Knowledge tracing method and system based on answering state and multidimensional item response theory

Abstract

The application provides a new knowledge tracking method and system based on an answering state and a multi-dimensional item response theory, and comprises the following steps: (1) modeling a student knowledge state; (2) modeling a student answering state; and (3) predicting a student answering performance. The application utilizes technical methods such as a gated recurrent unit, a multi-head attention mechanism and a multi-dimensional item response theory, analyzes student historical practice interaction data systematically, integrates the answering state into the knowledge tracking modeling with the knowledge state as the core, and combines the explainable advantage of the multi-dimensional item response theory, so that the accuracy and explainability of the student performance prediction are improved.

Description

Technical Field

[0001] This invention relates to the field of computer data processing technology, and in particular to a knowledge tracking method and system based on answer status and multidimensional item response theory. Background Technology

[0002] The rapid development of intelligent tutoring systems has created opportunities for personalized learning. Knowledge tracking and prediction, as a key technology in personalized learning, aims to accurately track learners' knowledge mastery levels and predict their responses, providing technical support for computer-aided education applications such as intelligent tutoring system development, course learning, and learning resource recommendation.

[0003] The core technology for personalized education services in intelligent tutoring systems is knowledge tracking, which involves modeling changes in students' knowledge status based on their practice and interaction data during the learning process to predict their future performance. Knowledge tracking has received widespread attention due to its significant role in improving student learning efficiency and teacher teaching effectiveness.

[0004] Currently, research on knowledge tracing mainly focuses on deep learning-based knowledge tracing models, which can be further subdivided into models based on recurrent neural networks (RNNs), memory-enhanced networks, and attention networks, according to their network structure. Deep Knowledge Tracing (DKT), a representative RNN-based model, was the first model to utilize deep neural networks in the field of knowledge tracing. DKT transforms student practice interaction time-series data into vector inputs through one-hot encoding or compressed sensing, and finally treats the hidden states of the RNN as the student's knowledge state. Dynamic Key-Value Memory Network (DKVMN) is a typical memory-enhanced network-based model, which adds a storage module and corresponding read / write mechanisms to the traditional neural network, providing a detailed characterization of the student's knowledge state. Self-Attentive model for Knowledge Tracing (SAKT) is a classic attention network-based model that effectively solves the problems of data sparsity and long-distance dependencies using self-attention mechanisms.

[0005] Although deep learning-based knowledge tracking models perform excellently in intelligent tutoring systems, they have the following shortcomings:

[0006] (1) Due to the complexity and opacity of the internal structure of neural networks, knowledge tracking models based on deep learning have poor interpretability and are difficult to effectively analyze students' knowledge status.

[0007] (2) This type of knowledge tracking model attributes students' performance solely to their knowledge state, while ignoring their non-knowledge state, making it difficult to further improve the model's prediction accuracy. The non-knowledge state during the student's answering process is defined as the answering state, which refers to the student's psychological reaction in a specific answering environment, including but not limited to the psychological state constituted by factors such as anxiety, concentration, and tension.

[0008] To address the above issues, this invention designs a knowledge tracking method based on answer states and multidimensional item response theory (MIRT). It integrates answer states into knowledge tracking modeling centered on knowledge states, while also leveraging the interpretability advantages of MIRT to improve the accuracy and interpretability of student performance prediction. Summary of the Invention

[0009] The main objective of this invention is to provide a knowledge tracking method and system based on answer status and multidimensional item response theory to solve the above-mentioned technical problems.

[0010] To achieve the above objectives, this invention provides a knowledge tracking method based on answer status and multidimensional item response theory, comprising the following steps:

[0011] S1, knowledge state modeling, includes:

[0012] S1, knowledge state modeling, includes:

[0013] S11, Construct a heterogeneous knowledge graph HKG, HKG = {V; E}, V = {V s V e V k} is a set composed of three different entities: students, exercises, and knowledge concepts. V s V represents the student entity. e V represents the practice entity. k Represents a knowledge concept entity; E = {E s-e E e-k Let E be the set of edges representing relationships between different entities. s-e E represents the student-practice response relationship. e-k This indicates the inclusion relationship between practice and knowledge concepts;

[0014] S12, generate practice node embeddings of relationships between aggregated nodes in the heterogeneous knowledge graph HKG using a graph convolutional network (GCN).

