tables 1 and 2 present in detail the performance of each cognitive diagnosis model in the out-of-distribution generalization (OOD) scenario of the SLP-Math and SLP-English datasets. In these evaluations, the present invention uses the average diagnostic accuracy on the test set as the measurement standard to objectively reflect the diagnostic effectiveness of each model.

[0125] Table 1 OOD experiments on the SLP-Math dataset

[0126]

[0127] Table 2 OOD experiments on the SLP-English dataset

[0128]

[0129] The experimental results show that the present invention has excellent performance in terms of accuracy and unbiasedness, and its debiasing performance is significantly better than other benchmark methods, and the advantages of the present invention are particularly obvious. These data strongly confirm the effectiveness of the present invention and highlight its application potential in the field of cognitive diagnosis.

[0130] In order to further explore the accuracy and unbiasedness of the present invention, the present invention conducted a case analysis. In this analysis, the present invention selected a student with a high overall accuracy rate in the SLP-Math dataset as the research object. Figure 4 As shown, Figure 4 Part (a) clearly shows the student’s average correctness on each knowledge point. Although the student’s overall performance is satisfactory, in some specific knowledge points (knowledge points and ) performed relatively weakly. Figure 4 Part (b) shows a radar chart comparing the original KSCD model training method and the method proposed in this invention in cognitive diagnosis. The original KSCD model training method tends to overestimate students' weak knowledge points (knowledge points and ). In contrast, the present invention can more accurately identify the actual ability differences of students on these knowledge points, effectively avoiding overestimation of students' abilities. This finding further verifies the advantages of the present invention in improving the accuracy of cognitive diagnosis and demonstrates its potential application value in educational assessment.

Claims

1. A debiased cognitive diagnostic method based on bias amplification and bias elimination, characterized in that: The steps are as follows: Step 1: Construct the student-exercise interaction log and divide the log: Step 1.1: Build a student-exercise interaction log: Define the student set as , the exercise set is , the knowledge point set is ,in, represents the nth student, represents the mth exercise, represents the kth knowledge point, , , ; N represents the number of students, M represents the number of exercises, and K represents the number of knowledge points; Collect student-exercise interaction logs Any interaction log in ,in, Indicates the nth student Answer the mth exercise The real label, Indicates the nth student Answer the mth exercise correctly ;like , then it means the nth student Wrong answer to question m ; The total number of interaction logs is recorded as ; Let the exercise-knowledge point matrix be ,in, Represents the mth exercise Contains the kth knowledge point ; Represents the mth exercise Does not include the kth knowledge point ; Step 1.2, log division: The nth student The set of logs for answering all exercises is recorded as ,according to Calculate the nth student The correct answer rate ; Based on the correct answer rate, the correct answer rate exceeds the threshold The students are divided into high accuracy student sets , the correct answer rate is lower than the threshold The students are divided into the low accuracy student set , the correct answer rate is between and The students between them are divided into a medium accuracy student set ; Assemble the students with high accuracy The logs with the true label of 1 and the low-accuracy student set in all interaction logs The logs with the true label of 0 in all interaction logs constitute the deviation consistent log set ; Assemble the students with high accuracy The logs with the true label of 0 and the low-accuracy student set in all interaction logs The logs with the true label of 1 in all interaction logs constitute the deviation conflict log set ; The medium accuracy students are grouped together All interaction logs constitute a set of deviation independent logs ; Step 2: Construct a cognitive diagnosis model , including: Student feature extractor , Exercise Feature Extractor , interaction function and diagnostic layer, and interact with the log Process and get the nth student Answer the mth exercise correctly Probability , used to construct the loss function ; Step 3: Based on the data enhancement strategy of deviation amplification and deviation elimination, construct an enhanced deviation consistent log set and deviation conflict log collection : Step 3.1: According to the cognitive diagnosis model Initialize two deviation models of the same structure and debiasing model ; Step 3.2: From the deviation consistent log collection Randomly extract 2 logs and input them into the deviation model The deviation is processed in the student feature extractor and the exercise feature extractor to obtain the deviation consistent log ; Thus constructing an enhanced deviation consistent log set ;in, Indicates students, Indicates Exercises, Indicates students Answer Exercises The synthetic label of Step 3.3: According to a certain probability, set the deviation consistent log and from the deviation conflict log collection Extract a log from each and input it into the debiasing model The deviation conflict log is obtained by processing in the student feature extractor and the exercise feature extractor ; Thus constructing an enhanced deviation conflict log set ;in, Indicates students, Indicates Exercises, Indicates students Answer Exercises The synthetic label of Step 4: Construct the total loss function , and use AdamW optimizer to optimize the total loss function Minimize the deviation model and debiasing model All parameters in are optimized and the deviation model after joint training is obtained and debiasing model , and use the jointly trained debiased model Achieve cognitive diagnosis of debiased students.

