Cognitive diagnosis method suitable for different knowledge point numbers under multiple learning tasks

Abstract

The invention discloses a cognitive diagnosis method suitable for different knowledge point numbers under multiple learning tasks, and the method comprises the steps: 1) obtaining a knowledge mastering level representation based on a complete cognitive state of a learner and a test Q matrix; 2) training the parameter matrix to obtain a variable-length knowledge weight vector; 3) based on the obtained knowledge mastering level representation, the variable-length knowledge weight vector and the number of the knowledge points, obtaining a position index in the knowledge mastering level representation and aligning the position index with the knowledge point subset; and 4) based on the knowledge mastering level position index matrix, converting the knowledge mastering level representation matrix into a knowledge mastering level function matrix, and aggregating the knowledge mastering level function matrix with the variable-length knowledge weight vector to obtain an ideal answer reaction of the learner. The variable-length knowledge weight vector representation module is introduced, the limitation of a traditional cognitive diagnosis model is overcome, the method can be suitable for cognitive diagnosis of different scenes of multiple learning task knowledge quantities, and the practicability is higher.

Description

Technical Field

[0001] The present invention belongs to the technical field of cognitive diagnosis under multiple learning tasks, and specifically relates to a cognitive diagnosis method that is adaptable to different numbers of knowledge points under multiple learning tasks. Background Art

[0002] With the continuous advancement of computer technology, deep learning technology has shown excellent performance in many fields, especially in mining nonlinear interactive relationships, processing sparse data, and predicting learners' performance. Therefore, some scholars realize that combining cognitive diagnosis with neural networks is the future direction for mining the value of educational data. In 2023, Zhang Suojuan et al. proposed a cognitive diagnosis deep model (CI-CDM) that integrates knowledge and the interactive relationship between knowledge sets into the process of cognitive diagnosis, and realizes the unified learning of learners' cognitive states and knowledge weights for learning tasks. CI-CDM first constructs a cognitive diagnosis model framework, adopts a knowledge weight learning algorithm based on multiple learning tasks, and uses it to characterize knowledge interaction relationships. Then, a deep neural network based on fuzzy measures (FM-Net) is used to predict learners' answer performance and realize the construction of a cognitive diagnosis deep model that integrates knowledge interaction.

[0003] Although CI-CDM takes into account the different knowledge weights and the interactive relationships between knowledge sets brought about by the knowledge interaction relationships for learning tasks in practical applications. However, CI-CDM adopts the technical means of screening fixed-length cognitive state vectors, fixed-length knowledge weight representation vectors, and unchanged index positions of knowledge point subsets in all learning tasks. In scenarios where learning tasks contain different numbers of knowledge points, the model needs to fully consider the interactive relationship of the knowledge set, resulting in inconsistent lengths of the knowledge weight vector and the knowledge mastery level function in the learning task, while FM-Net and the knowledge mastery level function in CI-CDM cannot adapt to the variable-length knowledge set in the learning task. Therefore, CI-CDM can only be applied to scenarios where all learning tasks contain the same number of knowledge points, and cannot be applied to actual scenarios where the number of knowledge points in learning tasks is usually different. Summary of the invention

[0004] The purpose of the present invention is to address the deficiencies of the prior art and provide a cognitive diagnosis method that is adaptable to different numbers of knowledge points under multiple learning tasks. By modifying and introducing a variable-length knowledge weight vector representation module, the cognitive diagnosis deep model can obtain associated cognitive state representations and variable-length knowledge weight representation vectors (the number of valid elements in the vector is directly determined by the number of knowledge points in each learning task) according to the characteristics of different learning tasks, and adaptively generate index positions for each knowledge point subset, thereby overcoming the limitations of CI-CDM and being able to adapt to cognitive diagnosis in scenarios with different amounts of knowledge in multiple learning tasks.

[0005] In order to achieve the above object, the present invention adopts the following technical solution.

