Explainable knowledge tracing method and device based on hypergraph convolution attention mechanism

Abstract

The application provides an explainable knowledge tracking method (HCAKT) and device based on hypergraph convolution attention mechanism. Wherein, by introducing the double parameter input mechanism of 2PLM (double parameter logic model), HCAKT effectively copes with the challenge of significant skill difficulty difference in the same learning project, not only improves the flexibility of the model, but also enhances the explainability of the model. In addition, a hypergraph network is also added to the model, which is a key tool for parameter balancing, effectively avoiding the potential interference of redundant information on the performance of the model, and further improving the accuracy of the prediction by optimizing the mutual relationship between the parameters. Finally, HCAKT also integrates a convolution attention mechanism to simulate and capture the forgetting law of students. This mechanism can deeply analyze the memory decay characteristics of students in the learning process, so as to more truly reflect the learning state of students.

Description

Technical Field

[0001] This application relates to an interpretable knowledge tracking method and apparatus based on a hypergraph convolutional attention mechanism, belonging to the field of artificial intelligence technology. Background Technology

[0002] Knowledge tracing (KT) is an intelligent education method that analyzes students' answer history to build a dynamic model of knowledge mastery, predict future learning performance, assist in personalized learning path planning, dynamically adjust teaching content, and evaluate learning effectiveness in real time, providing precise support for education.

[0003] The purpose of knowledge tracking is to predict students' real-time mastery of knowledge concepts by analyzing their performance on past problems, thereby promoting personalized and adaptive teaching. In the field of artificial intelligence, knowledge tracking is rapidly developing and playing a significant role in education. It is giving rise to various intelligent education service platforms, allowing students to significantly improve their learning efficiency and focus more on their weaknesses through unique and targeted services. Simultaneously, teachers can implement more precise and effective teaching guidance based on students' knowledge and ability assessment results, jointly promoting students' all-round development.

[0004] Currently, mainstream methods in the field of knowledge tracing fall into three categories: probabilistic methods, attention-based methods, and deep learning-based methods. The probabilistic method, Bayesian Knowledge Tracing (BKT), uses Hidden Markov Models to predict students' knowledge states. However, it is limited by insufficient exploration of the flexibility of students' knowledge changes and the single setting of skill dimensions, failing to capture multidimensional skills, resulting in significant prediction errors. Subsequent research introduced Item Response Theory (IRT) into the field of knowledge tracing. This not only improved the interpretability of the model parameters, making the model more convincing, but also demonstrated through experimental results that the introduction of this module significantly improved prediction performance. However, both BKT and IRT face the challenge of comprehensively capturing multidimensional skills, affecting the accuracy of predictions.

[0005] To overcome the limitations of traditional methods in capturing multidimensional skills, Piech et al. proposed Deep Knowledge Tracing (DKT), a deep learning-based knowledge tracing method. DKT uses Long Short-Term Memory (LSTM) to alleviate the limitations of skill segregation and binary assumptions. However, due to the limitations of LSTM, DKT lacks the ability to accurately process each student's skill. To further improve DKT, Dynamic Key-Value Memory Networks (DKVMN) were proposed. This method utilizes the latent relationships between skills and tracks the corresponding knowledge states. Moreover, DKVMN also has an update component to store the student's knowledge state during the learning process. Furthermore, Deep-IRT, improved based on DKVMN, enhances parameter interpretability for estimating item difficulty and skill difficulty, but it is still limited by the assumption that student ability is overly dependent on item features and fails to effectively address the challenge of items with the same skills but vastly different difficulties, affecting the accuracy and interpretability of predictions.

[0006] Ghosh et al. pointed out the error in early KT methods that assumed items with the same skill were equivalent, and proposed an attention-based KT method (Context-Aware Attentive Knowledge Tracing, AKT). It effectively filters and weights past learning data by finely optimizing parameter weights to reveal the deep connections between skills and solved items. While this method can provide state-of-the-art performance in predictions, AKT cannot directly reflect the trajectory of students' ability changes for each skill, thus limiting the interpretability of the model parameters.

[0007] In summary, the main challenge of KT at present is to improve the interpretability of the model and parameters, and reduce the impact of redundant information on the model without reducing prediction accuracy. Summary of the Invention

[0008] This application provides an interpretable knowledge tracking method and apparatus based on a hypergraph convolutional attention mechanism to solve the aforementioned problems in the prior art.

