Knowledge tracking method and system based on hypergraph neural network and emotion perception

Abstract

The present invention discloses a knowledge tracking method and system based on a hypergraph neural network and emotion perception, belonging to the field of knowledge tracking. The method of the present invention first obtains an exercise data set composed of exercise data of a student account, then converts the relationship between exercises and knowledge points in the exercise data into a matrix representation, constructs an exercise-knowledge point association matrix, and constructs an exercise-knowledge point hypergraph with the exercise data of the student account based on the exercise-knowledge point association matrix, constructs a knowledge tracking model including a hypergraph neural network, an emotion perception module, and a prediction module and trains it, inputs the exercise data to be tracked into the trained knowledge tracking model, and outputs the predicted value of the student's answer to the next exercise. The method of the present invention obtains the knowledge mastery of students in different emotional states through the combination of hypergraph structure and emotional factors, improves the prediction performance of knowledge tracking, and is of great significance to the precise and personalized development of intelligent guidance systems.

Description

Technical Field

[0001] The present invention belongs to the field of knowledge tracing, and in particular, relates to a knowledge tracing method and system based on a hypergraph neural network and emotion perception. Background Art

[0002] The intelligent guidance system integrates and analyzes a large amount of students' learning behavior data through artificial intelligence technology, provides students with personalized learning suggestions, and helps them learn efficiently according to their own knowledge level. Knowledge tracing (KT) is one of the core tasks of the intelligent guidance system. By mining the learner's historical practice records, the learner's knowledge mastery status is obtained and his future learning performance is predicted. Through knowledge tracing, a knowledge mastery architecture is built for learners, helping the intelligent guidance system to provide learners with personalized learning paths. Although the importance of knowledge tracing in the field of smart education has been widely recognized, its research still faces many challenges. Improving the prediction accuracy can not only better meet the learners' personalized needs, but also significantly improve learning efficiency and learning experience. With the accumulation of large-scale online education data, accurate modeling and tracking of students' knowledge status not only helps to improve learning effects, but also provides strong support for the personalized and intelligent development of the education system.

[0003] The knowledge tracing method has undergone many evolutions since its birth. In the early classic model, BKT predicted students' knowledge mastery through hidden variables and probability update mechanisms. However, the BKT model relies on fixed prior parameters and is difficult to flexibly cope with complex and diverse learning behavior patterns. In recent years, deep learning has become a key technology in knowledge tracing research with its powerful feature extraction capabilities and excellent prediction performance, providing important support for the precision and intelligent development of online education. With the maturity of graph structure data processing technology, graph neural networks (GNNs) have received increasing attention. As a deep learning method that can efficiently process graph structure data, GNNs perform well in tasks such as social relationship networks and recommendation tasks, showing excellent expression capabilities. Since knowledge tracing involves complex interactive relationships between students, questions and knowledge points, it is very suitable for modeling using graph neural networks. In this task, students, questions and knowledge points are used as nodes, and interactive behaviors are used as edges. GNNs can effectively capture the dependencies and dynamics between these entities and their relationships. However, existing graph-based knowledge tracking models still have the following limitations: 1) Models such as GKT only model the relationship graph between knowledge points, ignoring the association between exercises and knowledge points; 2) Models such as GIKT and PEBG use bipartite graphs, focusing on the relationship between exercises and knowledge points, ignoring the potential connections between exercises.

[0004] Pedro and Pardos' research pointed out that students' learning behavior and knowledge status are significantly affected by emotional factors (such as concentration, confusion, frustration, etc.). Under different emotional states, students' performance on the same question may be significantly different. However, existing deep knowledge tracking models often ignore the influence of emotional factors, resulting in limited prediction accuracy. Under the same knowledge point, different exercises are usually affected by similar emotional states, and graph-based models can effectively model the complex interactive relationships between students, exercises, and knowledge points. Through graph neural networks, students' emotional states can be propagated in the association graph between exercises and knowledge points, realizing emotional perception of knowledge status. However, most existing graph neural network methods focus on modeling simple structures and fail to fully explore high-order relationships in the learning process. Therefore, how to design a graph neural network model that can capture high-order interactive relationships based on process data and emotional information has become an important research task. This exploration is not only expected to improve the prediction accuracy of knowledge tracking, but also provide new ideas for the precise, personalized and systematic development of intelligent guidance systems. Summary of the invention

[0005] The purpose of the present invention is to solve the problems existing in the prior art and provide a knowledge tracking method and system based on hypergraph neural network and emotion perception.

[0006] The present invention captures and models students' emotional states by constructing an exercise-knowledge point hypergraph and combining it with an emotion perception module, thus providing support for personalized knowledge tracking. The exercise-knowledge point hypergraph is used to model the many-to-many relationship between exercises and knowledge points. Exercises are regarded as nodes in the hypergraph, and knowledge points are used as hyperedges to connect related exercises, thereby effectively representing the structured association between exercises and knowledge points. The emotion perception module generates a personalized emotion perception vector by integrating student characteristics, exercise difficulty characteristics, and number of attempts characteristics, providing emotional information support for the subsequent update of knowledge status.

[0007] The present invention designs a hypergraph convolution mechanism based on emotion perception, and adaptively adjusts the weight of the hyperedge so that the interaction information between nodes can reflect the individual differences of students. During feature aggregation, the knowledge tracking model not only considers the topological structure of the hypergraph, but also dynamically adjusts the influence of the node in combination with the current emotional state, so that the updated node representation can simultaneously contain knowledge structure and emotional dynamic information, significantly improving the generalization ability and prediction accuracy of the knowledge tracking model. The prediction module uses the updated emotion-aware exercise representation and the student's learning behavior state, and uses the gated recurrent unit (GRU) to dynamically update the student's knowledge state to capture the time dependency and emotional changes in the learning process. The knowledge tracking model optimization adopts a joint loss function, including prediction loss and emotion perception loss. The prediction loss is used to measure the error between the prediction result of the knowledge tracking model and the actual label to ensure the accuracy of knowledge tracking. The emotion perception loss is divided into supervised learning loss and unsupervised contrast learning loss, which are used to improve the accuracy and generalization ability of emotion perception respectively. By adjusting the weight parameters in the loss function, the prediction performance and the robustness of emotion modeling are balanced, and finally a higher personalized learning effect is achieved.

[0008] In order to achieve the above-mentioned invention object, the present invention specifically adopts the following technical solutions:

[0009] In a first aspect, the present invention provides a knowledge tracking method based on a hypergraph neural network and emotion perception, which comprises the following steps:

[0010] S1: Obtain an exercise dataset with or without real emotional state labels from an education platform or intelligent tutoring system. Both exercise datasets consist of exercise data from student accounts. Each exercise data contains exercises, real answer results, knowledge points included in the exercises, and the number of attempts to complete the exercises.

[0011] S2: Convert the relationship between exercises and knowledge points in the practice data into a matrix representation and construct an exercise-knowledge point association matrix;

[0012] S3: Based on the exercise-knowledge point association matrix, the exercise data of the student account is transformed into an exercise-knowledge point hypergraph, where each node in the exercise-knowledge point hypergraph is each exercise in the exercise data, and the knowledge point hyperedge between two or more nodes in the exercise-knowledge point hypergraph is each knowledge point in the exercise data;

[0013] S4: Construct a knowledge tracking model that includes a hypergraph neural network, a sentiment perception module, and a prediction module. Take an exercise dataset and an exercise-knowledge point hypergraph as input to train the knowledge tracking model.