[0006] A cognitive diagnosis method adapted to different numbers of knowledge points under multiple learning tasks, the cognitive diagnosis method is implemented based on a cognitive diagnosis deep model, the cognitive diagnosis deep model includes an associated cognitive state representation module, a variable-length knowledge weight vector representation module, a knowledge point subset position index module and a knowledge mastery level position index module; the cognitive diagnosis method includes the following steps: Step S1: The associated cognitive state representation module is based on the learner's complete cognitive state and test Matrix, to obtain the learner's knowledge mastery level representation on each learning task; The test matrix ;learner , learning task ; Conceptual knowledge ;Learning tasks Knowledge collection ;in, Indicates the number of knowledge points; Indicates the number of learning tasks; represents the number of learners; Indicates the task Measuring knowledge ; Indicates the task Unmeasured Knowledge ; Step S2, the variable-length knowledge weight vector representation module obtains a variable-length knowledge weight vector that maintains monotonicity constraints between the learner and each learning task knowledge point subset and adapts to different numbers of knowledge points by training the deep neural network parameter matrix; Step S3, the knowledge point subset position index module obtains the position index of the knowledge point subset of each learning task in the knowledge mastery level representation based on the obtained learner's knowledge mastery level representation on each learning task, the variable-length knowledge weight vector adapted to different numbers of knowledge points, and the number of knowledge points, and aligns the position index with the knowledge point subset at the same position in the variable-length knowledge weight vector; Step S4, the knowledge mastery level position index module converts the learner's knowledge mastery level representation matrix on each learning task into a knowledge mastery level function matrix based on the knowledge mastery level position index matrix, and then aggregates it with the variable-length knowledge weight vector of the corresponding learning task to predict the learner's ideal answer response.

[0007] Specifically, step S1 is based on the learner's complete cognitive state and test Matrix, to obtain the learner's knowledge mastery level representation on each learning task, including the following steps: Step S11: The associated cognitive state representation module converts the knowledge mastery level vector into With Quiz matrix Each row is multiplied element by element to obtain the learner exist The corresponding knowledge mastery level representation matrix for each learning task : ; In formula (1), express J OK K The set of real matrices of columns; Step S12: for learners The knowledge mastery level representation matrix corresponding to each learning task Sort the elements in each row in descending order to get the initial knowledge mastery level and location index matrix; Step S13: Remove the initial knowledge mastery level The all-0 columns in the table get the initial knowledge mastery level after removing the all-0 columns and location index ; Step S14: Expand the initial knowledge mastery level after removing all zero columns Add a column of all zero vectors to the dimension and get the knowledge mastery level ,in, , Indicates the maximum amount of knowledge that each learning task can contain in the current data set, Representation Learning Task The number of pieces of knowledge examined.

[0008] Specifically, the step S2 of obtaining a variable-length knowledge weight vector that maintains monotonicity constraints between the learner and each learning task knowledge point subset and adapts to different numbers of knowledge points includes the following steps: Step S21: Establish the index code of the fuzzy measure in the knowledge weight vector, and unify the number of elements in the knowledge weight set to The power set of knowledge, ,in, By A vector composed of the knowledge weights of each learning task; Step S22: Analyze the test Matrix, initialize the deep neural network parameter matrix combined with fuzzy measure learning , The matrix initialization formula is expressed as: hour, ; otherwise, ; Formula (2)-Formula (3), is the knowledge set corresponding to the learning task A proper subset of Representing a knowledge set In learning tasks The weight on ; Step S23: Matrix parameter training to obtain knowledge weight representation for each learning task ,in, The learning task The weight of knowledge The vector composed of It is expressed as: ; In formula (4), Representing a knowledge set correspond 1 row of elements in a matrix; Step S24: Introduction , Indicates taking The element with the smallest weight in the set, Representation Learning Task Knowledge Collection A subset of The set of weights in the matrix, i.e. The matrix Column; in formula (4) The condition is equivalent to , The condition is equivalent to ,according to and The value of is used as a condition to calculate the learning task The weights of the true subsets of each knowledge set above; Representation of knowledge weights in various learning tasks Add a row of all 0 elements at the bottom to set the knowledge for each learning task Weight The value is assigned to 1, and the corresponding initial knowledge weight representation is recorded as ; The number of non-zero elements or valid elements in each column is proportional to the number of knowledge points included in the learning task. Change with change; knowledge weight representation The values ​​of the elements will be continuously updated with the iteration of the program, thereby obtaining a variable-length knowledge weight vector that maintains monotonicity constraints with the knowledge point subsets of each learning task and adapts to different numbers of knowledge points.

