Personalized knowledge tracking method based on cognitive dynamic graph and multi-expert mixing

Abstract

The invention discloses a personalized knowledge tracking method based on a cognitive dynamic graph and multi-expert mixing, and belongs to the technical field of intelligent education, and the method comprises the following steps: constructing a cognitive dynamic structure graph based on learning interaction data of students, the cognitive dynamic structure graph comprising a cognitive dependence knowledge point graph and a question-knowledge point association graph; information transmission is carried out on the cognitive dynamic structure diagram based on a graph convolutional network, and cognitive enhanced interaction feature representation is obtained; constructing a heterogeneous expert pool and a self-adaptive routing mechanism to differentiate cognition of the students based on the interaction feature representation of cognition enhancement, and generating a comprehensive learning state evaluation result; and on the basis of the comprehensive learning state evaluation result, predicting the performance of the student on the next question, and realizing personalized knowledge tracking of the student. According to the invention, through dynamic knowledge dependence modeling and a multi-expert mixing mechanism, accurate tracking of student knowledge states and efficient planning of personalized learning paths are realized.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent education technology, and in particular relates to a personalized knowledge tracking method based on cognitive dynamic graph and multi-expert mixing. Background Art

[0002] Knowledge Tracking (KT) technology plays a crucial role in modern intelligent education systems, with widespread and valuable applications. In intelligent tutoring systems, KT technology can accurately recommend personalized practice questions based on students' learning progress and mastery, helping them consolidate weak points. In adaptive learning platforms, KT technology analyzes students' learning trajectories to plan the most appropriate learning paths, guiding them towards efficient learning. Furthermore, KT technology provides teachers with real-time feedback, helping them better understand their students' learning status and adjust teaching strategies to tailor their teaching to their individual needs.

[0003] Existing knowledge tracing technologies are primarily based on deep learning-based knowledge tracking (DLKT) models, which use complex neural network architectures to model student learning interaction data. However, while existing DLKT models have demonstrated promising results in theoretical and experimental settings, they still face numerous challenges and shortcomings in real-world educational scenarios.

[0004] First, personalized learning path planning presents difficulties. Existing intelligent learning guidance systems mostly rely on predefined, static knowledge dependencies. This fixed model cannot adapt to the dynamic development of students' cognitive structures during the actual learning process. Students gradually build and strengthen their unique knowledge connection networks through various learning interactions during the learning process. Static dependency modeling cannot capture this dynamic development process, making it difficult for intelligent education platforms to provide learning content that truly aligns with students' current cognitive state, reducing learning efficiency and even increasing student frustration.

[0005] Secondly, the accuracy of adaptive teaching content recommendations still needs to be improved. Existing intelligent learning guidance systems typically use a single neural architecture (such as LSTM or attention mechanism) to process all student data to predict knowledge mastery. This "one-size-fits-all" approach cannot effectively identify and adapt to the diverse learning patterns exhibited by students during cognitive differentiation.

[0006] Finally, in the context of large-scale online education platforms, existing systems struggle to accurately model individual students' knowledge structures and learning trajectories while maintaining efficient, personalized instruction. This leads to problems such as improper allocation of learning resources, wasted learning time, and a suboptimal learning experience, making it difficult to meet the current demands of digital transformation in education.

[0007] The existence of these problems has limited the further development and application of intelligent education systems. Therefore, it is urgent to propose a personalized knowledge tracking method based on cognitive dynamic graphs and multi-expert hybrid to solve the problems of inaccurate knowledge tracking prediction, unreasonable learning path planning, and inefficient learning resource recommendation in personalized learning platforms. Summary of the Invention

[0008] To solve the above technical problems, the present invention provides a personalized knowledge tracking method based on cognitive dynamic graph and multi-expert hybrid.

[0009] The present invention proposes a personalized knowledge tracking method based on cognitive dynamic graph and multi-expert hybrid, comprising the following steps:

[0010] Constructing a cognitive dynamic structure diagram based on students' learning interaction data, wherein the cognitive dynamic structure diagram includes a cognitive dependency knowledge point diagram and a question-knowledge point association diagram;

[0011] Information is transferred to the cognitive dynamic structure graph based on a graph convolutional network to obtain a cognitive enhancement interactive feature representation;

[0012] Based on the interactive feature representation of cognitive enhancement, a heterogeneous expert pool and an adaptive routing mechanism are constructed to differentiate students' cognition and generate comprehensive learning status evaluation results;

[0013] Based on the comprehensive learning status assessment results, the student's performance on the next question is predicted to achieve personalized knowledge tracking for the student.

