Explainable knowledge cognitive level mining method and system integrating learning cognitive elements

Abstract

This invention belongs to the field of personalized learning technology and discloses a method and system for mining interpretable knowledge cognitive levels that integrates learning cognitive elements. The method includes: constructing a learning resource-response matrix and a learning resource-knowledge point matrix to estimate error and guess parameters based on a cognitive diagnostic model; obtaining a feature vector that contains the learner's overall mastery of potential knowledge points; incorporating the error and guess parameters based on the cognitive diagnostic model to obtain a new feature vector to predict the learner's answer performance on a specific learning resource; and obtaining a knowledge growth vector, the learner's degree of forgetting of knowledge points, and the degree of memory reinforcement, to comprehensively update the learner's knowledge state after interacting with the learning resource. The present invention's method for mining interpretable knowledge cognitive levels that integrates learning cognitive elements is more effective than traditional knowledge cognitive level mining methods.

Description

Technical Field

[0001] The present invention belongs to the field of personalized learning technology, and in particular relates to an interpretable knowledge cognitive level mining method and system integrating learning cognitive elements. Background Art

[0002] At present, the online education model still continues the traditional offline education model. Although a series of online learning platforms, computer-assisted systems and other electronic learning software are used, the learning data generated by learners in the online education model are still not well utilized, nor are personalized learning services provided to each student.

[0003] Knowledge and cognitive level mining is a typical method for providing personalized learning services. Among existing knowledge and cognitive level mining methods, the most widely used ones are those that incorporate deep learning. This type of method is highly effective in characterizing changes in learners' knowledge states and predicting their performance. However, due to the "black box" nature of deep learning, the interpretability of this type of knowledge and cognitive level mining method in the field of education needs to be improved. Therefore, a number of studies have been conducted to improve the interpretability of knowledge and cognitive level mining models, but most of these are based on statistical methods. These methods typically quantify learners' cognitive elements based on common sense or educational theory. While this improves interpretability, the mined cognitive elements are not deep enough, failing to delve into the learners' cognitive level and remaining solely at the data-driven level.

[0004] Through the above analysis, the problems and defects of the existing technology are as follows: the existing knowledge and cognitive level mining systems or methods are less interpretable in the field of education; the mining of learners' cognitive elements is not deep enough and the mining methods are single. Summary of the Invention

[0005] In response to the problems existing in the prior art, the present invention provides an interpretable knowledge cognitive level mining method and system that integrates learning cognitive elements.

[0006] The present invention is achieved by providing a method for mining the cognitive level of interpretable knowledge by integrating learning cognitive elements. The method comprises:

[0007] Step 1: Based on the interaction sequence between learners and learning resources, a learning resource-response matrix and a learning resource-knowledge point matrix are constructed to estimate the error and guess parameters based on the cognitive diagnosis model.

[0008] Step 2: Create a key matrix to store the potential knowledge points of the learning resource answered by the learner, and a value matrix to store the mastery status of the potential knowledge points; obtain the knowledge points related to the learning resource answered by the learner from the key matrix; read the learner's mastery status of the relevant knowledge points from the value matrix, and obtain a feature vector containing the learner's overall mastery status of the potential knowledge points;

[0009] Step 3: learning the learner's personalized error parameters and guess parameters from the feature vector containing the learner's overall mastery of the potential knowledge points, integrating the error and guess parameters based on the cognitive diagnostic model to obtain a new feature vector to predict the learner's answer performance on a specific learning resource;

[0010] Step 4: Based on the embedding matrix vector representation of the interaction sequence between the learner and the learning resource, and based on the learner's current knowledge status, the knowledge growth vector is obtained; then, based on the learner's personalized forgetting rate, the learner's degree of forgetting of the knowledge points and the degree of memory reinforcement are obtained, and the knowledge status of the learner after the interaction with the learning resource is comprehensively updated.

[0011] Furthermore, the step 1 includes:

[0012] (1) Based on the interaction sequence between learners and learning resources, we make an embedding matrix vector representation of the learning resource sequence and an embedding matrix vector representation of the interaction sequence between learners and learning resources:

[0013] (1.1) Set the embedding matrix A of the learning resource and transform the learning resource sequence (q1,q2,...,q t ) is embedded with the embedding matrix A of the learning resource to obtain the embedding matrix vector representation K=(k1,k2,...,k t );

[0014] (1.2) Set the embedding matrix B of the interaction sequence between learners and learning resources, and transform the interaction sequence between learners and learning resources X = (x1, x2, ..., x t ) is embedded with the embedding matrix B of the interaction sequence between the learner and the learning resource to obtain the embedding matrix vector representation V = (v1, v2, ..., v t );

[0015] (2) Based on the interaction sequence between learners and learning resources, a learning resource-answer response matrix and a learning resource-knowledge point matrix are constructed; the learning resource-answer response matrix is ​​a matrix of learning resources answered by behavioral learners, and is a matrix of learners' answer responses on learning resources; the learning resource-knowledge point matrix is ​​a matrix of learning resources answered by behavioral learners, and is a matrix of knowledge points corresponding to learning resources;

[0016] (3) Based on the learning resource-response matrix and the learning resource-knowledge point matrix, a cognitive diagnosis model is introduced to estimate the error and guess parameters based on the cognitive diagnosis model:

[0017] (3.1) Initialize the learner's knowledge state α, and then combine it with the learning resource-knowledge point matrix Q to obtain the learner's ideal answer situation on the learning resource. The learner's ideal answer situation η on the learning resource is calculated as follows: Where N represents the number of knowledge points contained in the learning resource;

[0018] (3.2) Based on the learning resource-answer response matrix R and the ideal answer matrix η, we formally define the error parameter s and the guess parameter g, and obtain the probability model of the learning resource-answer response matrix R as follows:

[0019] s=P(R=0|η=1)

[0020] g=P(R=1|η=0)

[0021] P(α)=P(R=1|α)=g (1-η) (1-s) η

[0022] (3.3) Introducing the cognitive diagnosis model DINA, we obtain the total likelihood function of the DINA model:

[0023]

[0024] (3.4) Use the maximum likelihood function to estimate the learner's knowledge state, use the EM algorithm to estimate the error parameter and guess parameter, and obtain the error parameter s and guess parameter g based on the cognitive diagnosis model.

[0025] Furthermore, the step 2 includes:

[0026] 1) Create a key matrix to store the potential knowledge points of the learning resources answered by the learner and a value matrix to store the mastery status of the potential knowledge points; the key matrix is The value matrix is ​​KC v =(KC v (1),...,KC v (i),...,KCv (N));

[0027] 2) The embedding matrix vector of the learning resource sequence is represented by k t and value matrix Perform inner product and activate it through sigmoid function to obtain the knowledge points related to the learning resources answered by the learner and the weights of the knowledge points related to the learning resources answered by the learner;

[0028] The weight w of the knowledge points related to the learning resources answered by the learner t The calculation formula is: t =σ(k t KC k );

[0029] 3) Combine the acquired knowledge points related to the learning resources answered by the learners and the weights w of the knowledge points related to the learning resources answered by the learners t , from the value matrix KC v =(KC v (1),...,KC v (i),...,KC v (N)) Read the learner's mastery of relevant knowledge points

[0030] 4) Perform weighted summation on the learner's mastery of relevant knowledge points to obtain the overall mastery status of the knowledge points contained in the learning resource;

[0031] The overall mastery status of the knowledge points contained in the learning resource t The calculation formula is as follows:

[0032]

[0033] 5) The overall mastery status of the knowledge points contained in the learning resources and the embedding matrix vector representation k of the learning resource sequence t Connect and activate through Tanh function to obtain the feature vector containing the learner's overall mastery of the potential knowledge points;

[0034] The feature vector F containing the learner's overall mastery of the potential knowledge points t It is expressed as follows:

[0035] F t =Tanh(W1 T [r t ,k t ]+b1).

[0036] Furthermore, the step three includes:

[0037] First, learn the learner-based personalized error parameters and guess parameters from the feature vector:

[0038] s′ t =tanh(W1F t +b1)

[0039] g′ t =tanh(W2F t +b2)

[0040] Secondly, the learner-based personalized error parameter s' and guess parameter g' as well as the error parameter s and guess parameter g based on the cognitive diagnosis model are incorporated to obtain a new feature vector:

[0041]

[0042]

[0043] Finally, the new feature vector is activated through the sigmoid function to predict the learner's answer performance on a specific learning resource;

[0044] The predicted learner's answer performance P on a specific learning resource t The formula is as follows:

[0045]

[0046] Furthermore, the step 4 includes:

[0047] (1) Combining the learner’s current knowledge state, the new feature vector and the embedding matrix vector representation of the interaction sequence between the learner and the learning resource are obtained. V = (v1, v2, ..., v t ) to obtain the learner’s knowledge growth vector after answering the learning resources:

[0048] v′ t =[v t ,F t ];

[0049] (2) Obtain the knowledge growth vectors of all knowledge points contained in the learning resource, and distribute the knowledge growth vectors of all knowledge points contained in the learning resource in all memory fragments: The knowledge growth vectors of all knowledge points contained in the learning resource are obtained as w t v′ t , the knowledge growth vector is calculated according to the value matrix KC v =(KC v (1),...,KC v (i),...,KC v(N)) value slot is split to obtain the knowledge growth vector (w t v′ t (1),...,w t v′ t (i),...,w t v′ t (N));

[0050] (3) Take each value slot distribution fragment as the input of the LSTM model to obtain the learner's personalized forgetting rate LSTM (w t (i)v′ t );

[0051] (4) Based on the learner's personalized forgetting rate, obtain the learner's forgetting degree of the knowledge point after completing the learning resource:

[0052]

[0053] (5) Obtain the degree of memory enhancement after learners finish learning resources

[0054] (6) Comprehensively update the knowledge status of learners after interacting with learning resources based on the degree of forgetting and memory reinforcement after learners finish answering the learning resources:

[0055] (7) Define the loss function and use the Adma optimizer to update the weights; the loss function is as follows:

[0056]

[0057] (8) Add regularization terms l1 and l2 to the loss function: L' = L + λ1l1 + λ2l2; where λ1 and λ2 are the coefficients corresponding to the regularization terms.