[0009] Furthermore, the specific steps of step S3 are as follows: Step S31, establishing a knowledge mastering level position index matrix; First, the test The elements of each row of the matrix are summed and 1 is subtracted from each sum to obtain a column vector , Represents the corresponding learning task The amount of knowledge contained in The maximum weight of ; Step S32: Initialize an all-0 matrix ,right The matrix is ​​assigned from A sequence of integers decreasing to 0, represented by: ; In formula (6), , Indicates the number of the learning task; Step S33: Test Sort the matrix in descending order, set When , the all-0 subcolumns are deleted, and The subcolumn following the column; otherwise, the subcolumn is not deleted, and the result is ; Step S34: Since each learning task examines different knowledge, the knowledge obtained in step S13 is and Multiply the matrices and extract the position index of the valid elements to get The matrix is ​​represented as: ; Indicates the knowledge mastery level index from high to low of the learners' knowledge mastery level in each learning task; Step S35: Sort the matrix in descending order to get the position index matrix, is a value between 0 and The integer index of ; Step S36: The positions of corresponding elements in the matrix are The position index changes, and the corresponding formula is as follows: ; Combined with step S35 and Both are decreasing position weights, and knowledge weights The knowledge points corresponding to the binary codes from high to low are the learning tasks Attribute Vector The knowledge points examined from right to left can realize the index and knowledge weight of the knowledge mastery level function The index is consistent; Step S37, take Cumulative sum by column, then subtract 1 to obtain the knowledge mastery level position index matrix , which realizes the unification of the position index of the knowledge mastery level representation matrix in the knowledge mastery level function, where the matrix It is expressed as: .

[0010] Specifically, the step S4 of predicting the learner's ideal response includes the following steps: Step S41: Initialize the learner Function of knowledge mastery level in learning tasks : ; In specific learning tasks, The effective length is ; Step S42: the Subtract the vectors by offset to get the difference vector ; Step S43: Master the horizontal position index matrix based on the knowledge obtained in step S37 , the difference vector obtained in step S42 The elements in the matrix The index of the corresponding position is assigned to the knowledge mastery level function ; Step S44: knowledge mastery level function Representation of knowledge weights for various learning tasks Aggregation is performed to predict the learner's ideal response.

[0011] Compared with the prior art, the present invention has the following beneficial effects: The method of the present invention modifies and introduces a variable-length knowledge weight vector representation module, so that the cognitive diagnosis deep model can obtain the associated cognitive state representation and variable-length knowledge weight representation vector (the number of valid elements in the vector is directly determined by the number of knowledge points in each learning task) according to the characteristics of different learning tasks, and adaptively generate index positions for each knowledge point subset, thereby overcoming the limitations of CI-CDM and being able to adapt to cognitive diagnosis in scenarios with different amounts of knowledge in multiple learning tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present disclosure and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without paying creative work.

[0013] Figure 1 is an architectural diagram of the cognitive diagnosis deep model of the present invention; Figure 2 It is a comparison chart of the DOA performance of the cognitive diagnosis deep model of the present invention and the benchmark model on three data sets; Figure 3 It is a box plot comparing the clustering results of the cognitive diagnosis deep model of the present invention and the benchmark model on four real data sets. DETAILED DESCRIPTION

[0014] In order to facilitate those of ordinary skill in the art to understand and implement the present invention, the following detailed description is given to each step of the method proposed by the present invention, and it should be understood that these embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention. In addition, it should be understood that after reading the content taught by the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent forms also fall within the scope defined by the appended claims of the present application.

[0015] Example The present invention discloses a cognitive diagnosis method that is adapted to different numbers of knowledge points under multiple learning tasks. The cognitive diagnosis method is implemented based on a cognitive diagnosis deep model. Figure 1 As shown, the cognitive diagnosis deep model includes an associated cognitive state representation module, a variable-length knowledge weight vector representation module, a knowledge point subset position index module, and a knowledge mastery level position index module; the cognitive diagnosis method includes the following steps: Step S1: The associated cognitive state representation module is based on the learner's complete cognitive state and test Matrix, to obtain the learner's knowledge mastery level representation on each learning task; The test matrix ;learner , learning task ; Conceptual knowledge ;Learning tasks Knowledge collection ;in, Indicates the number of knowledge points; Indicates the number of learning tasks; represents the number of learners; Indicates the task Measuring knowledge ; Indicates the task Unmeasured Knowledge ; Step S2, the variable-length knowledge weight vector representation module obtains a variable-length knowledge weight vector that maintains monotonicity constraints between the learner and each learning task knowledge point subset and adapts to different numbers of knowledge points by training the deep neural network parameter matrix; Step S3, the knowledge point subset position index module obtains the position index of the knowledge point subset of each learning task in the knowledge mastery level representation based on the obtained learner's knowledge mastery level representation on each learning task, the variable-length knowledge weight vector adapted to different numbers of knowledge points, and the number of knowledge points, and aligns the position index with the knowledge point subset at the same position in the variable-length knowledge weight vector; Step S4, the knowledge mastery level position index module converts the learner's knowledge mastery level representation matrix on each learning task into a knowledge mastery level function matrix based on the knowledge mastery level position index matrix, and then aggregates it with the variable-length knowledge weight vector of the corresponding learning task to predict the learner's ideal answer response.