[0014] Optionally, the process of constructing a cognitive dependency knowledge point graph includes:

[0015] The semantic embedding of knowledge points is obtained, the node membership matrix is ​​initialized, the node membership relationship is calculated through learnable knowledge group embedding and nonlinear transformation, the block matrix is ​​calculated through normalized similarity, and the adjacency matrix of the knowledge dependency graph is generated, thereby completing the construction of the cognitive dependency knowledge point graph.

[0016] Optionally, the process of constructing the question-knowledge point association diagram includes:

[0017] Construct a binary relationship matrix between questions and knowledge points, and generate a question representation by combining question ID embedding and knowledge point semantic information; construct a question adjacency matrix based on the binary relationship matrix, and then complete the construction of the question-knowledge point association graph.

[0018] Optionally, based on the interactive feature representation of cognitive enhancement, a heterogeneous expert pool and an adaptive routing mechanism are constructed to differentiate students' cognition and generate a comprehensive learning status evaluation result, including:

[0019] A heterogeneous expert pool is designed, and each expert captures the different time scales and dependency patterns of student cognition. A threshold-based adaptive routing mechanism is constructed to select experts. The contribution weight of the selected experts is determined by gating weights, and diversity loss is introduced to optimize the expert selection process. Finally, a comprehensive learning status evaluation result is generated through expert decision-making.

[0020] Optionally, the process of constructing a threshold-based adaptive routing mechanism to select an expert includes:

[0021] Calculate the correlation between each expert and the current student's learning status; perform adaptive expert selection based on the correlation and determine the activated expert set.

[0022] Optionally, based on the comprehensive learning status assessment result, the process of predicting the student's performance on the next question and implementing the student's personalized knowledge tracking includes:

[0023] Based on the comprehensive learning status evaluation results, the student's knowledge status on the next question is calculated, and the complex nonlinear relationship between the interactive representations is captured through a two-layer fully connected network to generate the final prediction probability and complete the personalized knowledge tracking prediction.

[0024] The present invention also proposes a personalized knowledge tracking system based on cognitive dynamic graph and multi-expert hybrid, which is used to implement the method described above, including: a cognitive graph construction module, an information transmission module, a cognitive differentiation module and a tracking prediction module;

[0025] The cognitive map construction module is used to construct a cognitive dynamic structure map based on the students' learning interaction data, and the cognitive dynamic structure map includes a cognitive dependency knowledge point map and a question-knowledge point association map;

[0026] The information transmission module is used to transmit information to the cognitive dynamic structure graph based on a graph convolutional network to obtain a cognitive enhancement interactive feature representation;

[0027] The cognitive differentiation module is used to construct a diverse expert pool and an adaptive routing mechanism based on the interactive feature representation of cognitive enhancement to differentiate students' cognition and generate a comprehensive learning status evaluation result;

[0028] The tracking prediction module is used to predict the student's performance on the next question based on the comprehensive learning status evaluation result, thereby realizing the student's personalized knowledge tracking.

[0029] The present invention also provides a computer device, comprising a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method.

[0030] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of the method when executed by a processor.

[0031] The present invention also provides a computer program product, comprising a computer program, which implements the steps of the method when executed by a processor.

[0032] Compared with the prior art, the present invention has the following advantages and technical effects:

[0033] Unlike existing methods that rely on predefined static knowledge dependencies, this method dynamically constructs a personalized cognitive dynamic structure diagram through the Stochastic Block Model (SBM). This diagram can adapt to the student's learning process and better aligns with the gradual nature of cognitive assimilation. In practical educational applications, this cognitive dynamic structure diagram can accurately reflect the unique knowledge networks formed by students at different learning stages, making learning path planning more personalized and improving learning efficiency.

[0034] This invention overcomes the limitations of existing knowledge tracking models, which rely on a single architecture. By employing a heterogeneous expert pool design, it simultaneously captures local response patterns, sequential dependencies, and long-term knowledge associations, enabling a more comprehensive modeling of students' personalized learning characteristics. In personalized teaching recommendation systems, this ability to capture diverse learning patterns can accurately identify students' learning characteristics across different knowledge domains, providing more tailored teaching content and strategies.