[0014] S5: Input the exercise data to be tracked in the student account into the trained knowledge tracking model, output the predicted value of the student's answer to the next exercise, and push it to the user interface of the education platform or intelligent tutoring system in a preset manner.

[0015] Based on the above scheme, each step can be implemented in the following preferred specific manner.

[0016] As a preferred embodiment of the above-mentioned first aspect, in step S2, each element in the exercise-knowledge point association matrix is ​​used to indicate whether the exercise is associated with the knowledge point. When the element in the exercise-knowledge point association matrix is ​​1, it indicates that the exercise is associated with the knowledge point. Otherwise, when it is 0, it indicates that the exercise is not associated with the knowledge point.

[0017] As a preferred embodiment of the first aspect, in step S4, in the knowledge tracking model, the specific training process is as follows:

[0018] S41: Input the constructed exercise-knowledge point hypergraph into the hypergraph neural network and output the exercise embedding representation;

[0019] S42: The student account ID, exercise ID, and number of attempts to complete the exercise in the exercise data set are each embedded in a low dimension through an embedding layer, and a unique identification vector of the student account ID, a unique identification vector of the exercise ID, and a unique identification vector of the number of attempts are obtained accordingly. The unique identification vector of the student account ID, the unique identification vector of the exercise ID, and the unique identification vector of the number of attempts are input into the emotion perception module together, and the emotion knowledge state representation is output;

[0020] S43: The exercise embedding representation is fused with the emotional knowledge state representation through matrix multiplication to obtain the emotional enhanced exercise embedding representation;

[0021] S44: Input the sentiment-enhanced exercise embedding representation into the prediction module, firstly process the sentiment-enhanced exercise embedding representation by the gated recurrent unit to obtain the final hidden state, and then pass the final hidden state through a linear layer to map the sentiment-enhanced exercise embedding representation into the predicted value of the student's answer result for the next exercise;

[0022] S45: The cross entropy loss between the predicted value of the student's answer to the next exercise and the actual answer result is used as the prediction loss. When the exercise data set input to the knowledge tracking model is an exercise data set with a real emotional state label, the mean square error loss between the emotion perception embedding representation output by the emotion perception module and the real emotional state label is calculated as the emotion perception loss; when the exercise data set input to the knowledge tracking model is an exercise data set without a real emotional state label, a contrastive learning strategy is adopted to construct a contrastive loss with the cosine similarity of the positive and negative sample pairs and use it as the emotion perception loss; wherein the real emotional state label is confusion, concentration, boredom or frustration;

[0023] S46: Taking the weighted sum of the emotion perception loss and the prediction loss as the total loss, updating the knowledge tracking model parameters based on minimizing the total loss, and continuously iterating the training until the knowledge tracking model converges to obtain a trained knowledge tracking model.

[0024] As a preferred embodiment of the first aspect, in step S42, in the emotion perception module, the specific processing flow is as follows:

[0025] S421: multiply the unique identification vector of the student account ID by the trainable first weight matrix and pass it through a sigmoid activation function to output the student perception embedding vector;

[0026] S422: multiplying the unique identification vector of the number of attempts by the trainable second weight matrix and performing a sigmoid activation function to output an embedding vector of the number of attempts;

[0027] S423: multiplying the unique identification vector of the exercise ID by the trainable third weight matrix and then passing through a sigmoid activation function to output a knowledge point difficulty embedding vector;

[0028] S424: concatenating the student perception embedding vector, the attempt number embedding vector, and the knowledge point difficulty embedding vector as a fused embedding vector, and passing the fused embedding vector through a linear layer and a sigmoid activation function to output a sentiment perception embedding representation;

[0029] S425: Perform an element-by-element multiplication operation on the joint representation vector of the exercise and the answer result and the emotional perception embedding representation to obtain the emotional knowledge state representation.

[0030] As a preferred embodiment of the above-mentioned first aspect, in step S425, the specific method for generating the joint representation vector of the exercises and answer results is as follows: the exercise data in the exercise data set is divided into multiple groups of exercise data, the exercises and answer results in each group of exercise data are encoded, and the encoding vectors corresponding to each group of exercise data are obtained, and the encoding vectors of each group of exercise data constitute the joint representation vector of the exercises and answer results; wherein, the encoding vector of the g-th group is a 2F-dimensional vector, F represents the total number of exercises, and is encoded with 0 and 1 according to the actual answer results of the students: if the student answers the exercises correctly, the element value at the index of the exercise (ID-1) in the encoding vector of the g-th group is set to 1, and the remaining positions are set to 0; if the student answers the exercises incorrectly, the element value at the index of the exercise (ID+F-1) in the encoding vector of the g-th group is set to 1, and the remaining positions are set to 0.

[0031] As a preferred embodiment of the first aspect above, in step S45, the specific process of adopting the contrastive learning strategy and constructing the contrastive loss is as follows: first, determine whether the actual answer results of the students completing the two exercises are the same to construct a positive and negative sample pair: if the actual answer results of the students completing the two exercises are the same, then a positive sample pair is formed by their corresponding emotion-aware embedding representations; if the actual answer results of the students completing the two exercises are different, then a negative sample pair is formed by their corresponding emotion-aware embedding representations; then calculate the cosine similarity of the positive sample pair and the cosine similarity of the negative sample pair, and finally construct the contrastive loss based on the cosine similarity of the positive sample pair and the cosine similarity of the negative sample pair.

[0032] In a second aspect, the present invention provides a knowledge tracking system based on a hypergraph neural network and emotion perception, comprising:

[0033] A data acquisition module is used to obtain an exercise data set with a real emotional state label or an exercise data set without a real emotional state label from an education platform or an intelligent tutoring system. Both exercise data sets are composed of exercise data of a student account. Each exercise data includes an exercise, a real answer result, knowledge points included in the exercise, and the number of attempts to complete the exercise. The real emotional state label is confusion, concentration, boredom, or frustration.

[0034] The correlation matrix acquisition module is used to convert the relationship between exercises and knowledge points in the exercise data into a matrix representation and construct an exercise-knowledge point correlation matrix;

[0035] A hypergraph acquisition module is used to transform the exercise data of the student account into an exercise-knowledge point hypergraph according to the exercise-knowledge point association matrix, wherein each node in the exercise-knowledge point hypergraph is each exercise in the exercise data, and the knowledge point hyperedge between two or more nodes in the exercise-knowledge point hypergraph is each knowledge point in the exercise data;

[0036] A model training module is used to build a knowledge tracking model including a hypergraph neural network, a sentiment perception module, and a prediction module. An exercise data set and an exercise-knowledge point hypergraph are used as input to train the knowledge tracking model.