[0058] Furthermore, in step (4), based on the learner's personalized forgetting rate, obtaining the learner's forgetting degree of the knowledge point after completing the learning resource includes:

[0059] (4.1) Input any fragment of the value slot distribution into the LSTM and determine the discarded information: t (i)=σ(W f ·[h t-1 ,w(i)v']+b f ); where W represents the weight coefficient; b represents the corresponding bias vector;

[0060] (4.2) Determine the information to be updated: m t (i)=σ(W m ·[h t-1,w(i)v']+b m );

[0061] (4.3) Determine the information that needs to be updated:

[0062] (4.4) Update the status based on the discarded information and the information to be updated:

[0063] (4.5) Obtain the learner’s personalized forgetting rate:

[0064]

[0065] Among them, h t (i) Indicates the degree of forgetting acquired by the memory fragment of a certain value slot.

[0066] Another object of the present invention is to provide a system for mining the explainable knowledge cognitive level of integrated learning cognitive elements that implements the method for mining the explainable knowledge cognitive level of integrated learning cognitive elements. The system for mining the explainable knowledge cognitive level of integrated learning cognitive elements comprises:

[0067] The learning resource-answer response matrix and the learning resource-knowledge point matrix construction module are used to construct the learning resource-answer response matrix and the learning resource-knowledge point matrix based on the interaction sequence between learners and learning resources, and estimate the error and guess parameters based on the cognitive diagnosis model;

[0068] A feature vector extraction module is configured to create a key matrix storing the potential knowledge points of the learning resource answered by the learner and a value matrix storing the mastery status of the potential knowledge points; simultaneously obtain the knowledge points related to the learning resource answered by the learner from the key matrix; read the learner's mastery status of the relevant knowledge points from the value matrix; and obtain a feature vector containing the learner's overall mastery status of the potential knowledge points;

[0069] A response performance prediction module is configured to learn individualized error parameters and guess parameters based on the learner from the feature vector containing the learner's overall mastery of potential knowledge points, incorporate the error and guess parameters based on the cognitive diagnostic model, and obtain a new feature vector to predict the learner's response performance on a specific learning resource;

[0070] The knowledge state update module is used to obtain the knowledge growth vector based on the embedding matrix vector representation of the interaction sequence between the learner and the learning resource and the learner's current knowledge state. It then obtains the learner's degree of forgetting of knowledge points and the degree of memory reinforcement based on the learner's personalized forgetting rate, and comprehensively updates the learner's knowledge state after interaction with the learning resource.

[0071] Another object of the present invention is to provide a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the following steps of the method for mining the cognitive level of interpretable knowledge that integrates learning cognitive elements:

[0072] Step 1: Based on the interaction sequence between learners and learning resources, a learning resource-response matrix and a learning resource-knowledge point matrix are constructed to estimate the error and guess parameters based on the cognitive diagnosis model.

[0073] Step 2: Create a key matrix to store the potential knowledge points of the learning resource answered by the learner, and a value matrix to store the mastery status of the potential knowledge points; obtain the knowledge points related to the learning resource answered by the learner from the key matrix; read the learner's mastery status of the relevant knowledge points from the value matrix, and obtain a feature vector containing the learner's overall mastery status of the potential knowledge points;

[0074] Step 3: learning the learner's personalized error parameters and guess parameters from the feature vector containing the learner's overall mastery of the potential knowledge points, integrating the error and guess parameters based on the cognitive diagnostic model to obtain a new feature vector to predict the learner's answer performance on a specific learning resource;

[0075] Step 4: Based on the embedding matrix vector representation of the interaction sequence between the learner and the learning resource, and based on the learner's current knowledge status, the knowledge growth vector is obtained; then, based on the learner's personalized forgetting rate, the learner's degree of forgetting of the knowledge points and the degree of memory reinforcement are obtained, and the knowledge status of the learner after the interaction with the learning resource is comprehensively updated.

[0076] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the following steps of the method for mining the cognitive level of interpretable knowledge that integrates learning cognitive elements:

[0077] Step 1: Based on the interaction sequence between learners and learning resources, a learning resource-response matrix and a learning resource-knowledge point matrix are constructed to estimate the error and guess parameters based on the cognitive diagnosis model.

[0078] Step 2: Create a key matrix to store the potential knowledge points of the learning resource answered by the learner, and a value matrix to store the mastery status of the potential knowledge points; obtain the knowledge points related to the learning resource answered by the learner from the key matrix; read the learner's mastery status of the relevant knowledge points from the value matrix, and obtain a feature vector containing the learner's overall mastery status of the potential knowledge points;

[0079] Step 3: learning the learner's personalized error parameters and guess parameters from the feature vector containing the learner's overall mastery of the potential knowledge points, integrating the error and guess parameters based on the cognitive diagnostic model to obtain a new feature vector to predict the learner's answer performance on a specific learning resource;

[0080] Step 4: Based on the embedding matrix vector representation of the interaction sequence between the learner and the learning resource, and based on the learner's current knowledge status, the knowledge growth vector is obtained; then, based on the learner's personalized forgetting rate, the learner's degree of forgetting of the knowledge points and the degree of memory reinforcement are obtained, and the knowledge status of the learner after the interaction with the learning resource is comprehensively updated.

[0081] Another object of the present invention is to provide an information data processing terminal, which is used to implement the explainable knowledge cognitive level mining system that integrates learning cognitive elements.