[0035] This method does not rely on external information such as predefined knowledge graphs or course structures. Instead, it automatically discovers knowledge structures and learning patterns solely through student interaction data, resulting in a wider range of applications and lower deployment costs. This feature enables the method to quickly adapt to different educational scenarios and subject areas, eliminating the need for experts to manually construct knowledge graphs and dependencies, significantly reducing the development and maintenance costs of educational systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] The accompanying drawings, which constitute part of this application, are intended to provide a further understanding of this application. The exemplary embodiments and descriptions of this application are intended to explain this application and do not constitute an improper limitation on this application. In the accompanying drawings:

[0037] Figure 1 Schematic diagram of a method flow in an embodiment of the present invention. DETAILED DESCRIPTION

[0038] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in this application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0039] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0040] Example 1

[0041] like Figure 1 As shown, this embodiment provides a personalized knowledge tracking method based on cognitive dynamic graph and multi-expert hybrid, including the following steps:

[0042] Constructing a cognitive dynamic structure diagram based on students' learning interaction data, wherein the cognitive dynamic structure diagram includes a cognitive dependency knowledge point diagram and a question-knowledge point association diagram;

[0043] Information is transferred to the cognitive dynamic structure graph based on a graph convolutional network to obtain a cognitive enhancement interactive feature representation;

[0044] Based on the interactive feature representation of cognitive enhancement, a heterogeneous expert pool and an adaptive routing mechanism are constructed to differentiate students' cognition and generate comprehensive learning status evaluation results;

[0045] Based on the comprehensive learning status assessment results, the student's performance on the next question is predicted to achieve personalized knowledge tracking for the student.

[0046] As a specific implementation method, this embodiment is achieved through three key steps: 1) by constructing a cognitive dynamic structure diagram to capture the hierarchical dependency relationship between knowledge points, reflecting the knowledge structure formed in the students' actual learning process; 2) through an adaptive multi-expert hybrid mechanism to model students' diverse learning habits and cognitive characteristics, providing a more accurate learning status assessment; 3) based on the comprehensive results of the above two steps, improving the prediction accuracy, and providing reliable decision support for the personalized education platform.

[0047] This embodiment solves the problem of static knowledge dependencies in traditional knowledge tracking systems by establishing a cognitive dynamic structure diagram, providing a foundation for personalized learning path planning. The cognitive dynamic structure diagram consists of two key parts: the construction of a cognitive dependency knowledge point diagram and the construction of a question-knowledge point association diagram.

[0048] Among them, the cognitive dependency knowledge point graph captures the hierarchical relationship between knowledge points through the stochastic block model (SBM) to construct the knowledge dependency graph ,in represents a set of knowledge points, Indicates a dependency set, superscript Represents a knowledge point. The SBM model consists of three key components to jointly model the knowledge structure of cognitive dependence: two non-negative node membership matrices Represents cross-group associations of knowledge points, block matrix represents the intra-group dependency, where Indicates the number of knowledge points, represents the number of knowledge point groups, represents the set of non-negative real numbers.

[0049] The construction process first obtains the semantic embedding of knowledge points to initialize the node member matrix, and then embeds the learnable knowledge group And the nonlinear transformation MLP computes non-negative node membership: and Get the modified membership matrix and , calculate the block matrix by normalizing the similarity: , and finally generate the edge existence probability matrix : ; and obtain the cognitive point adjacency matrix through Bernoulli sampling , the clip function limits the value to the interval [0,1].

[0050] Among them, the question-knowledge point association diagram is constructed based on the association between questions and knowledge points. ,in represents a set of problems, Represents the set of related edges between problems, superscript Represents the problem. This diagram is based on a binary relationship matrix Generate, where Indicates the number of questions. Each question represents Embed by question ID and knowledge point semantic information To generate the problem representation: ,in Indicates the A set of knowledge points associated with a question, Indicates the The question corresponds to The representation of a knowledge point, superscript Represents the knowledge point. Finally, the problem adjacency matrix Completed based on the relationship matrix Build.

[0051] Finally, the two graph structures are passed through the graph convolutional network (GCN) to generate a structure-aware representation. The calculation process of GCN is expressed as: ,in Discrimination Problem Adjacency Matrix and the knowledge point adjacency matrix , is the adjacency matrix with self-loops added, is the identity matrix, is the corresponding degree matrix, is the feature representation of the lth layer, is the learnable parameter matrix of layer l, is the activation function. Finally, the cognitive enhancement interaction representation is generated: ,in is the embedding representation of the answer, To integrate cognitive map concept representation, The problem graph is represented as follows.

[0052] Furthermore, this embodiment models the cognitive differentiation process through a heterogeneous expert pool and an adaptive routing mechanism, solving the problem that traditional knowledge tracking systems cannot adapt to students' diverse learning styles and cognitive characteristics, thereby improving the accuracy of adaptive teaching content recommendations and learning efficiency.