[0037] The result acquisition module is used to input the exercise data of the student account to be tracked into the trained knowledge tracking model, output the predicted value of the student's answer to the next exercise, and push it to the user interface of the education platform or intelligent tutoring system in a preset manner;

[0038] The mastery probability acquisition module is used to pass the final hidden state output by the knowledge tracking model through an additional linear layer, map the final hidden state to the knowledge point dimension, and use the Sigmoid activation function to normalize the final hidden state after the linear layer to the mastery probability of each knowledge point, which is used to indicate the student's mastery of each knowledge point;

[0039] The learning suggestion generation module is used to compare the mastery probability output by the mastery probability acquisition module with a preset threshold, and regard the knowledge points with a mastery probability greater than or equal to the threshold as mastered knowledge points, and regard the knowledge points with a mastery probability less than the threshold as weak knowledge points. For the mastered knowledge points, the user interface displays suggestions for increasing the difficulty or expanding the knowledge, and for the weak knowledge points, the user interface displays suggestions for reviewing or performing intensive exercises.

[0040] In a third aspect, the present invention provides a computer program product, including a computer program / instruction, which, when executed by a processor, can implement the knowledge tracking method based on hypergraph neural network and emotion perception as described in any one of the solutions in the first aspect above.

[0041] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the knowledge tracking method based on a hypergraph neural network and emotion perception as described in any of the schemes in the first aspect above is implemented.

[0042] In a fifth aspect, the present invention provides a computer electronic device comprising a memory and a processor;

[0043] The memory is used to store computer programs;

[0044] The processor is used to implement the knowledge tracking method based on hypergraph neural network and emotion perception as described in any scheme of the first aspect above when executing the computer program.

[0045] Compared with the prior art, the present invention has the following beneficial effects:

[0046] By introducing an emotion perception module and integrating students' emotional states, exercise difficulty, and number of attempts, the modeling of personalized learning behaviors is achieved, which significantly improves the prediction accuracy of knowledge tracking. In addition, the present invention constructs an exercise-knowledge point hypergraph based on a hypergraph neural network, which can capture high-order dependencies between exercises and knowledge points, and has a stronger ability to express structured information compared to traditional graph structures. Through the emotion-weighted hypergraph convolution mechanism, the information propagation process is adaptively adjusted, so that the knowledge tracking model can optimize the knowledge state update in combination with the students' real-time emotional state, thereby improving the generalization ability and adaptability of the knowledge tracking model. In summary, the present invention effectively combines emotional factors with hypergraph structures, significantly improving the performance of the knowledge tracking model and the effect of personalized learning. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 is a flow chart of the steps of the method of the present invention;

[0048] Figure 2 Schematic diagram of the training process of the knowledge tracking model in the method of the present invention;

[0049] Figure 3 Schematic diagram of prediction results of this embodiment;

[0050] Figure 4 It is a system block diagram of the present invention. DETAILED DESCRIPTION

[0051] In order to make the above-mentioned purpose, features and advantages of the present invention more obvious and easy to understand, the specific implementation mode of the present invention is described in detail below in conjunction with the accompanying drawings. In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below. The technical features in each embodiment of the present invention can be combined accordingly without conflicting with each other.

[0052] In the description of the present invention, it should be understood that the terms "first" and "second" are only used for the purpose of distinguishing descriptions, and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features.

[0053] like Figure 1 As shown, in a preferred implementation of the present invention, the above-mentioned knowledge tracking method based on hypergraph neural network and emotion perception includes the following steps S1 to S5. The specific implementation process is described in detail below.

[0054] The method of the present invention first constructs an exercise-knowledge point hypergraph based on the association relationship between exercises and knowledge points, and captures the complex association relationship between questions and knowledge points through the hypergraph convolution operation in the hypergraph neural network. The construction process of the exercise-knowledge point hypergraph is divided into the following three stages: data collection, exercise-knowledge point association matrix construction, and exercise-knowledge point hypergraph structure generation.

[0055] 1) Data Collection

[0056] S1: Obtain an exercise dataset with or without real emotional state labels from an education platform or intelligent tutoring system. Both exercise datasets are composed of exercise data from student accounts. Each exercise data contains exercises, real answer results, knowledge points included in the exercises, and the number of attempts to complete the exercises. The real emotional state labels are confusion, concentration, boredom, or frustration.

[0057] It should be noted that in step S1, the students' practice data are extracted from the data set obtained from the education platform or intelligent tutoring system to construct the exercise-knowledge point association matrix. Each exercise data contains exercises, actual answer results, knowledge points contained in the exercises, and the number of attempts to complete the exercises. Among them, the knowledge point represents one or more knowledge points associated with each exercise; the student account ID is used to uniquely identify each student, reflecting their personalized learning characteristics; the exercise ID can uniquely identify the specific exercises practiced by the students and is associated with the knowledge points to reflect the difficulty information of the exercises; the number of attempts to complete the exercises records the students' multiple attempts to complete each exercise, helping the knowledge tracking model to quantify the students' proficiency in knowledge points. The real emotional state labels of students during the practice process are confused, concentrated, bored, and frustrated, which are used as supervised learning signals for the emotion perception module.

[0058] 2) Exercise-knowledge point association matrix construction

[0059] S2: Convert the relationship between exercises and knowledge points in the practice data into a matrix representation and construct an exercise-knowledge point association matrix .

[0060] In step S2 of the present invention, the first Line Elements of a column Indicates Is the exercise related to There is a connection between the knowledge points. Indicates Exercises and There is a connection between the knowledge points. Indicates Exercises and There is no correlation between the knowledge points. is the number of exercises, is the number of knowledge points.

[0061] 3) Exercises - Generation of knowledge point hypergraph structure

[0062] S3: Based on the exercise-knowledge point association matrix, the exercise data of the student account is constructed into an exercise-knowledge point hypergraph , each node in the exercise-knowledge point hypergraph is each exercise in the exercise data, and the knowledge point hyperedge between two or more nodes in the exercise-knowledge point hypergraph is each knowledge point in the exercise data; represents the node set consisting of the nodes in the exercise-knowledge point hypergraph, Represents the hyperedge set composed of knowledge point hyperedges in the exercise-knowledge point hypergraph.

[0063] S4: Construct a knowledge tracking model that includes a hypergraph neural network, a sentiment perception module, and a prediction module. Take an exercise dataset and an exercise-knowledge point hypergraph as input to train the knowledge tracking model.

[0064] It should be noted that in step S4, the present invention designs a knowledge tracking model based on the exercise-knowledge point hypergraph. In the knowledge tracking model, the embedded representation of each node (exercise) is updated through multiple convolutional layers in the hypergraph neural network (Hypergraph Neural Nerworks, HGNN). This process aggregates the knowledge point features related to each exercise, and combines the connection relationship between nodes in the exercise-knowledge point hypergraph to gradually update the representation of the exercise node, capturing the complex many-to-many relationship between the exercise and the knowledge point. The feature of each exercise node not only depends on itself, but also aggregates with the associated knowledge point features. The exercise-knowledge point association matrix in the exercise-knowledge point hypergraph links the relationship between the exercise and the knowledge point, promoting information dissemination and feature sharing. The core goal of the construction of the emotion perception module is to enhance the understanding of the student behavior of the knowledge tracking model by capturing the emotional changes of the students in the learning process. The emotion perception module combines information from multiple dimensions such as the personalized characteristics of the students, the difficulty of the knowledge points and the number of attempts, helping the knowledge tracking model to more accurately predict the changes in the students' emotional state, thereby improving the ability to predict learning behavior. The following is a detailed introduction to the specific process of the knowledge tracking model during the training process.