[0009] In a first aspect, embodiments of this application provide an interpretable knowledge tracing method based on a hypergraph convolutional attention mechanism, comprising:

[0010] Predicting students' skill mastery using an interpretable model based on hypergraph convolutional attention mechanism;

[0011] The interpretable model based on the hypergraph convolutional attention mechanism includes: an interpretable module based on 2PLM, a hypergraph convolutional balancing parameter module, and a convolutional attention memory component module.

[0012] The 2PLM-based interpretable module includes an extraction module and a student module. The extraction module independently extracts practice difficulty parameters and skill difficulty parameters based on the practice attempted by the student and the skills required to solve the practice, and then merges the practice difficulty parameters and skill difficulty parameters to obtain a comprehensive difficulty parameter. The student module calculates the student's potential knowledge state based on the knowledge the student has acquired over a set time period. The 2PLM-based interpretable module also predicts the response probability to the practice based on the comprehensive difficulty parameter and the student's potential knowledge state.

[0013] The hypergraph convolutional balancing parameter module optimizes the input parameters using a neural network model based on the hypergraph architecture to obtain optimized input parameters, and calculates retention parameters, forgetting parameters, and influence parameters based on the optimized input parameters; wherein, the input parameters of the neural network model based on the hypergraph architecture include the vector representation of the student's answer to the exercise and the student's past skill potential value;

[0014] The convolutional attention memory component module fuses the retention parameter, the forgetting parameter, and the influence parameter based on an attention mechanism to obtain a forgetting vector;

[0015] The interpretable model based on the hypergraph convolutional attention mechanism predicts the student's skill mastery based on the response probability to the exercise, the student's past skill potential value, and the forgetting vector, thereby obtaining the student's skill potential value.

[0016] Based on the above methods, optionally, the extraction module is based on the first... Embedded vectors of each exercise The calculation process for the practice difficulty parameter using a feedforward neural network is as follows:

[0017]

[0018]

[0019] In the formula, and Represents the weight vector. Indicates the number of hidden layers. The last layer of the hidden layer Then it means the first The difficulty parameter corresponding to each exercise. and Represents the bias parameter vector;

[0020] The extraction module is also based on solving the first Embedding vectors of skills required for each exercise The process of obtaining the skill difficulty parameters and calculating them using a feedforward neural network is as follows:

[0021]

[0022]

[0023] In the formula, and This represents the weight vector, the last layer of the hidden layer. Indicates the first The skill difficulty parameters corresponding to each skill and This represents the bias parameter vector.

[0024] Based on the above methods, optionally, in the student module, students can... Knowledge acquired over a period of time To calculate the students' potential changes in their knowledge status:

[0025]

[0026] in, Indicates the first OK vector, Indicates the first Attention weight for each skill;

[0027] The process for predicting student abilities is as follows:

[0028]

[0029]

[0030] in, This represents the weight vector, where m represents the number of hidden layers. ,and . , This represents the bias parameter vector.

[0031] Based on the above method, optionally, the 2PLM-based interpretable module uses students in time. Solve the exercises ability With two difficulty parameters and The difference between the sums is used to predict the probability of a student's response to an exercise:

[0032]

[0033] in, This represents the predicted response probability.

[0034] Based on the above method, optionally, the 2PLM-based interpretable module calculates the model loss using cross-entropy loss:

[0035]

[0036] in, This represents the cross-entropy loss.

[0037] Based on the above method, optionally, the student's past skill potential value optimized based on the neural network model of the hypergraph architecture. The calculation process is as follows:

[0038]

[0039] in, It is a weight vector. This is the bias parameter vector;

[0040] Vector representation of students' responses to exercises and students' past skill potential The optimization process in hypergraph networks is as follows:

[0041]

[0042]

[0043] in, , and It's a hyperparameter. This indicates the number of recursions in the recursive architecture.

[0044] Based on the above method, optionally, the reserved parameter Forgetting parameters and influencing parameters The calculation formula is:

[0045]

[0046]

[0047]

[0048] in, , , , , and Represents the weight vector. , and This represents the bias parameter vector.