[0082] In combination with the above technical solutions and the technical problems solved, please analyze the advantages and positive effects of the technical solutions to be protected by the present invention from the following aspects:

[0083] First, in view of the technical problems existing in the above-mentioned prior art and the difficulty of solving these problems, this paper closely combines the technical solutions to be protected by the present invention and the results and data during the research and development process, and analyzes in detail and in depth how the technical solutions of the present invention solve the technical problems and some creative technical effects brought about by solving the problems. The specific description is as follows:

[0084] On the one hand, the present invention combines the learner's cognitive elements estimated by the cognitive diagnosis model, and on the other hand, learns the learner's cognitive elements through a neural network. It not only solves the problem of insufficient mining of learner's cognitive elements in the current knowledge cognitive level field, but also increases the interpretability of the model from the aspects of parameter interpretability, prediction result interpretability and knowledge status update interpretability, and also improves the performance of the model in predicting learner performance.

[0085] The present invention improves the interpretability of the model in the field of education; the present invention provides a new development perspective for the development of the field of knowledge cognitive level mining; the present invention improves the performance of the knowledge cognitive level mining model in predicting learner performance; the present invention takes into account more learner cognitive parameters, deepens the learner's cognitive level, and better simulates the learner's learning process.

[0086] Second, considering the technical solution as a whole or from the perspective of the product, the technical effects and advantages of the technical solution to be protected by the present invention are described in detail as follows:

[0087] The interpretable knowledge cognitive level mining method that integrates learning cognitive elements proposed in the present invention is more effective than the traditional knowledge cognitive level mining method.

[0088] The interpretable cognitive level mining method provided by this invention, which integrates learning cognitive elements, not only helps learners gain a clearer understanding of their own knowledge structure but also provides insights into self-cognitive information, such as error probability, guess probability, and forgetting rate. Teachers can also better intervene and provide targeted instruction, thereby tailoring teaching to individual students' needs.

[0089] Third, as auxiliary evidence for the inventiveness of the claims of the present invention, it is also reflected in the following important aspects:

[0090] The technical solution of the present invention fills the technical gap in the industry at home and abroad: the present invention innovatively integrates the cognitive diagnosis method and the knowledge cognitive level mining method, and incorporates the learner's cognitive elements estimated by the cognitive diagnosis model.

[0091] The present invention uses a neural network to learn learner cognitive element parameters, so that the model has richer methods and more accurate representation in terms of learner cognitive element mining.

[0092] The present invention is based on cognitive diagnosis theory and takes learners' cognitive factors into consideration during prediction. It not only better simulates the learners' learning process, but also improves the interpretability of knowledge cognitive level mining methods in the field of education from the perspective of predictive interpretability.

[0093] When calculating the amount of knowledge growth of a learner, the present invention takes into account the learner's current knowledge status, rather than just focusing on the learner's answering situation.

[0094] Does the technical solution of the present invention solve the technical problems that people have always wanted to solve but have never been able to solve successfully?

[0095] The present invention solves the problem that previous knowledge cognition level mining methods are insufficient in mining learners' cognitive elements. On the one hand, it combines learners' cognitive elements estimated by cognitive diagnosis models, and on the other hand, it learns learners' cognitive elements through the network.

[0096] This invention solves the problem that previous knowledge and cognitive level mining methods use data-driven methods to predict learner performance, resulting in weak model interpretability. It increases the interpretability of the model from the levels of parameter interpretability, prediction result interpretability, and knowledge status update interpretability, and also improves the model's performance in predicting learner performance.

[0097] Whether the technical solution of the present invention overcomes technical prejudice:

[0098] Previous methods for mining knowledge and cognitive levels used deep learning to characterize learners' knowledge and cognitive levels. However, due to the black box nature of deep learning, its interpretability in the education field is poor. The technical solution of this invention integrates cognitive diagnostic theory and learners' cognitive factor parameters, increasing the interpretability of the model from the perspectives of parameter interpretability, prediction result interpretability, and knowledge state update interpretability. BRIEF DESCRIPTION OF THE DRAWINGS

[0099] Figure 1 This is a schematic diagram of the principle of the method for mining the explainable knowledge cognitive level by integrating learning cognitive elements provided by an embodiment of the present invention;

[0100] Figure 2 This is a flow chart of a method for mining explainable knowledge cognitive levels by integrating learning cognitive elements provided by an embodiment of the present invention;

[0101] Figure 3 Schematic diagram of the training process on a virtual data set of a model provided by an embodiment of the present invention;

[0102] Figure 4 This is a schematic diagram of the training process of the model provided by an embodiment of the present invention on a real C language dataset. DETAILED DESCRIPTION

[0103] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0104] 1. Explanatory Examples In order to enable those skilled in the art to fully understand how to implement the present invention, this section provides an illustrative example that expands upon the technical solutions of the claims.