[0015] S13, embed the practice node into the representation. The results of the practice exercises are used as input to a gated recurrent neural network (GRU) to update the student's knowledge status.

[0016] S2, answer state modeling, includes:

[0017] S21, filter out answer behavior characteristics that are closely related to students' answer results r;

[0018] S22 embeds answer behavior features into a low-dimensional space to reduce feature dimensionality, and captures the high-order relationship between practice and answer behavior features through a multi-head attention mechanism to model the student's answer state;

[0019] S3, Student Performance Prediction, includes:

[0020] The improved multidimensional item response theory model combines the outputs of the student knowledge state model in step S1 and the outputs of the answer state model in step S2 to jointly predict student answer performance.

[0021] Preferably, step S11 includes: defining two meta-paths in the heterogeneous knowledge graph HKG: ese and eke; wherein ese indicates that the neighboring nodes of the exercise are exercises answered by the same student; eke indicates that the neighboring nodes of the exercise can also be exercises that examine the same knowledge concept;

[0022] Step S12 includes: representing the i-th node in the heterogeneous knowledge graph HKG as n i n i The neighbor nodes are represented as N i The graph convolutional network GCN has n nodes in the l-th layer. i Represented as:

[0023]

[0024] Among them, c l and b l Let represent the aggregate weights and biases in the l-th layer, respectively, and ReLU represent the activation function;

[0025] After embedding and propagation through a graph convolutional network (GCN), we obtain

[0026] Preferably, step S13 includes:

[0027] vector Convert to a low-dimensional dense real-valued vector Secondly As the input vector of the gated recurrent neural network (GRU), it is then used to determine the student's knowledge state α. t Modeling:

[0028]

[0029]

[0030]

[0031]

[0032] α t =σ(W α h t +b h (6)

[0033] Where σ represents the activation function, ⊙ represents element-wise multiplication, and W z W r W h W α Let b be the weight matrix. h Indicates bias. s i In the hidden state at time step t, D h This indicates the size of the hidden layer in the GRU neural network.

[0034] Preferably, step S21 includes: calculating the correlation between student answer behavior characteristics and student answer results r using the Pearson correlation coefficient; the formula for calculating the Pearson correlation coefficient is as follows:

[0035]

[0036] Where P(X,Y) represents the correlation between variables X and Y when any two samples X and Y are used as variables; x i Let y represent the i-th variable data of sample X. i Let represent the i-th variable data in sample Y, and n represent the number of variables; This represents the mean of sample X. This represents the mean of sample Y;

[0037] The top L features F = {f1, f2, ..., f1} that are most strongly correlated with the student's answer result r, calculated by formula (7), are used to calculate the results. l ,…,f L This serves as the input data for the student's answer status model.

[0038] Preferably, step S22 includes:

[0039] Let the feature set F = {f1, f2, ..., f l ,…,f L} is represented as a vector, if f l If it belongs to the classification feature, it is represented as a one-hot vector, if f l If it is a numerical feature, it is represented as a scalar, and the feature f is further obtained. l The vector representation of f′ lTo avoid overfitting of the model due to sparse and high-dimensional vectors, f′ is... l Embedded into a low-dimensional continuous space, using low-dimensional feature vectors Represents the input vector f′ l :

[0040]

[0041] Among them, W l It is feature f l The embedding matrix;

[0042] Practice e alone t The embedding representation is used to obtain e′ t , will e t The knowledge concept embeddings k′ m and practice embedding e′ t By merging, we obtain exercise e. t Enhanced representation of c t :

[0043] c t =W e (e′ t ⊕k′ m )+b e (9)

[0044] Where ⊕ represents vector concatenation, W e and b e This represents the training parameters.