[0049] Based on the above method, optionally, the formula for obtaining the forgetting vector by the convolutional attention memory component module is expressed as:

[0050]

[0051]

[0052] in, It is the first The vector representing the factor. Reserved parameters from the above modules Forgetting parameters and influencing parameters composition, It refers to the number of attention layers; This indicates the proportion of the input in the attention layer;

[0053] In addition, a unique attention weight was assigned to the forgetting factor:

[0054]

[0055] in, This represents the attention weight.

[0056] Based on the above method, optionally, the formula for calculating the potential value of the skills mastered by students using the interpretable model based on the hypergraph convolutional attention mechanism is expressed as follows:

[0057]

[0058] in, This indicates that a one-dimensional convolutional network is used to capture and preserve parameters. With forgetting parameters The interrelationships within.

[0059] Secondly, embodiments of this application also provide an interpretable knowledge tracking device based on a hypergraph convolutional attention mechanism, comprising:

[0060] The prediction unit is used to predict students' skill mastery using an interpretable model based on a hypergraph convolutional attention mechanism.

[0061] The interpretable model based on the hypergraph convolutional attention mechanism includes: an interpretable module based on 2PLM, a hypergraph convolutional balancing parameter module, and a convolutional attention memory component module.

[0062] The 2PLM-based interpretable module includes an extraction module and a student module. The extraction module independently extracts practice difficulty parameters and skill difficulty parameters based on the practice attempted by the student and the skills required to solve the practice, and then merges the practice difficulty parameters and skill difficulty parameters to obtain a comprehensive difficulty parameter. The student module calculates the student's potential knowledge state based on the knowledge the student has acquired over a set time period. The 2PLM-based interpretable module also predicts the response probability to the practice based on the comprehensive difficulty parameter and the student's potential knowledge state.

[0063] The hypergraph convolutional balancing parameter module optimizes the input parameters using a neural network model based on the hypergraph architecture to obtain optimized input parameters, and calculates retention parameters, forgetting parameters, and influence parameters based on the optimized input parameters; wherein, the input parameters of the neural network model based on the hypergraph architecture include the vector representation of the student's answer to the exercise and the student's past skill potential value;

[0064] The convolutional attention memory component module fuses the retention parameter, the forgetting parameter, and the influence parameter based on an attention mechanism to obtain a forgetting vector;

[0065] The interpretable model based on the hypergraph convolutional attention mechanism predicts the student's skill mastery based on the response probability to the exercise, the student's past skill potential value, and the forgetting vector, thereby obtaining the student's skill potential value.

[0066] The interpretable knowledge tracing method (HCAKT) and apparatus based on hypergraph convolutional attention mechanism provided in this application effectively address the challenge of significant differences in skill difficulty within the same learning item by introducing a two-parameter input mechanism of 2PLM (two-parameter logical model). This not only improves the model's flexibility but also enhances its interpretability. Furthermore, a hypergraph network is incorporated into the model, serving as a key tool for parameter balancing. This effectively avoids the potential interference of redundant information on model performance and further improves prediction accuracy by optimizing the relationships between parameters. Finally, HCAKT also integrates a convolutional attention mechanism to simulate and capture students' forgetting patterns. This mechanism can deeply analyze the memory decay characteristics of students during the learning process, thus more realistically reflecting students' learning status. Attached Figure Description

[0067] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application. Furthermore, these drawings and textual descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concepts of this application to those skilled in the art through reference to specific embodiments.

[0068] Figure 1 A schematic diagram of the architecture of an interpretable model based on hypergraph convolutional attention mechanism provided in one embodiment of this application;

[0069] Figure 2 A two-parameter input framework diagram based on 2PLM is provided for one embodiment of this application;

[0070] Figure 3 This is a schematic diagram of the structure of a hypergraph balancing factor and a convolutional attention memory component provided in one embodiment of this application;

[0071] Figure 4 This is a schematic diagram of the convolution mapping of a convolutional attention memory component provided in one embodiment of this application. Detailed Implementation

[0072] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the embodiments of this application. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0073] To address the issues raised in the background section, this paper proposes a novel interpretable knowledge tracing method based on hypergraph convolutional attention mechanism. This method predicts students' skill mastery through an interpretable model based on hypergraph convolutional attention mechanism.

[0074] The interpretable model based on the hypergraph convolutional attention mechanism mainly consists of two modules: 1) a module for acquiring student learning features; and 2) a module for updating the learning process. This method uses a two-parameter input, which, compared to previous Deep-IRT models, provides better interpretability for acquiring student ability parameters and project difficulty features during the learning process.