[0105] like Figure 1-Figure 2 As shown, the method for mining the cognitive level of interpretable knowledge by integrating learning cognitive elements provided by the embodiment of the present invention includes:

[0106] S101, based on the interaction sequence between learners and learning resources, construct the learning resource-answer response matrix and the learning resource-knowledge point matrix, and estimate the error and guess parameters based on the cognitive diagnosis model;

[0107] S102, creating a key matrix storing the potential knowledge points of the learning resource answered by the learner and a value matrix storing the mastery status of the potential knowledge points; obtaining the knowledge points related to the learning resource answered by the learner from the key matrix; reading the learner's mastery status of the relevant knowledge points from the value matrix, and obtaining a feature vector containing the learner's overall mastery status of the potential knowledge points;

[0108] S103, learning the learner's personalized error parameters and guess parameters from the feature vector containing the learner's overall mastery of the potential knowledge points, integrating the error and guess parameters based on the cognitive diagnosis model to obtain a new feature vector to predict the learner's answer performance on the specific learning resource;

[0109] S104: Based on the embedding matrix vector representation of the interaction sequence between the learner and the learning resource, and based on the learner's current knowledge status, the knowledge growth vector is obtained; then, based on the learner's personalized forgetting rate, the learner's forgetting degree of the knowledge points and the degree of memory reinforcement are obtained, and the knowledge status of the learner after the interaction with the learning resource is comprehensively updated.

[0110] The method for mining the cognitive level of interpretable knowledge by integrating learning cognitive elements provided by the embodiment of the present invention specifically includes:

[0111] (1) Based on the interaction sequence between learners and learning resources, we make embedding matrix vector representations of the learning resource sequence and the interaction sequence between learners and learning resources, and construct the learning resource-answer response matrix and the learning resource-knowledge point matrix. We estimate the error and guess parameters based on the cognitive diagnosis model, and increase the interpretability of the model from the parameter interpretability level. Specifically, we:

[0112] Step 1.1: Based on the interaction sequence between learners and learning resources, make an embedding matrix-vector representation of the learning resource sequence and an embedding matrix-vector representation of the interaction sequence between learners and learning resources.

[0113] Step 1.1 provided in this embodiment of the present invention further includes:

[0114] Step 1.1.1: Set the embedding matrix A of the learning resource and transform the learning resource sequence (q1,q2,...,q t ) is embedded with the matrix to obtain the embedding matrix vector representation of the learning resource sequence K=(k1,k2,...,k t );

[0115] Step 1.1.2: Set the embedding matrix B of the interaction sequence between the learner and the learning resource, and transform the interaction sequence between the learner and the learning resource X = (x1, x2, ..., x t ) is embedded with the matrix to obtain the embedding matrix vector representation of the interaction sequence between the learner and the learning resource V=(v1,v2,...,v t );

[0116] Step 1.2: Based on the interaction sequence between learners and learning resources, construct a learning resource-response matrix and a learning resource-knowledge point matrix;

[0117] The learning resource-answer response matrix provided in the embodiment of the present invention is a learning resource answered by a behavioral learner, and is listed as a matrix R of the learner's answer response on the learning resource;

[0118] The learning resource-knowledge point matrix provided in the embodiment of the present invention is the learning resource answered by the behavioral learner, and is listed as a matrix Q of knowledge points corresponding to the learning resource.

[0119] Step 1.2 provided in this embodiment of the present invention further includes:

[0120] Step 1.2.1: There are two data sets used for the experiment provided in the embodiment of the present invention, including: a virtual data set and a real C language data set.

[0121] The virtual dataset provided by the embodiment of the present invention includes records of 2,000 learners answering questions on 50 learning resources in sequence, with a total of 100,000 learning data. The 50 learning resources are set to include 5 knowledge points.

[0122] The real C language dataset provided by the embodiment of the present invention includes records of 93 learners answering 36 C language exercises in sequence, with a total of 3348 learning data, and these 36 C language exercises include 12 knowledge points.

[0123] Step 1.2.2: Set the learning resource-answer response matrix R and the learning resource-knowledge point matrix Q of the virtual dataset and the real C language dataset.

[0124] Virtual dataset:

[0125]

[0126] Real C language dataset:

[0127]

[0128] Step 1.3: Based on the learning resource-response matrix and the learning resource-knowledge point matrix, a cognitive diagnostic model is introduced to estimate the error and guess parameters based on the cognitive diagnostic model, thereby increasing the model's interpretability from the perspective of parameter interpretability. Step 1.3 further includes:

[0129] Step 1.3.1: Initialize the learner's knowledge state α, and then combine it with the learning resource-knowledge point matrix Q to obtain the learner's ideal answer situation on the learning resource;

[0130] The ideal answering situation provided by the embodiment of the present invention is that when the learner has mastered all the knowledge points required to answer the learning resource, the answer is correct, otherwise the answer is incorrect. The specific formula is:

[0131]

[0132] Where N represents the number of knowledge points contained in the learning resource

[0133] Step 1.3.2: Formally define the error parameter s and the guess parameter g based on the learning resource-response matrix R and the ideal response matrix η, thereby obtaining a probabilistic model of the learning resource-response matrix R;

[0134] s=P(R=0|η=1)

[0135] g=P(R=1|η=0)

[0136] P(α)=P(R=1|α)=g (1-η) (1-s) η

[0137] Step 1.3.3: Introduce the cognitive diagnosis model DINA and obtain the total likelihood function of the DINA model;

[0138]

[0139] Step 1.3.4: Use the maximum likelihood function to estimate the learner's knowledge state, and use the EM algorithm to estimate the error parameter and guess parameter, thereby obtaining the error parameter s and guess parameter g based on the cognitive diagnosis model. The algorithm process of estimating the error parameter and guess parameter using the EM algorithm is shown in Table 1.

[0140] Table 1 Algorithm process of estimating error parameter s and guess parameter g using EM algorithm

[0141]

[0142] Step 1.3.5: Error parameters and guess parameters are incorporated to better simulate the learner's actual learning situation and increase the interpretability of the model at the parameter level.