[0065] In the knowledge tracking model, Figure 2 As shown, the specific training process is as follows:

[0066] S41: Input the constructed exercise-knowledge point hypergraph into the hypergraph neural network and output the exercise embedding representation.

[0067] It should be noted that in the present invention, the implementation method of the hypergraph neural network belongs to the prior art. The processing process of layer hypergraph convolution is as follows:

[0068]

[0069] in, Indicates Embedded representation of exercises generated by the layer; represents a learnable weight matrix used to weight the features of adjacent nodes; superscript Represents matrix transpose; Indicates Embedded representation of exercises generated by the layer; It is The learning parameters of the layer are used to linearly transform the node features; It is a nonlinear activation function, which is used to improve the nonlinear expression ability of the knowledge tracking model; represents the node degree matrix, represents the knowledge point hyperedge degree matrix, the node degree matrix is ​​a diagonal matrix, in which each diagonal element represents the degree of the exercise node, that is, the number of knowledge points associated with the exercise; the knowledge point hyperedge degree matrix is ​​a diagonal matrix, in which each diagonal element represents the degree of the knowledge point node, that is, the number of exercises associated with the knowledge point. These two degree matrices are used to standardize node features, ensure the balance of the feature aggregation process, and avoid the influence of node degree differences on feature updates.

[0070] In this embodiment, in the hypergraph-based knowledge tracking task, a hypergraph neural network (HGNN) is designed to model students' knowledge mastery. HGNN can handle complex learning data. Unlike traditional graph convolution, in each layer of hypergraph convolution, the embedded representation of the exercise (node) is aggregated and updated through the hyperedges related to it and the information of other nodes connected by the hyperedges. Hypergraph convolution can more naturally handle the complex associations between questions and knowledge points, and is particularly suitable for many-to-many association relationship modeling. Through multi-layer hypergraph convolution operations, the knowledge tracking model can refine the students' mastery of knowledge points layer by layer.

[0071] It should be noted that, in the present invention, the initial embedding representation of the exercise is generated through an embedding layer, and the embedding representation of each exercise is used as the input of the hypergraph convolution to capture the initial features of the nodes.

[0072] S42: The student account ID, exercise ID, and number of attempts to complete the exercise in the exercise data set are each embedded in a low-dimensional embedding layer to obtain a unique identification vector corresponding to the student account ID , unique identification vector of exercise ID and a unique identification vector of the number of attempts , the unique identification vector of the student account ID, the unique identification vector of the exercise ID, and the unique identification vector of the number of attempts are input into the emotion perception module, and the emotion knowledge state representation is output. Among them, Indicates the number of student account IDs; Indicates the number of exercise IDs; Indicates the number of attempts.

[0073] It should be noted that, in the present invention, by combining student characteristics, knowledge point difficulty and number of attempts with emotional state, the emotional perception module can capture emotional fluctuations in the learning process, thereby providing a more accurate prediction of the student's knowledge mastery status.

[0074] In this embodiment, the specific processing flow of the emotion perception module is as follows:

[0075] S421: Combine the unique identification vector of the student account ID with the trainable first weight matrix After multiplication, the sigmoid activation function is used to output the student perception embedding vector :

[0076]

[0077] in, Give students a sense of the dimensionality of the embedding vector.

[0078] In the emotion perception module, student characteristics are first introduced as a key factor. When facing the same exercises, different students may have different feelings and understandings of the exercises even if their knowledge levels are similar. Therefore, it is necessary to consider the personalized emotional response of each student in the problem-solving process. To capture this difference, the student perception embedding vector can be obtained by calculating the unique identification vector and weight matrix of the student account ID, which can better reflect the differences in emotional responses of different students when facing the same exercises.

[0079] S422: Combine the unique identification vector of the number of attempts with the trainable second weight matrix After multiplication, the sigmoid activation function is used to output the number of attempts embedded in the vector :

[0080]

[0081] in, Represents the dimension of the embedding vector of the number of attempts.

[0082] S423: Combine the unique identification vector of the exercise ID with the trainable third weight matrix After multiplication, the sigmoid activation function is used to output the knowledge point difficulty embedding vector :

[0083]

[0084] in, Represents the dimension of the embedding vector of the knowledge point difficulty.

[0085] It should be noted that in the present invention, students' multiple attempts at an exercise not only reflect their learning behavior, but also reveal their emotional state. Multiple attempts often mean that students have encountered difficulties in understanding the exercises or mastering the relevant knowledge points, which may cause confusion and emotional frustration. On the contrary, repeated successful attempts may enhance students' sense of accomplishment and concentration. By embedding the number of attempts as a feature into the knowledge tracking model, the knowledge tracking model can more accurately capture students' emotional fluctuations in multiple attempts and further improve the ability to predict learning behavior. The difficulty of the learning task also significantly affects the students' emotional state. Faced with more challenging tasks, students may experience more pressure or confusion, while less challenging tasks may enhance their confidence. In order to effectively capture this emotional fluctuation, the student perception embedding vector, the number of attempts embedding vector, and the knowledge point difficulty embedding vector are introduced in the emotional perception module, thereby improving the knowledge tracking model's ability to predict emotional states.

[0086] S424: The student perception embedding vector, the number of attempts embedding vector, and the knowledge point difficulty embedding vector are concatenated as a fused embedding vector, and the fused embedding vector is passed through a linear layer and then a sigmoid activation function is used to output the emotional perception embedding representation. , used to represent the students’ emotional state in the current learning task:

[0087]

[0088] in, and are the weight matrix and bias of the linear layer respectively; Represents a concatenation operation.

[0089] It should be noted that in the present invention, the emotional state of students is affected by multiple factors. After obtaining the student perception embedding vector, the number of attempts embedding vector, and the knowledge point difficulty embedding vector, these information together constitute the emotional perception embedding representation, which helps the knowledge tracking model to more comprehensively capture the emotional and behavioral changes of students in the learning process, thereby improving the ability to predict students' learning behavior.

[0090] S425: The joint representation vector of the exercise and the answer Perform element-wise product operation with the emotion perception embedding representation to obtain the emotion knowledge state representation :

[0091]

[0092] in, Represents an element-wise product operation.

[0093] The key to this fusion method is to closely associate emotional information with exercise features, thereby generating a feature representation that contains dual information. Through the element-by-element product operation, the two are combined to generate an emotional knowledge state representation, which not only retains the characteristics of the exercise itself, but also introduces the influence of emotional factors, so that the representation can more accurately reflect the student's contextual learning state on the exercise. The emotional knowledge state representation not only provides more fine-grained input features for the knowledge tracking model, but also provides additional information about the student's emotional changes for the subsequent prediction module, helping the knowledge tracking model to more flexibly and accurately model its dynamic learning process when predicting the student's knowledge state.