[0075] Furthermore, when updating students' knowledge mastery, relying solely on model input as the basis for forgetting and updating can negatively impact prediction performance if students' skills don't match their past knowledge well when facing new projects. To address this, this solution introduces a hypergraph neural network into the update module, achieving deep integration of students' historical knowledge and new project skills, thereby obtaining more accurate parameter settings. Simultaneously, combining this with an attention convolution module further enhances the ability to extract parameter features, significantly improving prediction accuracy.

[0076] The performance of our proposed method is compared with that of earlier KT methods on public datasets. The results show that our proposed method achieves significant improvements in both accuracy and interpretability. The main contributions of this approach are as follows:

[0077] 1) By establishing a two-parameter IRT logic module to estimate skill parameters and project parameters, the parameter interpretability of the KT model is improved. This can effectively address the challenge of large differences in skill difficulty within the same project, thereby improving the model's adaptability and accuracy, and making the KT model of this solution more convincing.

[0078] 2) By balancing the model's input with past knowledge parameters through a hypergraph neural network, the potential interference of redundant information on model performance is effectively mitigated. Furthermore, the memory update component composed of convolutional attention extracts features from the balanced parameters, which is more in line with the forgetting curve and significantly improves the model's performance and predictive ability.

[0079] Two-parameter IRT logic model:

[0080] As is well known, the IRT model has high interpretability. However, in the standard IRT model, learning ability remains constant throughout the learning process, which clearly does not conform to the real-world learning patterns. Although many recent studies have extended the standard IRT, these studies' IRT models all assume a single dimension of ability. Specifically, they only consider exercises with independent multiple skills, but not exercises with multiple different skills. Therefore, this approach adopts a more popular new IRT model: the Two-Parameter Logistic Model (2PLM). This new IRT model with two parameters ensures that exercises and skills can be mapped at each step, thus handling not only exercises with independent multiple skills but also exercises with multiple different skills, further improving the model's interpretability.

[0081] Hypergraph Neural Networks and Convolutional Attention:

[0082] In the field of natural language processing, many studies employ hypergraph networks to optimize the hidden layer weights and latent variables of LSTMs to express their contextual shifts. This approach utilizes the methods mentioned above, which not only optimizes the forgetting parameters used for subsequent updates to students' skill mastery but also balances past mastery levels, resulting in a significant improvement in experimental results.

[0083] Considering that the memory update module extracted using simple extraction methods cannot effectively highlight the important role of forgetting patterns in students' learning process, this solution introduces a convolutional attention mechanism. This mechanism aims to combine the obtained forgetting factors to simulate students' forgetting patterns using a convolutional attention-based memory component. The results are significantly better than those obtained using simple extraction methods, demonstrating the significant impact of forgetting factors on students' learning process.

[0084] For ease of explanation, the following assumptions are made:

[0085] There exists a set of exercises. Students proceed in a certain order Learn from the problems encountered in the exercises, among which Indicates that the student is in The questions answered by Bu This represents the total number of steps a student takes when answering a question. Indicates that the student is in Step's answer is correct. That is, when the student answers... When the answer is correct, ,otherwise, .

[0086] Previous studies have suffered from poor interpretability of model parameters, meaning they cannot represent the abilities of each student at every stage of their past learning. To address this issue, this paper proposes an interpretable model based on a hypergraph convolutional attention mechanism, distinct from other Deep-IRT approaches. The model diagram is shown below. Figure 1 As shown, the model mainly consists of two modules: 1) an interpretable module based on 2PLM; and 2) a hypergraph convolution balancing parameter module and a convolution attention memory component module. In summary, the structure and function of each module are as follows:

[0087] The 2PLM-based interpretable module includes an extraction module and a student module. The extraction module independently extracts the practice difficulty parameter and skill difficulty parameter based on the practice attempted by the student and the skills required to solve the practice, and then merges the practice difficulty parameter and skill difficulty parameter to obtain a comprehensive difficulty parameter. The student module calculates the student's potential knowledge state based on the knowledge mastered by the student in a set time period. The 2PLM-based interpretable module also predicts the response probability to the practice based on the comprehensive difficulty parameter and the student's potential knowledge state.