[0143] (2) Create a key matrix that stores the potential knowledge points of the learning resources answered by the learners and a value matrix that stores the mastery status of these potential knowledge points. Then obtain the knowledge points related to the learning resources answered by the learners from the key matrix, and read the learners' mastery status on the relevant knowledge points from the value matrix, thereby obtaining a feature vector containing the learners' overall mastery status of the potential knowledge points. Specifically, it includes:

[0144] Step 2.1: Create a key matrix that stores the potential knowledge points of the learning resources answered by the learners and a value matrix that stores the mastery status of these potential knowledge points;

[0145] The key matrix provided by the embodiment of the present invention is It uses N key slots to store potential knowledge points of learning resources;

[0146] The value matrix provided by the embodiment of the present invention is KC v =(KC v (1),...,KC v (i),...,KC v (N)), which contains N value slots, dynamically storing the learner's mastery status of these knowledge points.

[0147] Step 2.2: Represent the embedding matrix of the learning resource sequence as vector k t and value matrix Perform inner product and activate it through sigmoid function to obtain the knowledge points and their weights related to the learning resources answered by the learner;

[0148] w t =σ(k t KC k )

[0149] Step 2.3: Combine the knowledge points and their weights related to the learning resources answered by the learners, and get the value matrix KC v =(KC v (1),...,KC v (i),...,KC v (N)) Read the learner's mastery of relevant knowledge points;

[0150] The learners' mastery of relevant knowledge points is as follows:

[0151] Step 2.4: Take the weighted sum of the learners’ mastery of relevant knowledge points to obtain the overall mastery status of the knowledge points contained in the learning resource;

[0152]

[0153] Step 2.5: Combine the overall mastery status of the knowledge points contained in the learning resource with the embedding matrix vector representation k of the learning resource sequence t The connections are then activated through the Tanh function to obtain a feature vector containing the learner's overall mastery of the potential knowledge points.

[0154]

[0155] (3) Learn the learner's personalized error parameters and guess parameters from the feature vector. Then, based on cognitive diagnosis theory, integrate the error and guess parameters based on the cognitive diagnosis model to obtain a new feature vector, thereby predicting the learner's answer performance on a specific learning resource. On the one hand, the model's interpretability is increased from the parameter interpretability level, and on the other hand, the model's interpretability is increased from the prediction result interpretability level. Specifically, the following are some of the methods:

[0156] Step 3.1: Learn learner-based personalized error parameters and guess parameters from the feature vector;

[0157] s′ t =tanh(W1F t +b1)

[0158] g′ t =tanh(W2F t +b2)

[0159] Step 3.2: Based on cognitive diagnosis theory, the learner-based personalized error parameter s' and guess parameter g' are integrated with the error parameter s and guess parameter g based on the cognitive diagnosis model to obtain a new feature vector.

[0160]

[0161]

[0162] Step 3.3: Based on the obtained new feature vector, activate it through the sigmoid function to predict the learner's answer performance on the specific learning resource;

[0163]

[0164] Step 3.4: Incorporating the error and guess parameters estimated by the cognitive diagnosis model DINA and learned by the neural network, while ensuring the prediction performance of the deep knowledge tracking model, referring to cognitive diagnosis theory, the model prediction results are explained from the two aspects of learners' answering errors and guesses.

[0165] (4) Based on the embedding matrix vector representation of the interaction sequence between the learner and the learning resource, and considering the learner's current knowledge state, the knowledge growth vector is obtained. Then, considering the learner's personalized forgetting rate, the learner's forgetting degree of the knowledge point and the degree of memory reinforcement are obtained, thereby comprehensively updating the learner's knowledge state after the interaction with the learning resource, and then increasing the interpretability of the model from the perspective of knowledge state update interpretability, specifically including:

[0166] Step 4.1: Considering the current knowledge state of the learner, the new feature vector is obtained and the embedding matrix vector representation of the interaction sequence between the learner and the learning resource is V = (v1, v2, ..., v t ) to fuse, thereby obtaining the learner’s knowledge growth vector after answering the learning resources;

[0167] v′ t =[v t ,F t ]

[0168] Step 4.2: Get the knowledge growth vectors of all knowledge points contained in the learning resource and calculate them according to the value matrix KC v =(KC v (1),...,KC v (i),...,KC v (N)) value slots are divided, so that the knowledge growth vectors of all knowledge points contained in the learning resource are distributed in all memory fragments.

[0169] The knowledge growth vector of all knowledge points contained in the acquired learning resources provided by the embodiment of the present invention is w t v′ t .

[0170] The knowledge growth vector after segmentation provided by the embodiment of the present invention is (w t v′ t (1),...,w t v′ t (i),...,w t v′ t (N)).

[0171] Step 4.3: Use each value slot distribution segment as input to the LSTM model to obtain the learner's personalized forgetting rate;

[0172] The learner-specific forgetting rate obtained in a certain value slot provided by the embodiment of the present invention is LSTM(w t (i)v′ t ).