[0094] It should be noted that, in the present invention, the specific method for generating the joint representation vector of the exercises and the answer results is as follows: the exercise data in the exercise data set is divided into multiple groups of exercise data, the exercises and the answer results in each group of exercise data are encoded, and the encoding vectors corresponding to each group of exercise data are obtained, and the encoding vectors of each group of exercise data constitute the joint representation vector of the exercises and the answer results. The encoding vector of the group one dimensional vector, used to encode the known exercise ID and the student's answer result, Indicates the total number of questions. The actual answers of the students are coded with 0 and 1, 0 means the student answered the question incorrectly, and 1 means the student answered the question correctly. The encoding vector of the group Index to Exercises The element value at the position is set to 1, and the other positions are set to 0; if the student answers the question incorrectly, The encoding vector of the group Index to Exercises The element value at the position is set to 1, and the other positions are set to 0. By encoding the exercise ID and the actual answer result, the exercise information and the problem-solving status are jointly represented, which provides a basis for the integration of emotion and cognitive state.

[0095] S43: The exercise embedding representation is combined with the emotional knowledge state representation through matrix multiplication (matmul) to obtain the emotional enhanced exercise embedding representation :

[0096]

[0097] S44: The sentiment-enhanced exercise embedding representation is input into the prediction module. The sentiment-enhanced exercise embedding representation is first processed by a gated recurrent unit (GRU) to obtain a final hidden state. The final hidden state is then passed through a linear layer to map the sentiment-enhanced exercise embedding representation to a predicted value of the student's answer to the next exercise.

[0098] It should be noted that, in the present invention, the emotion-enhanced exercise embedding representation will be used as the input of the gated recurrent unit, thereby realizing the dynamic tracking and prediction of the student's learning process. The implementation method of the gated recurrent unit belongs to the prior art. The processing of a time step is as follows:

[0099]

[0100]

[0101]

[0102]

[0103] in, Reset gate. represents the update gate; , , , , , is the corresponding weight; , and is the bias term; For the time step input; For the The hidden state output at each time step; represents the candidate hidden state; is the hyperbolic tangent function; For the The hidden state at time step.

[0104] In the knowledge tracking task, GRU gradually tracks the changes in the student's knowledge state and generates a hidden state at each time step to represent the student's knowledge mastery of the question. Specifically, the reset gate determines the impact of the hidden state of the current time step input and the previous time step output on the candidate hidden state, helping the knowledge tracking model focus on the changes in the current knowledge state; the update gate determines how to combine the hidden state output of the previous time step and the candidate hidden state to generate the hidden state of the current time step, so that the knowledge tracking model can strike a balance between maintaining long-term information and updating new information; the hidden state of the current time step is applied to the candidate hidden state and the hidden state of the previous time step through the proportion controlled by the update gate, dynamically retaining past information or introducing new knowledge information.

[0105] Finally, the final hidden state output by the gated recurrent unit is mapped to the student's prediction of the answer to the next exercise through a linear transformation. :

[0106]

[0107] in, represents the final hidden state; denote the weights and biases of the linear layer respectively.

[0108] In this way, the prediction module comprehensively models students’ learning behaviors based on the combination of their emotional states, exercise characteristics, and historical performance, thereby improving the accuracy of predictions.

[0109] S45: The cross entropy loss between the predicted value of the student's answer to the next exercise and the actual answer result is used as the prediction loss to measure the gap between the accuracy of the knowledge tracking model's prediction and the actual label. When the exercise data set input to the knowledge tracking model is an exercise data set with real emotional state labels, the mean square error loss between the emotional perception embedding representation output by the emotional perception module and the real emotional state label is calculated as the emotional perception loss; when the exercise data set input to the knowledge tracking model is an exercise data set without real emotional state labels, a contrastive learning strategy is adopted to construct a contrastive loss based on the cosine similarity of positive and negative sample pairs and use it as the emotional perception loss.

[0110] It should be noted that, in the present invention, the prediction loss The function form is as follows:

[0111]

[0112] in, is the number of training data in the training dataset input to the knowledge tracing model; Indicates The actual answer results of the exercises; Represents the predicted value of the answer result output by the knowledge tracking model.

[0113] It should be noted that, in the present invention, depending on whether the training data set input to the knowledge tracking model contains the real emotional state label, the calculation of the emotion perception loss can be divided into two cases: supervised learning and unsupervised contrastive learning.

[0114] 1) Supervised Learning

[0115] When the training dataset with the real emotional state labels is input into the knowledge tracking model, the student's emotional state is labeled as a continuous value between [0,1]. The mean square error (MSE) loss can be used to measure the gap between the emotional perception embedding representation predicted by the emotion perception module and the real emotional state label. Loss as an emotion perception:

[0116]

[0117] in, is the number of training data in the training dataset with true emotional state labels; Indicates The true emotional state labels of the training data; Indicates the emotion perception module Emotion-aware embedding representations predicted from training data.

[0118] 2) Unsupervised Contrastive Learning

[0119] When the exercise data set without the real emotional state label is input into the knowledge tracking model, we first determine whether the real answer results of the students completing the two exercises are the same to construct positive and negative sample pairs: if the real answer results of the students completing the two exercises are the same, then the corresponding emotionally aware embedding representations constitute a positive sample pair; if the real answer results of the students completing the two exercises are different, then the corresponding emotionally aware embedding representations constitute a negative sample pair. Under this setting, the cosine similarity of the positive sample pair and the cosine similarity of the negative sample pair are then calculated:

[0120]

[0121]

[0122] in, Both represent the emotion-aware embedding representation predicted by the emotion-aware module; Indicates the calculation of positive sample pairs The cosine similarity between ; represents the L2 norm; Indicates the calculation of negative sample pairs The cosine similarity between .

[0123] Construct contrast loss based on the cosine similarity of positive and negative samples And as the sentiment perception loss:

[0124]

[0125] in, The temperature parameter of the contrast loss is used to adjust the distribution of the sample; Represents the set of all negative sample pairs; Represents the set of all positive sample pairs.

[0126] S46: Taking the weighted sum of the emotion perception loss and the prediction loss as the total loss, updating the knowledge tracking model parameters based on minimizing the total loss, and continuously iterating the training until the knowledge tracking model converges to obtain a trained knowledge tracking model.

[0127] It should be noted that, in the present invention, the form of the total loss function is as follows:

[0128]

[0129] in, Represents the emotion perception loss, which is used to guide the learning of the knowledge tracking model and capture the emotional changes of students during the learning process. It can be the mean square error loss of supervised learning or the contrast loss of unsupervised learning; is a weight parameter used to adjust the relative importance of prediction loss and sentiment perception loss in the total loss.

[0130] S5: Input the exercise data to be tracked in the student account into the trained knowledge tracking model, output the predicted value of the student's answer to the next exercise, and push it to the user interface of the education platform or intelligent tutoring system in a preset manner.

[0131] It should be noted that in step S5, the learner can log in to his account to view the prediction results displayed in the user interface of the education platform or the intelligent tutoring system. The recommendation mode in this embodiment can be to graphically display the push results in a pre-set functional interface module, or to push them in the form of system messages, etc.