[0045] Preferably, step S22 further includes: calculating the query vector corresponding to attention head h respectively. key vector and value vector

[0046]

[0047] in, These represent the projection matrices of the query vector, key vector, and value vector onto the attention head h, respectively, which project their respective vectors into their corresponding subspaces.

[0048] Calculate c under attention head h. t and similarity percentage between

[0049]

[0050] in, This represents calculating c using the inner product. t and Similarity;

[0051] By combining with c t All relevant features are used to represent the relationship between attentional head-down practice and answering behavior features.

[0052]

[0053] By creating three different attention heads, different higher-order relations are learned separately, and then the higher-order relations R learned from the three attention heads are collected. cf :

[0054]

[0055] To preserve the original feature vector c t The information is used to add standard residual connections to the network:

[0056]

[0057] Among them, W c It is a projection matrix;

[0058] By overlaying multiple interaction layers and using the output of the previous layer as the input of the next, the relationship between practice and answering behavior characteristics can be modeled. Nonlinear projection is used to model student behavior in e... t The degree of answering status β t :

[0059]

[0060] Among them, W R Let b be the projection vector. R It's a deviation.

[0061] Preferably, step S3 specifically includes: improving the multidimensional item response theory model to predict student s i At α t Below e t The degree of mastery θ t :

[0062]

[0063] in, For time step t, student s i In k m The degree of mastery on d tm For e t The knowledge concept k included m The difficulty, ω tm For k m The importance of, a t For e tDiscrimination, d tm ω tm and a t For training parameters;

[0064] Student performance is determined by both their mastery of the exercises and their answering style. Formula (17) can be used to predict student performance. i In e t The upper part is θ t and β t The performance of the answers determined jointly y t :

[0065]

[0066] Where, β t For s i In e t The level of answering status on the screen Represents e t The overall difficulty of all the knowledge concepts, combined with their importance, is e t The difficulty, λ t (0<λ t <1) indicates that in e t θ t The weights are the proportions of the training parameters.

[0067] The present invention also provides a knowledge tracking system based on answer state and multidimensional item response theory, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the knowledge tracking method based on answer state and multidimensional item response theory as described in any of the preceding claims.

[0068] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the knowledge tracking method based on answer state and multidimensional item response theory as described in any of the preceding claims:

[0069] This invention utilizes gated recurrent neural networks, multi-head attention mechanisms, and multidimensional item response theory to systematically analyze students' historical practice interaction data. It integrates the answer state into knowledge tracking modeling centered on knowledge state, and combines the interpretability advantages of multidimensional item response theory to improve the accuracy and interpretability of student performance prediction. Attached Figure Description

[0070] The accompanying drawings, as part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments and descriptions of the invention are used to explain the invention, but do not constitute an undue limitation of the invention. Obviously, the drawings described below are merely some embodiments, and those skilled in the art can obtain other drawings based on these drawings without creative effort. In the drawings:

[0071] Figure 1 This is a flowchart illustrating a knowledge tracking method based on answer status and multidimensional item response theory in one embodiment of the present invention.

[0072] Figure 2 This is a schematic diagram of the process for obtaining a student's knowledge status in one embodiment of the present invention.

[0073] Figure 3 This is a flowchart illustrating the process of obtaining a student's answer status in one embodiment of the present invention.

[0074] Figure 4 This is a schematic diagram of the hardware structure for running a knowledge tracking method based on answer status and multidimensional item response theory in one embodiment of the present invention.

[0075] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0076] The technical problems solved by the embodiments of the present invention, the technical solutions adopted, and the technical effects achieved will be clearly and completely described below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other equivalent or obvious variations of embodiments obtained by those skilled in the art without creative effort fall within the protection scope of the present invention. The embodiments of the present invention can be embodied in various different ways as defined and covered by the claims.

[0077] It should be noted that many specific details are given in the following description for ease of understanding. However, it is obvious that the present invention may be implemented without these specific details.

[0078] It should be noted that, in the absence of explicit limitations or conflicts, the various embodiments and their technical features in this invention can be combined with each other to form a technical solution.