[0088] The hypergraph convolutional balancing parameter module optimizes the input parameters using a neural network model based on the hypergraph architecture to obtain optimized input parameters, and calculates retention parameters, forgetting parameters, and influence parameters based on the optimized input parameters. The input parameters of the neural network model based on the hypergraph architecture include the vector representation of the student's response to the exercise and the student's past skill potential values.

[0089] The convolutional attention memory component module fuses the retention parameters, forgetting parameters, and influencing parameters based on the attention mechanism to obtain the forgetting vector;

[0090] An interpretable model based on hypergraph convolutional attention mechanism predicts students’ skill mastery based on the response probability to practice, students’ past skill potential, and forgetting vector, thus obtaining the skill potential value mastered by students.

[0091] Specifically, for interpretable modules based on 2PLM:

[0092] Previous studies using IRT generally assumed that each parameter was item-dependent and that the implied problem difficulty and skill level were constant. This assumption fails when there is a significant difference between problem difficulty and skill mastery. Therefore, predicting student learning behavior becomes difficult in such situations. To address this issue, this paper proposes an interpretable IRT module based on 2PLM, comprising two sub-modules: 1) an independent module for extracting problem difficulty / skill difficulty; and 2) a student module based on DKVMN, such as... Figure 2 As shown, the extraction module takes the student's attempted exercises and the skills required to solve them as inputs, extracting the difficulty factor for the exercises and the mastery factor for the skills, respectively. Then, by combining this with the student module and the student's actual situation, the probability of the student answering the exercises correctly can be calculated. Because the proposed module extracts exercises and skills separately, the prediction accuracy is not reduced.

[0093] In the extraction module, IRT has two difficulty parameters, one of which is the practice difficulty parameter. The other is the skill difficulty parameter. This module combines these two difficulty features to generate entirely new difficulty features. This refers to the overall difficulty parameter. For a more intuitive explanation, the input in the extraction module is the [missing information - likely a number or parameter]. Embedded vectors of each exercise The calculation process using a feedforward neural network is as follows:

[0094] (1)

[0095] (2)

[0096] In the formula, and Represents the weight vector. Indicates the number of hidden layers. The last layer of the hidden layer Then it means the first The difficulty parameter corresponding to each exercise. and This represents the bias parameter vector.

[0097] Similarly, to calculate the difficulty of a skill, the same method is used based on solving the first... Embedding vectors of skills required for each exercise The process of obtaining the skill difficulty parameters and calculating them using a feedforward neural network is as follows:

[0098] (3)

[0099] (4)

[0100] in, and This represents the weight vector, the last layer of the hidden layer. Indicates the first The skill difficulty parameters corresponding to each skill and This represents the bias parameter vector.

[0101] In the student module, this method is implemented through students'... Knowledge acquired over a period of time To calculate students' potential changes in their knowledge status of:

[0102] (5)

[0103] in, Indicates the first OK Vector, and Indicates the first Attention weight for each skill.

[0104] Similar to the weight calculation method in DKVMN, the prediction process for student ability is as follows:

[0105] (6)

[0106] (7)

[0107] in, This represents the weight vector, where m represents the number of hidden layers. ,and , This represents the bias parameter vector. Unlike previous Deep-IRT methods, this approach predicts student abilities through a separate student module, independent of the project difficulty parameter. Furthermore, The value also represents the student's potential learning ability. In other words, It can be considered a measurement model.

[0108] The proposed method uses students in time Solve the exercises ability With two difficulty parameters and The difference between the sums is used to predict the probability of a student's response to an exercise. :

[0109] (8)

[0110] Then, the cross-entropy of the classification error is mapped using the loss function employed in this scheme, i.e., by predicting the response probability. and true response probability To perform the calculation:

[0111] (9)

[0112] in, This represents the cross-entropy loss.

[0113] Student Hypergraph Network and Convolutional Attention Mechanism Module

[0114] During long-term learning, it's crucial not only to make good use of current input but also to consider how to leverage past information to optimize forgetting parameters and update students' knowledge status. The simplest method is to use an LSTM forgetting gate for basic data processing; another approach is to add weights to balance the input data. and past potential values However, as the learning process progresses, the number of weights increases dynamically, resulting in too many parameters and making it impossible to successfully estimate the result.