[0016] Specifically, step S1 is based on the learner's complete cognitive state and test Matrix, to obtain the learner's knowledge mastery level representation on each learning task, including the following steps: Step S11: The associated cognitive state representation module converts the knowledge mastery level vector into the attribute vector of each learning task according to the attribute vector examined. and Each row is multiplied element by element to obtain the learner exist The corresponding knowledge mastery level representation matrix for each learning task : ; Assume learner The initial representation vector of the knowledge mastery level , then the corresponding test The matrix is ​​represented as: ;

[0017] learner exist The knowledge mastery level representation matrix corresponding to the learning task It is expressed as: ; Step S12: for learners The knowledge mastery level representation matrix corresponding to each learning task Sort the elements in each row in descending order to get the initial knowledge mastery level and location index Matrix; among them, the initial knowledge mastery level It is expressed as: ;

[0018] Position Index It is expressed as: ; Step S13: Remove the initial knowledge mastery level The all-0 columns in the table get the initial knowledge mastery level after removing the all-0 columns and location index : ; ; Step S14: Expand the initial knowledge mastery level after removing all zero columns Add a column of all zero vectors to the dimension and get the knowledge mastery level ,in, , Indicates the maximum amount of knowledge that each learning task can contain in the current data set, Representation Learning Task Number of knowledge items tested; level of knowledge mastered It is expressed as: .

[0019] Specifically, the step S2 of obtaining a variable-length knowledge weight vector that maintains monotonicity constraints between the learner and each learning task knowledge point subset and adapts to different numbers of knowledge points includes the following steps: Step S21: Establish the index code of the fuzzy measure in the knowledge weight vector, and unify the number of elements in the knowledge weight set to The power set of knowledge, ,in, By is a vector composed of the knowledge weights of the learning tasks; then the learning task , , The knowledge weight vectors are expressed as: ; ; ;

[0020] In the above formula, represents the vector element, in the first equation The subscripts are learning tasks The knowledge subset under consideration, in the second equation The subscripts are binary coded. In the third equation The subscript of the vector element at the corresponding position in the second equation is converted from binary code to decimal minus 1, which is used to represent the vector Position index in ; Step S22: Analyze the test Matrix, initialize the deep neural network parameter matrix combined with fuzzy measure learning , The matrix initialization formula is expressed as: hour, ; otherwise, ;

[0021] In the above formula, is the knowledge set corresponding to the learning task A proper subset of Representing a knowledge set In learning tasks The weight on ; Learning tasks , , The initial parameter matrix is ​​expressed as: ;

[0022] Step S23: Matrix parameter training to obtain knowledge weight representation for each learning task ,in, The learning task The weight of knowledge The vector composed of It is expressed as: ;

[0023] In the above formula, Representing a knowledge set correspond One row of elements in the matrix, that is correspond The third row of elements in the matrix is ​​based on the learning task in step S23 , , The initial parameter matrix It can be concluded that ; Step S24: Introduction , Indicates taking The element with the smallest weight in the set, Representation Learning Task Knowledge Collection A subset of The set of weights in the matrix, i.e. The matrix Column; knowledge weight in step S23 Expression The condition is equivalent to , The condition is equivalent to , and then according to and The value of is used as a condition to calculate the learning task The weights of the true subsets of each knowledge set above; Representation of knowledge weights in various learning tasks Add a row of all 0 elements at the bottom to set the knowledge for each learning task Weight Assign a value of 1, and the corresponding initial knowledge weight representation It is expressed as: ;

[0024] In the above formula, The number of non-zero elements or valid elements in each column is proportional to the number of knowledge points included in the learning task. Change and change, The knowledge weight is In column 1, The knowledge weight is In the second column, The knowledge weight is Column 3; Knowledge weight representation The values ​​of the elements will be continuously updated with the iteration of the program, thereby obtaining a variable-length knowledge weight vector that maintains monotonicity constraints with the knowledge point subsets of each learning task and adapts to different numbers of knowledge points.