[0079] This invention proposes a knowledge tracking method based on answer status and multidimensional item response theory, including the following steps:

[0080] S1, knowledge state modeling, includes:

[0081] S11, Construct a heterogeneous knowledge graph HKG, HKG = {V; E}, V = {V s V e V k} is a set composed of three different entities: students, exercises, and knowledge concepts. V s V represents the student entity. e V represents the practice entity. k Represents a knowledge concept entity; E = {E s-e E e-k Let E be the set of edges representing relationships between different entities. s-e E represents the student-practice response relationship. e-k This indicates the inclusion relationship between practice and knowledge concepts;

[0082] S12, using a graph convolutional network (GCN) to generate practice node embeddings representing heterogeneous knowledge graphs (HKG) that aggregate different entity relationships.

[0083] S13, embed the practice node into the representation. The student's knowledge status is updated by using the answer result r from the exercise as the input of the gated recurrent neural network (GRU).

[0084] In the student practice interaction data, the three entities of student, practice, and knowledge concept can form a heterogeneous knowledge graph (HKG).

[0085] To further model the rich semantic information and heterogeneous relationships in HKG, two meta-paths are defined in HKG: ese and eke.

[0086] Preferably, step S11 includes: defining two meta-paths in HKG: ese and eke; wherein ese indicates that the neighboring nodes of the exercise are exercises answered by the same student; eke indicates that the neighboring nodes of the exercise can also be exercises that test the same knowledge concept. By defining meta-paths, complex interaction relationships between nodes can be explored and neighboring node selection can be guided, thereby enriching the node embedding representation.

[0087] Heterogeneous graph embedding is based on meta-paths to embed the rich structural and semantic information of heterogeneous graphs into low-dimensional node representations. This invention, based on two proposed meta-paths, generates embedded representations of practice nodes in HKG that aggregate student responses and knowledge concept information through a Graph Convolutional Network (GCN). Step S12 includes: representing the i-th node in the heterogeneous knowledge graph HKG as n i n i The neighbor nodes are represented as N i The graph convolutional network GCN has n nodes in the l-th layer.i Represented as:

[0088]

[0089] Among them, c l and b l Let represent the aggregation weights and biases in the l-th layer, respectively, and ReLU represent the activation function; after propagation through the graph convolutional network GCN, we obtain...

[0090] This invention embeds practice nodes into a representation. The exercise response result r is used as the input to a gated recurrent neural network (GRU). Preferably, step S13 includes:

[0091] vector Convert to a low-dimensional dense real-valued vector Secondly As the input vector of the gated recurrent neural network (GRU), it is then used to determine the student's knowledge state α. t Modeling:

[0092]

[0093]

[0094]

[0095]

[0096] α t =σ(W α h t +b h (6)

[0097] Where σ represents the activation function, ⊙ represents element-wise multiplication, and W z W r W h W α Let b be the weight matrix. h Indicates bias. s i In the hidden state at time step t, D h This indicates the size of the hidden layer in the GRU neural network.

[0098] S2, answer state modeling, includes:

[0099] S21, filter out answer behavior features related to student answer results r;

[0100] S22 embeds the answer behavior features into a low-dimensional space to reduce the feature dimensionality, and captures the high-order relationship between practice and answer behavior features through a multi-head attention mechanism to model the student's answer state.

[0101] Preferably, step S21 includes: calculating the correlation between student answering behavior characteristics and student answering results r using the Pearson correlation coefficient. The formula for calculating the Pearson correlation coefficient is as follows:

[0102]

[0103] Where P(X,Y) represents the correlation between variables X and Y when any two samples X and Y are used as variables. Specifically, it involves substituting all the answer behavior features into X and the student's answer results into Y for calculation; x i Let y represent the i-th variable data of sample X. i Let represent the i-th variable data in sample Y, and n represent the number of variables; This represents the mean of sample X. This represents the mean of sample Y;

[0104] The top L features F = {f1, f2, ..., f1} that are most strongly correlated with the student's answer result r, calculated by formula (7), are used to calculate the results. l ,…,f L This serves as the input data for the student's answer status model.