[0115] In standard LSTM, latent variables change over time, but their update weights remain fixed, lacking good targeting for different time points. To address this issue, various hypergraph networks have been proposed across different fields to optimize LSTM. These networks optimize the non-shared weights at each time point in the hidden layers to support the main recurrent neural network, and these studies have shown that the results of this novel LSTM outperform those of traditional LSTM. Inspired by these studies, this approach also incorporates this novel hypergraph network into the memory update component. It not only generates corresponding non-shared weights for each time point but also effectively avoids parameter redundancy.

[0116] Figure 3 The hypergraph architecture and the proposed convolutional attention memory component module are presented. The hypergraph optimizes the weights generated based on past data, which not only improves the interpretability of the parameters in the model, but also provides more accurate input for the subsequent memory component to update the student's knowledge state.

[0117] Specifically, in the method update Previously, the Hypergraph architecture used time points... arrive The obtained latent variables To balance the current input and past potential values .in This indicates the number of times a student has tried a potential variable in the past.

[0118] The input to the hypergraph architecture is an embedding vector. and past potential values When students answer questions When, according to the input Calculate the embedding vector .also, The calculation process is as follows:

[0119] (10)

[0120] in, It is a weight vector. This is the bias parameter vector;

[0121] Vector representation of students' responses to exercises and students' past skill potential The optimization process in hypergraph networks is as follows:

[0122] (11)

[0123] (12)

[0124] in, , and It's a hyperparameter. This indicates the number of recursions in the recursive architecture. and This represents the weight vector.

[0125] if ,So ,and Because of the repeated multiplication in formulas (11) and (12), ... and A balance was achieved. This scheme further optimized the number of recursions in subsequent experiments.

[0126] Optimized in the hypergraph architecture and As input to estimate the retention parameters and forgetting parameters These two parameters play a crucial role in the convolutional attention memory module. Compared to the previous Deep-IRT, this optimized solution... and This allows for a better balance of parameters, resulting in improved accuracy and results. Furthermore, this method also extracts an influencing parameter. right Optimize. For these three parameters (retain parameters) Forgetting parameters and influencing parameters The calculation process for ) is as follows:

[0127] (13)

[0128] (14)

[0129] (15)

[0130] in, , , , , and Represents the weight vector. , and This represents the bias parameter vector.

[0131] After obtaining the required forgetting parameters, and considering how to make the forgetting module better match the students' forgetting patterns, this scheme uses convolutional attention. This can fuse the obtained parameters to improve their correlation and reduce noise to mitigate its impact on the data, making it more accurate. The formula for fusing the forgetting vector using the convolutional attention memory component module is expressed as follows:

[0132] (16)

[0133] (17)

[0134] in, It is the first The vector representing the factor. Reserved parameters from the above modules Forgetting parameters and influencing parameters composition, It refers to the number of attention layers; This indicates the proportion of the input in the attention layer;

[0135] In addition, a unique attention weight was assigned to the forgetting factor. :

[0136] (18)

[0137] This scheme uses the reweighting in formula (18) to simulate the forgetting pattern of students. Specifically, it uses a one-dimensional convolutional network to capture and retain parameters. With forgetting parameters Interrelationships, such as Figure 4 As shown, it can be simply defined as:

[0138] (19)

[0139] Finally, the obtained parameters are denoted as the forgetting vector. Together with the skill parameters and past latent values, they are used to predict student performance and update the student's skill latent values. That is, the formula for calculating the student's skill latent values ​​using the interpretable model based on the hypergraph convolutional attention mechanism is expressed as:

[0140] (20)

[0141] During the update process, the model can acquire more potential value information, enabling it to better and more accurately capture changes in students' knowledge status.

[0142] In summary, this paper proposes a knowledge tracing method (HCAKT) that integrates hypergraph networks and convolutional attention mechanisms. This method significantly enhances the adaptability and prediction accuracy of traditional knowledge tracing (KT) models. By introducing a two-parameter input mechanism of 2PLM (two-parameter logistic model), HCAKT effectively addresses the challenge of significant differences in skill difficulty within the same learning item, improving both the model's flexibility and interpretability. Furthermore, the inclusion of a hypergraph network, a key tool for parameter balancing, effectively avoids the potential interference of redundant information on model performance and further improves prediction accuracy by optimizing the relationships between parameters. Finally, HCAKT incorporates a convolutional attention mechanism to simulate and capture students' forgetting patterns. This mechanism can deeply analyze the memory decay characteristics of students during the learning process, thus more realistically reflecting their learning status.