[0025] Specifically, the specific steps of step S3 are as follows: Step S31, establishing a knowledge mastering level position index matrix; First, the test The elements of each row of the matrix are summed and 1 is subtracted from each sum to obtain a column vector , Represents the corresponding learning task The amount of knowledge contained in The maximum weight of ;

[0026] Further The vector is represented as: ; Step S32: Initialize an all-0 matrix ,right The matrix is ​​assigned from A sequence of integers decreasing to 0, represented by: ; ;

[0027] In the above formula, , Indicates the number of the learning task; Step S33: Test Sort the matrix in descending order, set When , the all-0 subcolumns are deleted, and The subcolumn following the column; otherwise, the subcolumn is not deleted, and the result is ; ; Step S34: Since each learning task examines different knowledge, the knowledge obtained in step S13 is and Multiply the matrices and extract the position index of the valid elements to get The matrix is ​​represented as: ; ;

[0028] In the above formula, Indicates the knowledge mastery level index from high to low of the learners' knowledge mastery level in each learning task; Step S35: Sort the matrix in descending order to get the position index matrix, is a value between 0 and The integer index of , represented as: ; Step S36: The positions of the corresponding elements in the matrix are The position index changes, and the corresponding formula is as follows: ; ; Combined with step S35 and Both are descending position weights, and the weight of knowledge The knowledge points corresponding to the binary codes from high to low are the learning tasks Attribute Vector The knowledge points examined from right to left can realize the index and knowledge weight of the knowledge mastery level function The index is consistent; Step S37, take Cumulative sum by column, then subtract 1 to obtain the knowledge mastery level position index matrix , which realizes the unification of the position index of the knowledge mastery level representation matrix in the knowledge mastery level function, where the matrix It is expressed as: ; .

[0029] Specifically, the step S4 of predicting the learner's ideal response includes the following steps: Step S41: Initialize the learner Function of knowledge mastery level in learning tasks : ;

[0030] In specific learning tasks, The effective length is According to step S21: In learning tasks superior: ; In learning tasks superior: ;

[0031] In learning tasks superior: ;

[0032] In the above formula, represents the vector element, in the first equation The subscripts are learning tasks The knowledge subset under consideration; in the second equation The subscript of The position index of Step S42: the Subtract the vectors by offset to get the difference vector : ;

[0033] Step S43: Master the horizontal position index matrix based on the knowledge obtained in step S37 , the difference vector obtained in step S42 The elements in the matrix The index of the corresponding position is assigned to the knowledge mastery level function , let the learner In learning tasks The knowledge mastery level position index is a vector , then: ;

[0034] Get learners In learning tasks Knowledge level function on for: ;

[0035] Similarly, learners In learning tasks Knowledge level function on for: ; ;

[0036] learner In learning tasks Knowledge level function on for: ; ;

[0037] Step S44: knowledge mastery level function Representation of knowledge weights for various learning tasks Aggregate and predict the learner's ideal response: learner In learning tasks The ideal response is ; learner In learning tasks The ideal response is ; learner In learning tasks The ideal response is .

[0038] The following is a specific experimental process to further illustrate the superiority of the cognitive diagnosis deep model and cognitive diagnosis method proposed in the present invention.

[0039] 1. Dataset and Preprocessing Dataset: In order to verify the effectiveness of the deep model of cognitive diagnosis proposed in this invention, this experiment conducted a comparative study in two scenarios: learning tasks with the same number of knowledge points and learning tasks with different numbers of knowledge points. The performance of the model was verified on three public datasets: Math1, Math2, FrcSub, and extracted Math_A (each learning task on Math1 was extracted to contain only 3 pieces of knowledge) and Math_B (each learning task on Math2 was extracted to contain only 4 pieces of knowledge). The statistical information of the dataset used in this experiment is shown in Table 1 below.

[0040] Table 1. Experimental dataset statistics ; Preprocessing: In all data sets, the same learning task may be answered multiple times by the learner, but only the learner's first answer record is retained; in order to ensure that each learner has enough answer records for diagnosis, only learners who answered more than 15 learning tasks are retained; the ratio of training set and test set in the data set is 0.2, that is, 0.2 is the test set and the rest is the training set.