[0173] Step 4.4: Based on the learner's personalized forgetting rate, obtain the learner's forgetting degree of the knowledge point after completing the learning resource, providing a certain degree of explanation for the learner's knowledge status update;

[0174]

[0175] Step 4.4 provided in this embodiment of the present invention further includes:

[0176] Step 4.4.1: Input a fragment of a value slot distribution into the LSTM and determine the discarded information;

[0177] f t (i)=σ(W f ·[h t-1 ,w(i)v']+b f )

[0178] Step 4.4.2: Determine the information to be updated;

[0179] m t (i)=σ(W m ·[h t-1 ,w(i)v']+b m )

[0180] Step 4.4.3: Determine the information to be updated;

[0181]

[0182] Step 4.4.4: Update the status based on the discarded information and the information to be updated;

[0183]

[0184] Step 4.4.5: Output the learner's personalized forgetting rate.

[0185] g t (i)=σ(W g ·[h t-1 ,w(i)v']+b g )

[0186] h t (i) = g t (i)·tanh(U t )

[0187] Step 4.5: Obtain the learner's memory enhancement level after completing the learning resource;

[0188]

[0189] Step 4.6: Comprehensively update the learner's knowledge status after interacting with the learning resource based on the learner's degree of forgetting and memory reinforcement after completing the learning resource.

[0190]

[0191] Step 4.7: Define the loss function and use Adma optimizer to update the weights.

[0192] The loss function is the binary cross entropy between the true output and the predicted output, as follows:

[0193]

[0194] Step 4.8: Add regularization terms l1 and l2 to the loss function to make the prediction results of the knowledge cognition level mining model more stable.

[0195] L'=L+λ1l1+λ2l2

[0196] Among them, λ1 and λ2 are the coefficients corresponding to the regularization terms.

[0197] The interpretable knowledge cognitive level mining system integrating learning cognitive elements provided by the embodiment of the present invention includes:

[0198] The learning resource-answer response matrix and the learning resource-knowledge point matrix construction module are used to construct the learning resource-answer response matrix and the learning resource-knowledge point matrix based on the interaction sequence between learners and learning resources, and estimate the error and guess parameters based on the cognitive diagnosis model;

[0199] A feature vector extraction module is configured to create a key matrix storing the potential knowledge points of the learning resource answered by the learner and a value matrix storing the mastery status of the potential knowledge points; simultaneously obtain the knowledge points related to the learning resource answered by the learner from the key matrix; read the learner's mastery status of the relevant knowledge points from the value matrix; and obtain a feature vector containing the learner's overall mastery status of the potential knowledge points;

[0200] A response performance prediction module is configured to learn individualized error parameters and guess parameters based on the learner from the feature vector containing the learner's overall mastery of potential knowledge points, incorporate the error and guess parameters based on the cognitive diagnostic model, and obtain a new feature vector to predict the learner's response performance on a specific learning resource;

[0201] The knowledge state update module is used to obtain the knowledge growth vector based on the embedding matrix vector representation of the interaction sequence between the learner and the learning resource and the learner's current knowledge state. It then obtains the learner's degree of forgetting of knowledge points and the degree of memory reinforcement based on the learner's personalized forgetting rate, and comprehensively updates the learner's knowledge state after interaction with the learning resource.

[0202] The symbols involved in the embodiments of the present invention are as follows:

[0203]

[0204]

[0205] 2. Application Examples: In order to demonstrate the creativity and technical value of the technical solution of the present invention, this section provides application examples of the claimed technical solution on specific products or related technologies.

[0206] The method and system for mining the cognitive level of interpretable knowledge that integrates learning cognitive elements specifically include:

[0207] (1) Based on the interaction sequence between learners and learning resources, we make embedding matrix vector representations of the learning resource sequence and the interaction sequence between learners and learning resources, and construct a learning resource-answer response matrix and a learning resource-knowledge point matrix to estimate the error and guess parameters based on the cognitive diagnosis model, thereby increasing the interpretability of the model from the perspective of parameter interpretability;

[0208] (2) Create a key matrix that stores the potential knowledge points of the learning resources answered by the learners and a value matrix that stores the mastery status of these potential knowledge points. Then, obtain the knowledge points related to the learning resources answered by the learners from the key matrix, and read the learners' mastery status on the relevant knowledge points from the value matrix, thereby obtaining a feature vector containing the learners' overall mastery status of the potential knowledge points.

[0209] (3) Learn the learner's personalized error parameters and guess parameters from the feature vector, and then, based on cognitive diagnosis theory, integrate the error and guess parameters based on the cognitive diagnosis model to obtain a new feature vector, thereby predicting the learner's answer performance on a specific learning resource. This increases the model's interpretability from the perspective of parameter interpretability and the perspective of prediction result interpretability.

[0210] (4) Based on the embedding matrix vector representation of the interaction sequence between the learner and the learning resource, and considering the learner's current knowledge state, the knowledge growth vector is obtained. Then, considering the learner's personalized forgetting rate, the learner's forgetting degree of the knowledge points and the degree of memory reinforcement are obtained, thereby comprehensively updating the learner's knowledge state after the interaction with the learning resource, and then increasing the interpretability of the model from the level of knowledge state update interpretability.