[0143] Furthermore, this application provides an interpretable knowledge tracking device based on a hypergraph convolutional attention mechanism, including a prediction unit, for predicting students' skill mastery using an interpretable model based on a hypergraph convolutional attention mechanism.

[0144] Among them, the interpretable model based on the hypergraph convolutional attention mechanism includes: an interpretable module based on 2PLM, a hypergraph convolutional balancing parameter module, and a convolutional attention memory component module;

[0145] The 2PLM-based interpretable module includes an extraction module and a student module. The extraction module independently extracts the practice difficulty parameter and skill difficulty parameter based on the practice attempted by the student and the skills required to solve the practice, and then merges the practice difficulty parameter and skill difficulty parameter to obtain a comprehensive difficulty parameter. The student module calculates the student's potential knowledge state based on the knowledge mastered by the student in a set time period. The 2PLM-based interpretable module also predicts the response probability to the practice based on the comprehensive difficulty parameter and the student's potential knowledge state.

[0146] The hypergraph convolutional balancing parameter module optimizes the input parameters using a neural network model based on the hypergraph architecture to obtain optimized input parameters, and calculates retention parameters, forgetting parameters, and influence parameters based on the optimized input parameters. The input parameters of the neural network model based on the hypergraph architecture include the vector representation of the student's response to the exercise and the student's past skill potential values.

[0147] The convolutional attention memory component module fuses the retention parameters, forgetting parameters, and influencing parameters based on the attention mechanism to obtain the forgetting vector;

[0148] An interpretable model based on hypergraph convolutional attention mechanism predicts students’ skill mastery based on the response probability to practice, students’ past skill potential, and forgetting vector, thus obtaining the skill potential value mastered by students.

[0149] The specific implementation methods of each module of the interpretable knowledge tracking device based on the hypergraph convolutional attention mechanism can be found in the corresponding content of the aforementioned method embodiments, and will not be repeated here.

[0150] It is understood that the same or similar parts in the above embodiments can be referred to each other, and the contents not described in detail in some embodiments can be referred to the same or similar contents in other embodiments.

[0151] It should be noted that in the description of this invention, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance. Furthermore, in the description of this invention, unless otherwise stated, "a plurality of" means at least two.

[0152] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process, and the scope of the preferred embodiments of the invention includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as will be understood by those skilled in the art to which embodiments of the invention pertain.

[0153] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0154] Those skilled in the art will understand that all or part of the steps of the methods implementing the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0155] Furthermore, the functional units in the various embodiments of this invention can be integrated into a single processing module, or each unit can exist physically separately, or two or more units can be integrated into a single module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. The aforementioned storage medium can be a read-only memory, a disk, or an optical disk, etc.

[0156] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0157] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. An interpretable knowledge tracing method based on hypergraph convolutional attention mechanism, characterized in that, include: Predicting students' skill mastery using an interpretable model based on hypergraph convolutional attention mechanism; The interpretable model based on the hypergraph convolutional attention mechanism includes: an interpretable module based on 2PLM, a hypergraph convolutional balancing parameter module, and a convolutional attention memory component module. The 2PLM-based interpretable module includes an extraction module and a student module. The extraction module independently extracts practice difficulty parameters and skill difficulty parameters based on the practice attempted by the student and the skills required to solve the practice, and then merges the practice difficulty parameters and skill difficulty parameters to obtain a comprehensive difficulty parameter. The student module calculates the student's potential knowledge state based on the knowledge the student has acquired over a set time period. The 2PLM-based interpretable module also predicts the response probability to the practice based on the comprehensive difficulty parameter and the student's potential knowledge state. The hypergraph convolutional balancing parameter module optimizes the input parameters using a neural network model based on the hypergraph architecture to obtain optimized input parameters, and calculates retention parameters, forgetting parameters, and influence parameters based on the optimized input parameters; wherein, the input parameters of the neural network model based on the hypergraph architecture include the vector representation of the student's answer to the exercise and the student's past skill potential value; The convolutional attention memory component module fuses the retention parameter, the forgetting parameter, and the influence parameter based on an attention mechanism to obtain a forgetting vector; The interpretable model based on the hypergraph convolutional attention mechanism predicts the student's skill mastery based on the response probability to the exercise, the student's past skill potential value, and the forgetting vector, thereby obtaining the student's skill potential value.