[0041] 2. Benchmark Model NCD model: A deep learning cognitive diagnosis model with good generalization ability, which realizes cognitive diagnosis through complex interaction functions through neural networks.

[0042] IRT (Unidimensional Item Response Theory): A unidimensional item response theory model method that uses linear functions to model learners' potential traits and task characteristic parameters (such as difficulty and discrimination) to achieve response prediction.

[0043] MIRT (Multidimensional Item Response Theory), as a multidimensional extension of IRT, models the proficiency of multidimensional knowledge of learners and learning tasks.

[0044] Probabilistic Matrix Factorization (PMF), by decomposing the learner score matrix, obtains the latent feature vectors of learners and learning tasks.

[0045] 3. Evaluation Metrics Considering that the learning tasks in the dataset used in this experiment are objective, the scores of the learning tasks are 0 or 1, indicating wrong or correct answers respectively. Classification and regression evaluation indicators are used in this experiment. In terms of classification evaluation, this experiment uses prediction accuracy (ACC) and F1 score (F1-score) to evaluate the prediction performance. In terms of regression evaluation indicators, this experiment uses mean absolute error (MAE) to measure the gap between the predicted score and the actual score.

[0046] 4. Comparison of experimental results This experiment sets up two scenarios, one is a learning task scenario with the same number of knowledge points, and the other is a learning task scenario with different numbers of knowledge points.

[0047] ① The comparison of experimental results in learning task scenarios with the same number of knowledge points is shown in Table 2 below.

[0048] Table 2. Comparison of experimental results in learning task scenarios with the same number of knowledge points ; It can be seen from Table 2 that in the learning task scenario with the same number of knowledge points, the cognitive diagnosis deep model and the CI-CDM model provided by the present invention have basically the same prediction performance. This shows that the cognitive diagnosis deep model provided by the present invention is effective and is effective for generalizing the CI-CDM model.

[0049] ② The comparison of experimental results in learning task scenarios with different numbers of knowledge points is shown in Table 3 below.

[0050] Table 3. Comparison of experimental results under learning task scenarios with different numbers of knowledge points ; It can be seen from Table 3 that in the learning task scenarios with different numbers of knowledge points, the prediction performance of the cognitive diagnosis deep model provided by the present invention on the three data sets is significantly better than that of the four benchmark models, namely NCD, MIRT, IRT, and PMF; this shows that the cognitive diagnosis deep model provided by the present invention is superior in predicting learners' answer performance.

[0051] 5. Model Interpretability Analysis The interpretability analysis in this experiment includes DOA comparative analysis and learners’ knowledge mastery cluster analysis.

[0052] DOA Comparative Analysis Monotonicity is one of the basic assumptions of cognitive diagnostic model theory. The interpretability of the model depends on whether the diagnostic results conform to the monotonicity assumption. According to the monotonicity assumption, the more proficient a learner is in a certain knowledge, the higher the learner's score in the learning task containing this knowledge should be. Intuitively, if a learner Compare More knowledgeable ,Right now ,So Including Learning tasks The score should also be higher than exist The score on In order to evaluate the interpretability of cognitive diagnosis models, technicians in this field have proposed a comparative analysis of the degree of consistency (DOA). The larger the DOA, the more the model conforms to the monotonicity assumption and the better the interpretability of the model.

[0053] The DOA calculation formula is as follows: ; ; ; In the above formula, , Indicates the number of knowledge points; ,in Indicates the number of learning tasks; ,in represents the number of learners; yes exist proficiency in express exist score on; if , ;otherwise, ;if Include , ;otherwise ;if and All answered , ,otherwise .

[0054] like Figure 2 The DOA performance comparison diagram of the cognitive diagnosis deep model of the present invention and the benchmark model on three data sets is shown in FIG. Figure 2 It can be seen that the DOA performance of the cognitive diagnosis deep model of the present invention on the three data sets is better than that of the NCD model, and its DOA value is close to 0.9, while the DOA value of the NCD model is between 0.75 and 0.85; overall, the performance of the cognitive diagnosis deep model of the present invention on each data set is more stable and outstanding, indicating that the interpretability of the model is stronger than that of the NCD model.

[0055] Cluster analysis of learners' knowledge mastery According to the learners' cognitive states, the K-means algorithm is used to cluster the learners into four clusters. Indicates clusters; learners The average proficiency in all knowledge is express, The average proficiency express; and The calculation formula is as follows: ; ; Then, according to The clusters are sorted from small to large, corresponding to four categories: poor, medium, good, and excellent.