[0132] It should be noted that, in the present invention, the exercise data to be tracked for knowledge may be labeled with a real emotional state or may not be labeled with a real emotional state, without limitation.

[0133] The present invention will now use a specific example to demonstrate the application effect of a knowledge tracking method based on a hypergraph neural network and emotion perception described in S1 to S5 in the above embodiments on a specific data set, so as to facilitate understanding of the essence of the present invention.

[0134] Example

[0135] The specific implementation process of a knowledge tracking method based on hypergraph neural network and emotion perception adopted in this embodiment is as described above and will not be repeated here. To verify the effectiveness of the present invention, this embodiment is experimentally evaluated on the widely used online education data sets ASSIST2009, ASSIST2012 and ASSIST2017 to evaluate the prediction performance of the knowledge tracking model. These data sets contain a large number of interactive records of students at different times, problems and skills (knowledge points), which can comprehensively reflect the students' knowledge status and behavior patterns, and are suitable for testing knowledge tracking tasks. In this embodiment, the performance of the knowledge tracking model is measured by two indicators: accuracy (ACC) and area under the receiver operating characteristic curve (AUC). Among them, ACC represents the accuracy of the knowledge tracking model's prediction of the correctness of the student's answer, that is, the proportion of correct answers predicted by the knowledge tracking model. AUC represents the discrimination ability of the knowledge tracking model prediction, which reflects the performance of the knowledge tracking model under different thresholds by calculating the discrimination effect of the predicted positive and negative samples, and more comprehensively evaluates the performance of the knowledge tracking model. The closer the AUC is to 1, the better the knowledge tracking model is at distinguishing correct and incorrect answers, and the higher its discriminative power.

[0136] Table 1 shows the prediction results of the present invention on the above three data sets, including ACC and AUC scores. The test results show that the performance of this embodiment on different data sets is stable, and relatively high ACC and AUC values ​​have been achieved, proving the effectiveness and reliability of the knowledge tracking model of the present invention in knowledge tracking tasks. This also shows the advantages of this embodiment in capturing students' knowledge status and its dynamic changes, and provides reliable support for personalized predictions in educational scenarios. In Table 1, the implementation methods of the DKT model, DKVMN model, SAKT model, AKT model, GKT model and DGEKT model all belong to the prior art and will not be repeated here. OOM in Table 1 refers to the situation in which, during the reproduction of the existing model, the memory required for program execution exceeds the available memory resources, resulting in the inability of the program to continue execution.

[0137] Table 1. Comparison results of different knowledge tracking models

[0138]

[0139] In order to clearly display the results, this embodiment randomly selects four skills (knowledge points) from the skills answered by the students, with IDs of 34, 39, 40, and 41, corresponding to multiplication, subtraction, addition, and division. Figure 3 The horizontal axis in the middle represents the exercise ID in the test set, and the vertical axis represents the knowledge point ID. The color depth in the heat map represents the predicted probability of the student's correct use of the corresponding skill in the next step. The darker the color, the higher the student's mastery of the skill.

[0140] from Figure 3 Some key information can be observed in the data: First, as the number of exercises for a specific concept increases, the mastery probability predicted by the knowledge tracking model gradually increases, showing that the learner gradually masters the knowledge after multiple exercises. In addition, when a student answers a question involving a skill, not only the knowledge state of the knowledge point will change, but also the mastery state of other knowledge that has an implicit relationship with the knowledge point will change accordingly. Based on these data, more explanatory feedback can be generated on the student's skill mastery. In multiplication skills, the learner's performance is stable, and the probability of correctly using the skill next time is predicted to be close to 90%. It is recommended to try more difficult questions; in subtraction skills, the accuracy rate is low, and the predicted accuracy rate is about 60%. It is recommended to review related concepts and increase practice; the mastery of addition skills is gradually improved, and it is recommended to maintain the current progress; while in division skills, the predicted accuracy rate is only 50%, and it is recommended to concentrate on practice to enhance confidence and proficiency. Through this visualization and explanatory feedback, not only can the learner's dynamic knowledge mastery be clearly tracked, but also targeted learning path suggestions can be provided to help learners improve their overall learning performance more effectively.

[0141] In summary, the innovation of the method of the present invention lies in:

[0142] 1) Integrate emotional factors to model knowledge status

[0143] Unlike traditional knowledge tracking models that only focus on students' historical answer records, this invention innovatively incorporates students' emotional states (such as concentration, confusion, frustration, etc.) into the modeling process. By integrating student characteristics, problem difficulty characteristics, and number of attempts characteristics, a personalized emotion perception module is constructed, which achieves accurate capture of emotional dynamics and modeling of personalized learning behaviors.

[0144] 2) Hypergraph-based high-order dependency modeling

[0145] Traditional graph structures have limitations when modeling complex relationships between exercises and knowledge points, especially when it is difficult to reflect the many-to-many relationships brought about by multiple questions sharing knowledge points. The present invention uses a hypergraph structure to model the many-to-many complex relationship between exercises and knowledge points, treating exercises as nodes and knowledge points as hyperedges, breaking through the limitation that traditional graph neural networks can only handle one-to-one or one-to-many relationships. Hypergraph neural networks can capture high-order dependencies and improve the representation capabilities of knowledge tracking models in complex knowledge structures.

[0146] 3) Sentiment-weighted hypergraph convolution mechanism

[0147] An innovative emotion-weighted hypergraph convolution mechanism is proposed to dynamically integrate students' emotional states into the information dissemination process, realizing personalized regulation of feature aggregation between nodes. In the case of large emotional fluctuations, the knowledge tracking model can adaptively adjust the influence of different nodes, thereby highlighting the role of key knowledge points and enhancing the prediction accuracy and generalization ability of the knowledge tracking model.

[0148] 4) Jointly optimize emotion perception and knowledge prediction

[0149] By designing a joint loss function, the prediction loss of the knowledge tracking task and the optimization loss of the emotion perception module are comprehensively considered to balance the modeling process of emotion and knowledge state. This method not only improves the accuracy of prediction, but also enhances the robustness and generalization ability of the knowledge tracking model on data without real emotion labels.

[0150] It should also be noted that the knowledge tracking method based on hypergraph neural network and emotion perception in the above embodiment can essentially be executed by a computer program or module. Therefore, similarly, based on the same inventive concept, another preferred embodiment of the present invention also provides a knowledge tracking system based on hypergraph neural network and emotion perception corresponding to the knowledge tracking method based on hypergraph neural network and emotion perception provided in the above embodiment, such as Figure 4 As shown, it includes:

[0151] A data acquisition module is used to obtain an exercise data set with a real emotional state label or an exercise data set without a real emotional state label from an education platform or an intelligent tutoring system. Both exercise data sets are composed of exercise data of a student account. Each exercise data includes an exercise, a real answer result, knowledge points included in the exercise, and the number of attempts to complete the exercise. The real emotional state label is confusion, concentration, boredom, or frustration.