2. The method according to claim 1, characterized in that, The extraction module is based on the first Embedded vectors of each exercise The calculation process for the practice difficulty parameter using a feedforward neural network is as follows: In the formula, and Represents the weight vector. Indicates the number of hidden layers. The last layer of the hidden layer Then it means the first The difficulty parameter corresponding to each exercise. and Represents the bias parameter vector; The extraction module is also based on solving the first Embedding vector of skills required for each exercise The process of obtaining the skill difficulty parameters and calculating them using a feedforward neural network is as follows: In the formula, and This represents the weight vector, the last layer of the hidden layer. Indicates the first The skill difficulty parameters corresponding to each skill and This represents the bias parameter vector.

3. The method according to claim 2, characterized in that, In the student module, students can... Knowledge acquired over a period of time To calculate the students' potential changes in their knowledge status: in, Indicates the first OK vector, Indicates the first Attention weight for each skill; The process for predicting student abilities is as follows: in, This represents the weight vector, where m represents the number of hidden layers. ,and , This represents the bias parameter vector.

4. The method according to claim 3, characterized in that, The 2PLM-based interpretable module uses students in time Solve the exercises ability With two difficulty parameters and The difference between the sums is used to predict the probability of a student's response to an exercise: in, This represents the predicted response probability.

5. The method according to claim 4, characterized in that, The 2PLM-based interpretable module calculates the model loss using cross-entropy loss: in, Represents cross-entropy loss, This represents the probability of the actual response.

6. The method according to claim 5, characterized in that, The optimized past skill potential of students based on the neural network model of the hypergraph architecture. The calculation process is as follows: in, It is a weight vector. For the bias parameter vector; Vector representation of students' responses to exercises and students' past skill potential The optimization process in hypergraph networks is as follows: in, , and It's a hyperparameter. This indicates the number of recursions in the recursive architecture. and This represents the weight vector.

7. The method according to claim 6, characterized in that, The reserved parameters Forgetting parameters and influencing parameters The calculation formula is: in, , , , , and Represents the weight vector. , and This represents the bias parameter vector.

8. The method according to claim 7, characterized in that, The formula for obtaining the forgetting vector by fusing the convolutional attention memory component module is expressed as follows: in, It is the first The vector representing the factor. Reserved parameters from the above modules Forgetting parameters and influencing parameters composition, It refers to the number of attention layers; This indicates the proportion of the input in the attention layer; In addition, a unique attention weight was assigned to the forgetting factor: in, This represents the attention weight.

9. The method according to claim 8, characterized in that, The formula for calculating the potential value of skills mastered by students using the interpretable model based on the hypergraph convolutional attention mechanism is expressed as follows: in, This indicates that a one-dimensional convolutional network is used to capture and preserve parameters. With forgetting parameters The interrelationships within.

10. An interpretable knowledge tracking device based on a hypergraph convolutional attention mechanism, characterized in that, include: The prediction unit is used to predict students' skill mastery using an interpretable model based on a hypergraph convolutional attention mechanism. The interpretable model based on the hypergraph convolutional attention mechanism includes: an interpretable module based on 2PLM, a hypergraph convolutional balancing parameter module, and a convolutional attention memory component module. The 2PLM-based interpretable module includes an extraction module and a student module. The extraction module independently extracts practice difficulty parameters and skill difficulty parameters based on the practice attempted by the student and the skills required to solve the practice, and then merges the practice difficulty parameters and skill difficulty parameters to obtain a comprehensive difficulty parameter. The student module calculates the student's potential knowledge state based on the knowledge the student has acquired over a set time period. The 2PLM-based interpretable module also predicts the response probability to the practice based on the comprehensive difficulty parameter and the student's potential knowledge state. The hypergraph convolutional balancing parameter module optimizes the input parameters using a neural network model based on the hypergraph architecture to obtain optimized input parameters, and calculates retention parameters, forgetting parameters, and influence parameters based on the optimized input parameters; wherein, the input parameters of the neural network model based on the hypergraph architecture include the vector representation of the student's answer to the exercise and the student's past skill potential value; The convolutional attention memory component module fuses the retention parameter, the forgetting parameter, and the influence parameter based on an attention mechanism to obtain a forgetting vector; The interpretable model based on the hypergraph convolutional attention mechanism predicts the student's skill mastery based on the response probability to the exercise, the student's past skill potential value, and the forgetting vector, thereby obtaining the student's skill potential value.

Images