[0105] Preferably, step S22 includes:

[0106] Let the feature set F = {f1, f2, ..., f l ,…,f L} is represented as a vector, if f l If it belongs to the classification feature, it is represented as a one-hot vector, if f l If it is a numerical feature, it is represented as a scalar, and the feature f is further obtained. l The vector representation of f′ l To avoid overfitting of the model due to sparse and high-dimensional vectors, f′ is... l Embedded into a low-dimensional continuous space, using low-dimensional feature vectors Represents the input vector f′ l :

[0107]

[0108] Among them, W l It is feature f l The embedding matrix;

[0109] Practice e alone t The embedding representation is used to obtain e′t , will e t The knowledge concept embeddings k′ m and practice embedding e′ t By merging, we obtain exercise e. t Enhanced representation of c t :

[0110] c t =W e (e′ t ⊕k′ m )+b e (9)

[0111] Where ⊕ represents vector concatenation, W e and b e This represents the training parameters.

[0112] This invention captures the relationship between answering behavior features and practice through a multi-head attention mechanism. Preferably, step S22 further includes: calculating the query vector corresponding to each attention head h. key vector and value vector

[0113]

[0114] in, These represent the projection matrices of the query vector, key vector, and value vector onto the attention head h, respectively, which project their respective vectors into their corresponding subspaces.

[0115] Calculate c under attention head h. t and similarity percentage between

[0116]

[0117] in, This represents calculating c using the inner product. t and Similarity;

[0118] By combining with c t All relevant features are used to represent the relationship between attentional head-down practice and answering behavior features.

[0119]

[0120] because It is C t It is a high-order combination of related answering behavior features (under attention head h), thus representing the complex relationship the model has learned between the two. Furthermore, considering... Different higher-order relations may be involved in different representation spaces. This invention achieves this by using a three-head attention mechanism. Three different attention heads are created to learn different higher-order relations respectively, and then the higher-order relations R learned from the three attention heads are collected. cf :

[0121]

[0122] To preserve the original feature vector c t The information is used to add standard residual connections to the network:

[0123]

[0124] Among them, W c It is a projection matrix;

[0125] By overlaying multiple interaction layers and using the output of the previous layer as the input of the next, the relationship between practice and answering behavior characteristics can be modeled. Nonlinear projection is used to model student behavior in e... t The degree of answering status β t :

[0126]

[0127] Among them, W R Let b be the projection vector. R It's a deviation.

[0128] S3, Student Performance Prediction, includes:

[0129] The improved multidimensional item response theory model combines the outputs of the student knowledge state model in step S1 and the outputs of the answer state model in step S2 to jointly predict student answer performance.

[0130] A student's mastery of the exercises depends not only on their own knowledge level (α) t It is also related to the characteristics of the exercise itself, such as the discrimination of the exercise, the difficulty and importance of the knowledge points tested in the exercise, etc. Given the good interpretability of parameters in the Multidimensional Item Response Theory (MIRT), preferably, step S3 specifically includes: improving the MIRT to predict student s i At α t Below e t The degree of mastery θ t :

[0131]

[0132] in, For time step t, student s i In km The degree of mastery on d tm For e t The knowledge concept k included m The difficulty, ω tm For k m The importance of, a t For e t Discrimination, d tm ω tm and a t For training parameters;

[0133] Student performance is determined by both their mastery of the exercises and their answering style. Formula (17) can be used to predict student performance. i In e t The upper part is θ t and β t The performance of the answers determined jointly y t :

[0134]

[0135] Where, β t For s i In e t The level of answering status on the screen Represents e t The overall difficulty of all the knowledge concepts, combined with their importance, is e t The difficulty, λ t (0<λ t <1) indicates that in e t θ t The weights are the proportions of the training parameters.