2. A debiased cognitive diagnostic method based on bias amplification and bias elimination according to claim 1, characterized in that: Described step 2 is carried out as follows: Step 2.1: The student feature extractor Extract the nth student Ability to master various knowledge points ; Step 2.2: The exercise feature extractor Extract the mth exercise The difficulty information and discrimination information of The feature representation ; Step 2.3: The interaction function Will , as well as After fusion, an interactive fusion feature representation is generated; Step 2.4: The diagnostic layer Depend on The nth student Answer the mth question correctly The predicted label ; Step 2.5: Use formula (1) to build a cognitive diagnosis model The loss function ; (1) In formula (1), represents cross entropy.

3. A debiased cognitive diagnostic method based on bias amplification and bias elimination according to claim 2, characterized in that: The step 3.2 is carried out as follows: Step 3.2.1: From the deviation consistent log set Randomly select students Answer Exercises Log and students Answer Exercises Log ,and ,in, Indicates students Answer Exercises The real label, Indicates students Answer Exercises The true label of Will Input to the bias model The student feature extractor and exercise feature extractor are processed and the corresponding students Ability to master and Exercises The feature representation ; Will Input to the bias model The student feature extractor and exercise feature extractor are processed and the corresponding students Ability to master and Exercises The feature representation ; Step 3.2.2: Use formula (2) to get students Ability to master , No. Exercises The feature representation , No. students Answer Exercises Synthetic tags , and form a consistent log of deviations ; (2) Mode middle, Represents a subject parameter is a uniformly distributed random variable.

4. A debiased cognitive diagnostic method based on bias amplification and bias elimination according to claim 3, characterized in that: The step 3.3 is carried out as follows: Step 3.3.1: Collect the deviation consistent logs Input Bias Model Diagnose and obtain The probability set of correct diagnosis in the log ; Collect the deviation conflict log Input Bias Model Diagnose and obtain The probability set of correct diagnosis in the log ; Step 3.3.2, follow From the deviation consistent log collection Extract the students Answer Exercises Log , and input into the debiasing model The student feature extractor and exercise feature extractor are processed and the corresponding students Ability to master and Exercises The feature representation ; according to From the deviation conflict log collection Extract from students Answer Exercises Log , and input into the debiasing model The student feature extractor and exercise feature extractor are processed and the corresponding students Ability to master and Exercises The feature representation ; Step 3.3.3, use formula Get the first students Ability to master , No. Exercises The feature representation , No. students Answer Exercises Synthetic tags , and form a deviation conflict log ; (3) Mode middle, Represents a subject parameter is a uniformly distributed random variable.

5. The debiasing cognitive diagnosis method based on bias amplification and bias elimination according to claim 4, characterized in that: The total loss function in step 4 It is constructed in the following steps; Step 4.1: Use formula (4) to construct the deviation model The loss function ; (4) Mode middle, The deviation model The cross entropy loss function is: express The number of logs in is the bias parameter that controls the blending; Step 4.2: Use formula (5) to build a debiasing model The loss function ; (5) In formula (5), For the debiasing model The cross entropy loss function is: express The number of logs in is the debiasing parameter that controls the blend; is a set of weights, and , Indicates interaction log The weights are: (6) In formula (6), Representation bias model Output the nth student Answer the mth question correctly The predicted label of Represents the debiased model Output the nth student Answer the mth question correctly The predicted label of Step 4.3: Use formula (7) to construct the total loss function : (7)。 6. An electronic device, comprising a memory and a processor, characterized in that: The memory is used to store a program that supports the processor to execute the debiased cognitive diagnosis method according to any one of claims 1 to 5, and the processor is configured to execute the program stored in the memory.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the debiased cognitive diagnosis method according to any one of claims 1 to 5 are executed.

Images