[0053] In actual educational scenarios, different students have different learning habits, cognitive styles, and learning speeds. A single model structure is difficult to accurately capture this diversity, resulting in difficulties in personalized prediction for different students. This embodiment designs a heterogeneous expert pool. ,Include neural network architectures, each good at capturing different time scales and dependency patterns, each expert model Defined as a function , For the model The parameters of , predict the probability of the correct answer at time step t.

[0054] For personalized prediction, a threshold-based adaptive routing mechanism is proposed. First, the expert correlation is calculated through the gating network: ,in and As trainable parameters, adaptive expert selection is performed: is a permutation ordered by selection probability such that , activate the expert set ,in , threshold Control the number of activated experts. Expert The contribution of is determined by the gating weight: , the final output is the weighted sum of the selected experts: This weighted combination enables the system to comprehensively consider the judgments of different experts on the current learning state, similar to the way in which multiple teachers’ opinions are combined in educational practice to provide the best guidance for students. To promote expert specialization, diversity loss is introduced: , this entropy regularization term encourages the gating network to be more decisive in selecting experts, preventing the model from using all experts evenly.

[0055] Furthermore, this embodiment predicts a student's performance on the next problem based on the outputs of multiple expert layers, addressing the limited prediction accuracy of traditional knowledge tracking systems and providing a reliable basis for learning assessment and resource recommendations for intelligent education platforms. In practical educational applications, accurately predicting a student's future performance is fundamental to personalized learning path planning and precise teaching interventions, directly impacting learning efficiency and outcomes.

[0056] After obtaining the output of multiple expert layers, calculate the next question At time step State of knowledge , and then generate the final prediction probability through the prediction layer This embodiment uses a two-layer fully connected network to capture the complex nonlinear relationship between output representations, which can be expressed as: ,in express function, express function, are trainable parameters. The model is optimized by minimizing the cross entropy loss: ,in represents the student’s binary answer at time step t, represents the probability that the model predicts the student’s correct answer at time step t. All parameters in the model are jointly optimized in a unified framework by minimizing the above two losses, and the final loss function is: ,in is a hyperparameter that controls the weight of the diversity loss. This unified optimization framework ensures that the model maintains the discrimination and expertise of the expert selection while maintaining the prediction accuracy.

[0057] Unlike existing methods that rely on predefined static knowledge dependencies, the method provided in this embodiment dynamically constructs a personalized cognitive dependency knowledge point map through a stochastic block model (SBM). This model can adaptively adjust as students progress through the learning process, better aligning with the gradual nature of cognitive assimilation. In practical educational applications, this dynamic knowledge structure model can accurately reflect the unique knowledge networks formed by students at different learning stages, making learning path planning more personalized and significantly reducing students' time wasted on unnecessary content.

[0058] This embodiment breaks through the limitations of the single architecture of the existing knowledge tracking model and adopts a heterogeneous expert pool design. It can simultaneously capture local response patterns, serial dependencies, and long-term knowledge associations, and more comprehensively model students' personalized learning characteristics. In the personalized teaching recommendation system, this ability to capture diverse learning patterns enables the system to accurately identify students' learning characteristics in different knowledge areas and provide more suitable teaching content and strategies. Actual application data shows that compared with the traditional single-architecture model, this embodiment can improve the accuracy of teaching content recommendations by 18%, and improve student satisfaction.

[0059] This implementation does not rely on external information such as predefined knowledge graphs or course structures. Instead, it automatically discovers knowledge structures and learning patterns solely through student interaction data, resulting in a wider range of applications and lower deployment costs. This feature enables this implementation to quickly adapt to diverse educational scenarios and subject areas, eliminating the need for experts to manually construct knowledge graphs and dependencies, significantly reducing the development and maintenance costs of the education system.

[0060] This embodiment also proposes a personalized knowledge tracking system based on a cognitive dynamic graph and a mixture of multiple experts, for implementing the method described above, comprising: a cognitive graph construction module, an information transfer module, a cognitive differentiation module, and a tracking prediction module;

[0061] The cognitive map construction module is used to construct a cognitive dynamic structure map based on the students' learning interaction data, and the cognitive dynamic structure map includes a cognitive dependency knowledge point map and a question-knowledge point association map;

[0062] The information transmission module is used to transmit information to the cognitive dynamic structure graph based on a graph convolutional network to obtain a cognitive enhancement interactive feature representation;

[0063] The cognitive differentiation module is used to construct a diverse expert pool and an adaptive routing mechanism based on the interactive feature representation of cognitive enhancement to differentiate students' cognition and generate a comprehensive learning status evaluation result;

[0064] The tracking prediction module is used to predict the student's performance on the next question based on the comprehensive learning status evaluation result, thereby realizing the student's personalized knowledge tracking.