[0136] This invention utilizes gated recurrent neural networks, multi-head attention mechanisms, and multidimensional item response theory to systematically analyze students' historical practice interaction data. It integrates the answer state into knowledge tracking modeling centered on knowledge state, and combines the interpretability advantages of multidimensional item response theory to improve the accuracy and interpretability of student performance prediction.

[0137] The present invention also provides a knowledge tracking system based on answer state and multidimensional item response theory, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the knowledge tracking method based on answer state and multidimensional item response theory as described in any of the preceding claims.

[0138] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the knowledge tracking method based on answer state and multidimensional item response theory as described in any of the preceding claims.

[0139] Figure 4 This is a schematic diagram of the hardware structure for running a knowledge tracking method based on answer status and multidimensional item response theory, according to an embodiment of the present invention. Figure 3 As shown, this embodiment / computer 6 includes: a processor 60, a memory 61, and a computer program 62 stored in the memory 61 and executable on the processor 60, such as a program for running a computational thinking evaluation method based on BP neural network and multi-source data fusion. When the processor 60 executes the computer program 62, it implements the steps described in the embodiments of running the computational thinking evaluation method based on BP neural network and multi-source data fusion. Alternatively, when the processor 60 executes the computer program 62, it implements the functions of each module / unit in the above-described device embodiments.

[0140] For example, the computer program 62 may be divided into one or more modules / units, which are stored in the memory 61 and executed by the processor 60 to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program 62 in the computer 6.

[0141] The computer 6 can be a desktop computer, laptop, handheld computer, cloud server, or other computing device. The computer 6 may include, but is not limited to, a processor 60 and a memory 61. Those skilled in the art will understand that... Figure 3 This is merely an example of computer 6 and does not constitute a limitation on computer 6. It may include more or fewer components than shown, or combine certain components, or different components. For example, computer 6 may also include input / output devices, network access devices, buses, etc.

[0142] The processor 60 may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0143] The memory 61 can be an internal storage unit of the computer 6, such as a hard drive or memory. The memory 61 can also be an external storage device of the terminal device 6, such as a plug-in hard drive, smart media card (SMC), secure digital card (SD), flash card, etc., equipped on the terminal device. Furthermore, the memory 61 can include both internal storage units and external storage devices of the computer 6. The memory 61 is used to store the computer program and other programs and data required by the terminal device. The memory 61 can also be used to temporarily store data that has been output or will be output.

[0144] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0145] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0146] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0147] In the embodiments provided by this invention, it should be understood that the disclosed apparatus / terminal devices and methods can be implemented in other ways. For example, the apparatus / terminal device embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0148] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0149] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0150] If the integrated module / unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0151] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A knowledge tracking method based on answer status and multidimensional item response theory, characterized in that, Including the following steps: S1, knowledge state modeling, includes: S11, Constructing the heterogeneous knowledge graph HKG, , It is a collection of three different entities: students, exercises, and knowledge concepts. Represents the student entity. Represents practice entities, Represents knowledge concept entities; It is a set of edges representing relationships between different entities, where, This indicates a student-practice response relationship. This indicates the inclusion relationship between practice and knowledge concepts; S12, generate practice node embeddings of relationships between aggregated nodes in the heterogeneous knowledge graph HKG using a graph convolutional network (GCN). ; S13, embed the practice node into the representation. Based on the results of the practice exercises As input to a gated recurrent neural network (GRU), to update the student's knowledge state; S2, answer state modeling, includes: S21, Filter out results that match student responses. Closely related characteristics of answering behaviors; S22 embeds answer behavior features into a low-dimensional space to reduce feature dimensionality, and captures the high-order relationship between practice and answer behavior features through a multi-head attention mechanism to model the student's answer state; S3, Student Performance Prediction, includes: The improved multidimensional item response theory model is combined with the output results of the student knowledge state model in step S1 and the output results of the answer state model in step S2 to jointly predict the student's answer performance. Step S11 includes: defining two meta-paths in the heterogeneous knowledge graph HKG: and ;in, This indicates that the neighboring nodes of a practice exercise are all answered by the same student. The neighboring nodes of a practice can also be practices that examine the same knowledge concept; Step S12 includes: transferring the first... from the heterogeneous knowledge graph HKG... Each node is represented as , The neighbor nodes are represented as Graph Convolutional Network (GCN) Layer nodes Represented as: in, and They represent the first Aggregation weights and biases in the layer Indicates the activation function; After embedding and propagation through a graph convolutional network (GCN), we obtain = }; Step S13 includes: vector Convert to a low-dimensional dense real-valued vector Secondly As the input vector of the gated recurrent neural network (GRU), it is used to assess the student's knowledge state. Modeling: in, This represents the activation function. This represents element-wise multiplication. , , , This is the weight matrix. Indicates bias. express At time step The hidden state of time, This indicates the size of the hidden layer in the GRU neural network.