[0152] The correlation matrix acquisition module is used to convert the relationship between exercises and knowledge points in the exercise data into a matrix representation and construct an exercise-knowledge point correlation matrix;

[0153] A hypergraph acquisition module is used to transform the exercise data of the student account into an exercise-knowledge point hypergraph according to the exercise-knowledge point association matrix, wherein each node in the exercise-knowledge point hypergraph is each exercise in the exercise data, and the knowledge point hyperedge between two or more nodes in the exercise-knowledge point hypergraph is each knowledge point in the exercise data;

[0154] A model training module is used to build a knowledge tracking model including a hypergraph neural network, a sentiment perception module, and a prediction module. An exercise data set and an exercise-knowledge point hypergraph are used as input to train the knowledge tracking model.

[0155] The result acquisition module is used to input the exercise data of the student account to be tracked into the trained knowledge tracking model, output the predicted value of the student's answer to the next exercise, and push it to the user interface of the education platform or intelligent tutoring system in a preset manner;

[0156] The mastery probability acquisition module is used to pass the final hidden state output by the knowledge tracking model through an additional linear layer, map the final hidden state to the knowledge point dimension, and use the Sigmoid activation function to normalize the final hidden state after the linear layer to the mastery probability of each knowledge point, which is used to indicate the student's mastery of each knowledge point;

[0157] The learning suggestion generation module is used to compare the mastery probability output by the mastery probability acquisition module with a preset threshold, and regard the knowledge points with a mastery probability greater than or equal to the threshold as mastered knowledge points, and regard the knowledge points with a mastery probability less than the threshold as weak knowledge points. For the mastered knowledge points, the user interface displays suggestions for increasing the difficulty or expanding the knowledge, and for the weak knowledge points, the user interface displays suggestions for reviewing or performing intensive exercises.

[0158] It should be noted that, for the knowledge tracking system of the present invention, the mastery probability acquisition module processes the student's practice data through the GRU in the prediction module to obtain the final hidden state, which can capture the student's dynamic learning process and knowledge state in the time series. Next, the final hidden state is mapped to the knowledge point dimension through a linear layer to generate a feature representation corresponding to each knowledge point. In addition, in order to make the output knowledge mastery reflect the student's mastery of each knowledge point, the Sigmoid activation function is used to normalize the output of the linear layer to the [0,1] interval to obtain the mastery probability of each knowledge point. These probability values ​​represent the student's current mastery of each knowledge point, that is, the higher the mastery, the closer the probability value is to 1; the lower the mastery, the closer the probability value is to 0.

[0159] It should be noted that the learning suggestion generation module in the knowledge tracking system will identify students' weak knowledge points and mastered knowledge points based on the students' mastery probability of each knowledge point by setting appropriate thresholds, thereby generating personalized learning suggestions. Specifically, for knowledge points with low mastery (mastery probability is lower than the set threshold), these knowledge points will be identified as students' weak links, and targeted review and reinforcement exercises will be recommended to help students fill in knowledge gaps. Such suggestions may include repeated exercises, providing relevant learning resources, or increasing guided feedback to gradually improve students' mastery. For knowledge points with high mastery (mastery probability is higher than the set threshold), these knowledge points are identified as mastered content, and students are advised to challenge more difficult tasks or expand their knowledge to further enhance the depth of learning. These tasks may involve more complex problems, comprehensive applications across knowledge points, or guide students to explore more advanced learning content independently.

[0160] It is understandable that the knowledge tracking method based on hypergraph neural network and emotion perception described in S1 to S5 above can be implemented by computer program. Therefore, based on the same inventive concept, another preferred embodiment of the present invention also provides a computer program product corresponding to the knowledge tracking method based on hypergraph neural network and emotion perception provided in the above embodiment, which includes a computer program / instruction. When the computer program / instruction is executed by the processor, the knowledge tracking method based on hypergraph neural network and emotion perception as described in the above embodiment can be implemented.

[0161] Similarly, based on the same inventive concept, another preferred embodiment of the present invention also provides a computer electronic device corresponding to the knowledge tracking method based on hypergraph neural network and emotion perception provided in the above embodiment, which includes a memory and a processor;

[0162] The memory is used to store computer programs;

[0163] The processor is used to implement the knowledge tracking method based on hypergraph neural network and emotion perception in the above embodiment when executing the computer program.

[0164] In addition, the logic instructions in the above-mentioned memory can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention.

[0165] Therefore, based on the same inventive concept, another preferred embodiment of the present invention also provides a computer-readable storage medium corresponding to the knowledge tracking method based on hypergraph neural network and emotion perception provided in the above embodiment, and a computer program is stored on the storage medium. When the computer program is executed by the processor, it can implement the knowledge tracking method based on hypergraph neural network and emotion perception in the above embodiment.

[0166] It is understandable that the above storage medium may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage. The storage medium may also be a U disk, a mobile hard disk, a magnetic disk or an optical disk, etc., which can store program codes.

[0167] It is understandable that the above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0168] It should also be noted that those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working process of the system described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here. In the various embodiments provided in this application, the division of steps or modules in the system and method is only a logical function division, and there may be other division methods in actual implementation, such as multiple modules or steps can be combined or integrated together, and a module or step can also be split.

[0169] The above-described embodiment is only a preferred solution of the present invention, but it is not intended to limit the present invention. A person skilled in the relevant technical field may make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, any technical solution obtained by equivalent replacement or equivalent transformation falls within the protection scope of the present invention.

Claims

1. A knowledge tracking method based on hypergraph neural network and emotion perception, characterized in that: The following steps are involved: S1: Obtain an exercise dataset with or without real emotional state labels from an education platform or intelligent tutoring system. Both exercise datasets consist of exercise data from student accounts. Each exercise data contains exercises, real answer results, knowledge points included in the exercises, and the number of attempts to complete the exercises. S2: Convert the relationship between exercises and knowledge points in the practice data into a matrix representation and construct an exercise-knowledge point association matrix; S3: Based on the exercise-knowledge point association matrix, the exercise data of the student account is transformed into an exercise-knowledge point hypergraph, where each node in the exercise-knowledge point hypergraph is each exercise in the exercise data, and the knowledge point hyperedge between two or more nodes in the exercise-knowledge point hypergraph is each knowledge point in the exercise data; S4: Construct a knowledge tracking model that includes a hypergraph neural network, a sentiment perception module, and a prediction module. Take an exercise dataset and an exercise-knowledge point hypergraph as input to train the knowledge tracking model. S5: Input the exercise data to be tracked in the student account into the trained knowledge tracking model, output the predicted value of the student's answer to the next exercise, and push it to the user interface of the education platform or intelligent tutoring system in a preset manner.

2. A knowledge tracking method based on hypergraph neural network and emotion perception as claimed in claim 1, characterized in that: In step S2, each element in the exercise-knowledge point association matrix is ​​used to indicate whether an exercise is associated with a knowledge point. When an element in the exercise-knowledge point association matrix is ​​1, it indicates that the exercise is associated with the knowledge point. Otherwise, when it is 0, it indicates that the exercise is not associated with the knowledge point.