[0056] like Figure 3 The box plots are shown to compare the clustering results of the cognitive diagnosis deep model of the present invention and the benchmark model on four real data sets. Figure 3It can be seen that the cognitive diagnosis deep model of the present invention shows a greater ability to distinguish between three data sets (Math1, Math2, FrcSub), and the performance in different data sets shows a larger range of variation as the learner's proficiency increases, especially for learners with high proficiency, the discreteness in the box plot is larger; this shows that the cognitive diagnosis deep model of the present invention can more accurately distinguish the cognitive state of high-level learners. The NCD model, on the other hand, performs relatively smoothly, and the box plots in all tasks have a smaller range of variation, indicating that the NCD model is not as good as the cognitive diagnosis deep model provided by the present invention in distinguishing between learners with different proficiency. Whether it is a low-level or high-level learner, the prediction results of the NCD model do not fluctuate much, indicating that its cognitive diagnosis of learners of different levels is relatively vague.

[0057] The above is only a preferred embodiment of the present invention, and does not limit the present invention in other forms. Any technician familiar with the profession may use the above disclosed technical content to change or modify it into an equivalent embodiment with equivalent changes. However, any simple modification, equivalent change and modification made to the above embodiment according to the technical essence of the present invention without departing from the technical solution of the present invention still belongs to the protection scope of the technical solution of the present invention.

Claims

1. A cognitive diagnosis method that adapts to different numbers of knowledge points in multiple learning tasks, characterized in that: The cognitive diagnosis method is implemented based on a cognitive diagnosis deep model, wherein the cognitive diagnosis deep model includes an associated cognitive state representation module, a variable-length knowledge weight vector representation module, a knowledge point subset position index module, and a knowledge mastery level position index module; the cognitive diagnosis method includes the following steps: Step S1: The associated cognitive state representation module is based on the learner's complete cognitive state and test Matrix, to obtain the learner's knowledge mastery level representation on each learning task; The test matrix ;learner , learning task ; Conceptual knowledge ;Learning tasks Knowledge collection ;in, Indicates the number of knowledge points; Indicates the number of learning tasks; represents the number of learners; Representation Task Measuring knowledge ; Representation Task Unmeasured Knowledge ; Step S2, the variable-length knowledge weight vector representation module obtains a variable-length knowledge weight vector that maintains monotonicity constraints between the learner and each learning task knowledge point subset and adapts to different numbers of knowledge points by training the deep neural network parameter matrix; Step S3, the knowledge point subset position index module obtains the position index of the knowledge point subset of each learning task in the knowledge mastery level representation based on the obtained learner's knowledge mastery level representation on each learning task, the variable-length knowledge weight vector adapted to different numbers of knowledge points, and the number of knowledge points, and aligns the position index with the knowledge point subset at the same position in the variable-length knowledge weight vector; Step S4, the knowledge mastery level position index module converts the learner's knowledge mastery level representation matrix on each learning task into a knowledge mastery level function matrix based on the knowledge mastery level position index matrix, and then aggregates it with the variable-length knowledge weight vector of the corresponding learning task to predict the learner's ideal answer response.

2. According to claim 1, a cognitive diagnosis method adapted to different numbers of knowledge points in multiple learning tasks is characterized in that: Step S1 is based on the learner's complete cognitive state and test Matrix, to obtain the learner's knowledge mastery level representation on each learning task, including the following steps: Step S11: The associated cognitive state representation module converts the knowledge mastery level vector into With Quiz matrix Each row is multiplied element by element to obtain the learner exist The corresponding knowledge mastery level representation matrix for each learning task : ； In formula (1), express J OK K The set of real matrices of columns; Step S12: for learners The knowledge mastery level representation matrix corresponding to each learning task Sort the elements in each row in descending order to get the initial knowledge mastery level and location index matrix; Step S13: Remove the initial knowledge mastery level The all-0 columns in the table get the initial knowledge mastery level after removing the all-0 columns and location index ; Step S14: Expand the initial knowledge mastery level after removing all zero columns Add a column of all zero vectors to the dimension and get the knowledge mastery level ,in , Indicates the maximum amount of knowledge that each learning task can contain in the current data set, Representation Learning Task The number of pieces of knowledge examined.