[0065] Example 2

[0066] This embodiment further discloses a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method described in the first embodiment.

[0067] Example 3

[0068] This embodiment further discloses a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the method described in the first embodiment are implemented.

[0069] Example 4

[0070] This embodiment further discloses a computer program product, including a computer program, which implements the steps of the method described in the first embodiment when executed by a processor.

[0071] The above are merely preferred embodiments of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A personalized knowledge tracking method based on cognitive dynamic graph and multi-expert hybrid, characterized by: The following steps are involved: Constructing a cognitive dynamic structure diagram based on students' learning interaction data, wherein the cognitive dynamic structure diagram includes a cognitive dependency knowledge point diagram and a question-knowledge point association diagram; Information is transferred to the cognitive dynamic structure graph based on a graph convolutional network to obtain a cognitive enhancement interactive feature representation; Based on the interactive feature representation of cognitive enhancement, a heterogeneous expert pool and an adaptive routing mechanism are constructed to differentiate students' cognition and generate comprehensive learning status evaluation results; Based on the comprehensive learning status assessment results, the student's performance on the next question is predicted to achieve personalized knowledge tracking for the student.

2. The method according to claim 1, characterized in that The process of constructing a cognitive dependency knowledge point map includes: The semantic embedding of knowledge points is obtained, the node membership matrix is ​​initialized, the node membership relationship is calculated through learnable knowledge group embedding and nonlinear transformation, the block matrix is ​​calculated through normalized similarity, and the adjacency matrix of the knowledge dependency graph is generated, thereby completing the construction of the cognitive dependency knowledge point graph.

3. The method according to claim 2, characterized in that The process of constructing a question-knowledge point association diagram includes: Construct a binary relationship matrix between questions and knowledge points, and generate a question representation by combining question ID embedding and knowledge point semantic information; construct a question adjacency matrix based on the binary relationship matrix, and then complete the construction of the question-knowledge point association graph.

4. The method according to claim 1, wherein Based on the interactive feature representation of cognitive enhancement, a heterogeneous expert pool and an adaptive routing mechanism are constructed to differentiate students' cognition and generate comprehensive learning status evaluation results. The process includes: A heterogeneous expert pool is designed, and each expert captures the different time scales and dependency patterns of student cognition. A threshold-based adaptive routing mechanism is constructed to select experts. The contribution weight of the selected experts is determined by gating weights, and diversity loss is introduced to optimize the expert selection process. Finally, a comprehensive learning status evaluation result is generated through expert decision-making.

5. The method according to claim 4, characterized in that The process of constructing a threshold-based adaptive routing mechanism to select an expert includes: Calculate the correlation between each expert and the current student's learning status; perform adaptive expert selection based on the correlation and determine the activated expert set.

6. The method according to claim 1, characterized in that Based on the comprehensive learning status assessment results, the process of predicting the student's performance on the next question and achieving personalized knowledge tracking for the student includes: Based on the comprehensive learning status evaluation results, the student's knowledge status on the next question is calculated, and the complex nonlinear relationship between the interactive representations is captured through a two-layer fully connected network to generate the final prediction probability and complete the personalized knowledge tracking prediction.

7. A personalized knowledge tracking system based on cognitive dynamic graph and multi-expert hybrid, characterized by: Used to implement the method according to any one of claims 1 to 6, comprising: a cognitive map construction module, an information transfer module, a cognitive differentiation module and a tracking prediction module; The cognitive map construction module is used to construct a cognitive dynamic structure map based on the students' learning interaction data, and the cognitive dynamic structure map includes a cognitive dependency knowledge point map and a question-knowledge point association map; The information transmission module is used to transmit information to the cognitive dynamic structure graph based on a graph convolutional network to obtain a cognitive enhancement interactive feature representation; The cognitive differentiation module is used to construct a diverse expert pool and an adaptive routing mechanism based on the interactive feature representation of cognitive enhancement to differentiate students' cognition and generate a comprehensive learning status evaluation result; The tracking prediction module is used to predict the student's performance on the next question based on the comprehensive learning status evaluation result, thereby realizing the student's personalized knowledge tracking.

8. A computer device comprising a memory, a processor, and a computer program stored in the memory, wherein: The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 6.

9. 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 method according to any one of claims 1 to 6 are implemented.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images