3. The knowledge tracking method based on hypergraph neural network and emotion perception as claimed in claim 1, characterized in that: In step S4, in the knowledge tracking model, the specific training process is as follows: S41: Input the constructed exercise-knowledge point hypergraph into the hypergraph neural network and output the exercise embedding representation; S42: The student account ID, exercise ID, and number of attempts to complete the exercise in the exercise data set are each embedded in a low dimension through an embedding layer, and a unique identification vector of the student account ID, a unique identification vector of the exercise ID, and a unique identification vector of the number of attempts are obtained accordingly. The unique identification vector of the student account ID, the unique identification vector of the exercise ID, and the unique identification vector of the number of attempts are input into the emotion perception module together, and the emotion knowledge state representation is output; S43: The exercise embedding representation is fused with the emotional knowledge state representation through matrix multiplication to obtain the emotional enhanced exercise embedding representation; S44: Input the sentiment-enhanced exercise embedding representation into the prediction module, firstly process the sentiment-enhanced exercise embedding representation by the gated recurrent unit to obtain the final hidden state, and then pass the final hidden state through a linear layer to map the sentiment-enhanced exercise embedding representation into the predicted value of the student's answer result for the next exercise; S45: The cross entropy loss between the predicted value of the student's answer to the next exercise and the actual answer result is used as the prediction loss. When the exercise data set input to the knowledge tracking model is an exercise data set with a real emotional state label, the mean square error loss between the emotion perception embedding representation output by the emotion perception module and the real emotional state label is calculated as the emotion perception loss; when the exercise data set input to the knowledge tracking model is an exercise data set without a real emotional state label, a contrastive learning strategy is adopted to construct a contrastive loss with the cosine similarity of the positive and negative sample pairs and use it as the emotion perception loss; wherein the real emotional state label is confusion, concentration, boredom or frustration; S46: Taking the weighted sum of the emotion perception loss and the prediction loss as the total loss, updating the knowledge tracking model parameters based on minimizing the total loss, and continuously iterating the training until the knowledge tracking model converges to obtain a trained knowledge tracking model.

4. The knowledge tracking method based on hypergraph neural network and emotion perception as claimed in claim 3, characterized in that: In step S42, in the emotion perception module, the specific processing flow is as follows: S421: multiply the unique identification vector of the student account ID by the trainable first weight matrix and pass it through a sigmoid activation function to output the student perception embedding vector; S422: multiplying the unique identification vector of the number of attempts by the trainable second weight matrix and performing a sigmoid activation function to output an embedding vector of the number of attempts; S423: multiplying the unique identification vector of the exercise ID by the trainable third weight matrix and then passing through a sigmoid activation function to output a knowledge point difficulty embedding vector; S424: concatenating the student perception embedding vector, the attempt number embedding vector, and the knowledge point difficulty embedding vector as a fused embedding vector, and passing the fused embedding vector through a linear layer and a sigmoid activation function to output a sentiment perception embedding representation; S425: Perform an element-by-element multiplication operation on the joint representation vector of the exercise and the answer result and the emotional perception embedding representation to obtain the emotional knowledge state representation.

5. The knowledge tracking method based on hypergraph neural network and emotion perception as claimed in claim 4, characterized in that: In step S425, the specific method for generating the joint representation vector of the exercises and answer results is as follows: divide the exercise data in the exercise data set into multiple groups of exercise data, encode the exercises and answer results in each group of exercise data, and obtain the encoding vectors corresponding to each group of exercise data, and the encoding vectors of each group of exercise data constitute the joint representation vector of the exercises and answer results; wherein, the encoding vector of the g-th group is a 2F-dimensional vector, F represents the total number of exercises, and is encoded with 0 and 1 according to the actual answer results of the students: if the student answers the exercises correctly, the element value at the index of the exercise (ID-1) in the encoding vector of the g-th group is set to 1, and the remaining positions are set to 0; if the student answers the exercises incorrectly, the element value at the index of the exercise (ID+F-1) in the encoding vector of the g-th group is set to 1, and the remaining positions are set to 0.

6. The knowledge tracking method based on hypergraph neural network and emotion perception as claimed in claim 3, characterized in that: In step S45, the specific process of adopting the contrastive learning strategy and constructing the contrastive loss is as follows: first, determine whether the actual answer results of the students completing the two exercises are the same to construct positive and negative sample pairs: if the actual answer results of the students completing the two exercises are the same, then the corresponding emotion-aware embedding representations constitute a positive sample pair; if the actual answer results of the students completing the two exercises are different, then the corresponding emotion-aware embedding representations constitute a negative sample pair; then calculate the cosine similarity of the positive sample pair and the cosine similarity of the negative sample pair, and finally construct the contrastive loss based on the cosine similarity of the positive sample pair and the cosine similarity of the negative sample pair.

7. A knowledge tracking system based on hypergraph neural network and emotion perception, characterized in that: include: A data acquisition module is used to obtain an exercise data set with a real emotional state label or an exercise data set without a real emotional state label from an education platform or an intelligent tutoring system. Both exercise data sets are composed of exercise data of a student account. Each exercise data includes an exercise, a real answer result, knowledge points included in the exercise, and the number of attempts to complete the exercise. The real emotional state label is confusion, concentration, boredom, or frustration. The correlation matrix acquisition module is used to convert the relationship between exercises and knowledge points in the exercise data into a matrix representation and construct an exercise-knowledge point correlation matrix; A hypergraph acquisition module is used to transform the exercise data of the student account into an exercise-knowledge point hypergraph according to the exercise-knowledge point association matrix, wherein each node in the exercise-knowledge point hypergraph is each exercise in the exercise data, and the knowledge point hyperedge between two or more nodes in the exercise-knowledge point hypergraph is each knowledge point in the exercise data; A model training module is used to build a knowledge tracking model including a hypergraph neural network, a sentiment perception module, and a prediction module. An exercise data set and an exercise-knowledge point hypergraph are used as input to train the knowledge tracking model. The result acquisition module is used to input the exercise data of the student account to be tracked into the trained knowledge tracking model, output the predicted value of the student's answer to the next exercise, and push it to the user interface of the education platform or intelligent tutoring system in a preset manner; The mastery probability acquisition module is used to pass the final hidden state output by the knowledge tracking model through an additional linear layer, map the final hidden state to the knowledge point dimension, and use the Sigmoid activation function to normalize the final hidden state after the linear layer to the mastery probability of each knowledge point, which is used to indicate the student's mastery of each knowledge point; The learning suggestion generation module is used to compare the mastery probability output by the mastery probability acquisition module with a preset threshold, and regard the knowledge points with a mastery probability greater than or equal to the threshold as mastered knowledge points, and regard the knowledge points with a mastery probability less than the threshold as weak knowledge points. For the mastered knowledge points, the user interface displays suggestions for increasing the difficulty or expanding the knowledge, and for the weak knowledge points, the user interface displays suggestions for reviewing or performing intensive exercises.

8. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instruction is executed by a processor, it can implement the knowledge tracking method based on hypergraph neural network and emotion perception as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by the processor, the knowledge tracking method based on hypergraph neural network and emotion perception as described in any one of claims 1 to 6 is implemented.

10. A computer electronic device, characterized in that: including memory and processor; The memory is used to store computer programs; The processor is used to implement the knowledge tracking method based on hypergraph neural network and emotion perception as described in any one of claims 1 to 6 when executing the computer program.

Images