3. A cognitive diagnosis method adapted to different numbers of knowledge points in multiple learning tasks according to claim 2, characterized in that: The step S2 of obtaining a variable-length knowledge weight vector that maintains monotonicity constraints between the learner and each learning task knowledge point subset and adapts to different numbers of knowledge points includes the following steps: Step S21: Establish the index code of the fuzzy measure in the knowledge weight vector, and unify the number of elements in the knowledge weight set to The power set of knowledge, ,in, By A vector composed of the knowledge weights of each learning task; Step S22: Analyze the test Matrix, initialize the deep neural network parameter matrix combined with fuzzy measure learning , The matrix initialization formula is expressed as: hour, ； otherwise, ； Formula (2)-Formula (3), is the knowledge set corresponding to the learning task A proper subset of Representing a knowledge set In learning tasks The weight on ; Step S23: Matrix parameter training to obtain knowledge weight representation for each learning task ,in, The learning task The weight of knowledge The vector composed of It is expressed as: ； In formula (4), Representing a knowledge set correspond 1 row of elements in a matrix; Step S24: Introduction , Indicates taking The element with the smallest weight in the set, Representation Learning Task Knowledge Collection A subset of The set of weights in the matrix, i.e. The matrix Column; in formula (4) The condition is equivalent to , The condition is equivalent to ,according to and The value of is used as a condition to calculate the learning task The weights of the true subsets of each knowledge set above; Representation of knowledge weights in various learning tasks Add a row of all 0 elements at the bottom to set the knowledge for each learning task Weight The value is assigned to 1, and the corresponding initial knowledge weight representation is recorded as ; The number of non-zero elements or valid elements in each column is proportional to the number of knowledge points included in the learning task. Change with change; knowledge weight representation The values ​​of the elements will be continuously updated with the iteration of the program, thereby obtaining a variable-length knowledge weight vector that maintains monotonicity constraints with the knowledge point subsets of each learning task and adapts to different numbers of knowledge points.

4. A cognitive diagnosis method adapted to different numbers of knowledge points in multiple learning tasks according to claim 3, characterized in that: The specific steps of step S3 are as follows: Step S31, establishing a knowledge mastering level position index matrix; First, the test The elements of each row of the matrix are summed and 1 is subtracted from each sum to obtain a column vector , Represents the corresponding learning task The amount of knowledge contained in The maximum weight of ； Step S32: Initialize an all-0 matrix ,right The matrix is ​​assigned from A sequence of integers decreasing to 0, represented by: ； In formula (6), , Indicates the number of the learning task; Step S33: Test Sort the matrix in descending order, set When , the all-0 subcolumns are deleted, and The subcolumn following the column; otherwise, the subcolumn is not deleted, and the result is ; Step S34: Since each learning task examines different knowledge, the knowledge obtained in step S13 is and Multiply the matrices and extract the position index of the valid elements to get The matrix is ​​represented as: ； Indicates the knowledge mastery level index from high to low of the learners' knowledge mastery level in each learning task; Step S35: Sort the matrix in descending order to get the position index matrix, is a value between 0 and integer index of ; Step S36: The positions of the corresponding elements in the matrix are The position index changes, and the corresponding formula is as follows: ； Combined with step S35 and Both are decreasing position weights, and knowledge weights The knowledge points corresponding to the binary codes from high to low are the learning tasks Attribute Vector The knowledge points examined from right to left can realize the index and knowledge weight of the knowledge mastery level function The index is consistent; Step S37, take Cumulative sum by column, then subtract 1 to obtain the knowledge mastery level position index matrix , which realizes the unification of the position index of the knowledge mastery level representation matrix in the knowledge mastery level function, where the matrix It is expressed as: 。 5. According to claim 4, a cognitive diagnosis method adapted to different numbers of knowledge points in multiple learning tasks is characterized in that: The step S4 of predicting the learner's ideal response comprises the following steps: Step S41: Initialize the learner Function of knowledge mastery level in learning tasks : ； In specific learning tasks, The effective length is ; Step S42: the Subtract the vectors by offset to get the difference vector ; Step S43: Master the horizontal position index matrix based on the knowledge obtained in step S37 , the difference vector obtained in step S42 The elements in the matrix The index of the corresponding position is assigned to the knowledge mastery level function ; Step S44: knowledge mastery level function Representation of knowledge weights for various learning tasks Aggregation is performed to predict the learner's ideal response.

Images