2. The knowledge tracking method based on answer status and multidimensional item response theory according to claim 1, characterized in that, Step S21 includes: calculating the student's answering behavior characteristics and student's answering results using the Pearson correlation coefficient. The correlation between them; the Pearson correlation coefficient is calculated as follows: Wherein, P(X,Y) represents the correlation between variables X and Y when any two samples X and Y are used as variables; This represents the data of the i-th variable in sample X. Let represent the i-th variable data in sample Y, and n represent the number of variables; Indicates sample The mean, Indicates sample The mean; Formula Calculated results and student responses The most strongly correlated Features This serves as input data for the student's answer status model.

3. The knowledge tracking method based on answer status and multidimensional item response theory according to claim 2, characterized in that, Step S22 includes: feature set Represented as a vector, if If it belongs to the classification feature, it is represented as a one-hot vector. If it is a numerical feature, it is represented as a scalar, and further features are obtained. vector representation To avoid overfitting of the model due to sparse and high-dimensional vectors, the following measures will be taken: Embedded into a low-dimensional continuous space, using low-dimensional feature vectors Represents the input vector : in, It is a feature The embedding matrix; Practice alone Embedded representation is obtained ,Will The knowledge concept embeddings contained and practice embedding To integrate and obtain practice Enhanced representation : in, This indicates vector concatenation. and This represents the training parameters.

4. The knowledge tracking method based on answer status and multidimensional item response theory according to claim 3, characterized in that, Step S22 further includes: calculating attention heads respectively. The corresponding query vector Key vector and value vector : in, , , These represent the query vector, key vector, and value vector in the attention head, respectively. The projection matrices below project their respective vectors into their corresponding subspaces; Calculation in attention head Down, similarity percentage between : in, The representation is calculated using the inner product. and Similarity; By combining All relevant features to represent the attention head The relationship between practice and answering behavior characteristics : By creating three different attention heads, different higher-order relations are learned separately, and then the higher-order relations learned from the three attention heads are collected. : To preserve the original feature vector The information is used to add standard residual connections to the network: in, It is a projection matrix; By overlaying multiple interaction layers and using the output of the previous layer as the input of the next, the relationship between practice and answering behavior characteristics is modeled. Nonlinear projection is used to model student behavior... The level of answering status on : in, For projection vectors, It's a deviation.

5. The knowledge tracking method based on answer status and multidimensional item response theory according to claim 4, characterized in that, Step S3 specifically includes: improving the multidimensional project response theory model to predict student responses. In Below level of mastery : in, For time step Students exist The degree of mastery of the above, for The knowledge concepts contained The difficulty for The importance of for Distinctiveness , as well as For training parameters; Students' performance in answering questions is determined by both their mastery of the exercises and their answering style. Formula (17) can be used to predict students' performance. exist Up by and Answer performance : in, for exist The level of answering status on the screen represent The overall difficulty of all the knowledge concepts, combined with their importance, is... The difficulty (0) 1) Indicates in The weights are the proportions of the training parameters.

6. A knowledge tracking system based on answer status and multidimensional item response theory, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the knowledge tracking method based on answer state and multidimensional item response theory as described in any one of claims 1 to 5.

7. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the knowledge tracking method based on answer status and multidimensional item response theory as described in any one of claims 1 to 5.

Images