[0211] As a preferred embodiment of the present invention, based on the interaction sequence between learners and learning resources, an embedding matrix-vector representation of the learning resource sequence and the interaction sequence between learners and learning resources is made, and a learning resource-answer response matrix and a learning resource-knowledge point matrix are constructed to estimate the error and guess parameters based on the cognitive diagnosis model, thereby increasing the interpretability of the model from the parameter interpretability level, specifically including:

[0212] Step 1.1: Based on the interaction sequence between the learner and the learning resource, make an embedding matrix vector representation of the learning resource sequence and an embedding matrix vector representation of the interaction sequence between the learner and the learning resource. Step 1.1 further includes:

[0213] Step 1.1.1: Set the embedding matrix A of the learning resource and transform the learning resource sequence (q1,q2,...,q t ) is embedded with the matrix to obtain the embedding matrix vector representation of the learning resource sequence K=(k1,k2,...,k t );

[0214] Step 1.1.2: Set the embedding matrix B of the interaction sequence between the learner and the learning resource, and transform the interaction sequence between the learner and the learning resource X = (x1, x2, ..., x t ) is embedded with the matrix to obtain the embedding matrix vector representation of the interaction sequence between the learner and the learning resource V=(v1,v2,...,v t );

[0215] Step 1.2: Based on the interaction sequence between learners and learning resources, construct a learning resource-response matrix and a learning resource-knowledge point matrix;

[0216] The learning resource-answer response matrix is ​​the learning resource answered by the behavioral learner, and is listed as the matrix R of the learner's answer response on the learning resource;

[0217] The learning resource-knowledge point matrix is ​​the learning resource answered by the behavioral learner, and is listed as a matrix Q of knowledge points corresponding to the learning resource.

[0218] Step 1.2 further includes:

[0219] Step 1.2.1: There are two data sets used in the experiment of this embodiment, including: a virtual data set and a real C language data set.

[0220] The virtual dataset contains records of 2,000 learners answering questions on 50 learning resources in sequence, with a total of 100,000 learning data. The 50 learning resources are set to contain 5 knowledge points.

[0221] The real C language dataset contains records of 93 learners answering 36 C language exercises in sequence, with a total of 3348 learning data, and these 36 C language exercises contain 12 knowledge points.

[0222] Step 1.2.2: Set the learning resource-answer response matrix R and the learning resource-knowledge point matrix Q of the virtual dataset and the real C language dataset.

[0223] Virtual dataset:

[0224]

[0225] Real C language dataset:

[0226]

[0227] Step 1.3: Based on the learning resource-response matrix and the learning resource-knowledge point matrix, a cognitive diagnostic model is introduced to estimate the error and guess parameters based on the cognitive diagnostic model, thereby increasing the model's interpretability from the perspective of parameter interpretability. Step 1.3 further includes:

[0228] Step 1.3.1: Initialize the learner's knowledge state α, and then combine it with the learning resource-knowledge point matrix Q to obtain the learner's ideal answer situation on the learning resource;

[0229] The ideal answer is that if the learner has mastered all the knowledge points required to answer the learning resource, the answer is correct; otherwise, the answer is incorrect. The specific formula is:

[0230]

[0231] Where N represents the number of knowledge points contained in the learning resource

[0232] Step 1.3.2: Formally define the error parameter s and the guess parameter g based on the learning resource-response matrix R and the ideal response matrix η, thereby obtaining a probabilistic model of the learning resource-response matrix R;

[0233] s=P(R=0|η=1)

[0234] g=P(R=1|η=0)

[0235] P(α)=P(R=1|α)=g (1-η) (1-s) η

[0236] Step 1.3.3: Introduce the cognitive diagnosis model DINA and obtain the total likelihood function of the DINA model;

[0237]

[0238] Step 1.3.4: Use the maximum likelihood function to estimate the learner's knowledge state and the EM algorithm to estimate the error parameter and guess parameter, thereby obtaining the error parameter s and guess parameter g based on the cognitive diagnosis model. The algorithm process of estimating the error parameter and guess parameter using the EM algorithm is shown in Table 2.

[0239] Table 2 Algorithm process of estimating error parameter s and guess parameter g using EM algorithm

[0240]

[0241]

[0242] Step 1.3.5: Error parameters and guess parameters are incorporated to better simulate the learner's actual learning situation and increase the interpretability of the model at the parameter level.

[0243] As a preferred embodiment of the present invention, a key matrix storing the potential knowledge points of the learning resource answered by the learner and a value matrix storing the mastery status of these potential knowledge points are created respectively, and the knowledge points related to the learning resource answered by the learner are obtained from the key matrix, and the mastery status of the learner on the relevant knowledge points is read from the value matrix, thereby obtaining a feature vector containing the learner's overall mastery status of the potential knowledge points, specifically including:

[0244] Step 2.1: Create a key matrix that stores the potential knowledge points of the learning resources answered by the learners and a value matrix that stores the mastery status of these potential knowledge points;

[0245] The bond matrix is It uses N key slots to store potential knowledge points of learning resources;

[0246] The value matrix is ​​KC v =(KC v (1),...,KC v (i),...,KC v (N)), which contains N value slots, dynamically storing the learner's mastery status of these knowledge points.

[0247] Step 2.2: Represent the embedding matrix of the learning resource sequence as vector k t and value matrix Perform inner product and activate it through sigmoid function to obtain the knowledge points and their weights related to the learning resources answered by the learner;

[0248] w t =σ(k t KC k )

[0249] Step 2.3: Combine the knowledge points and their weights related to the learning resources answered by the learners, and get the value matrix KC v =(KC v (1),...,KC v (i),...,KC v(N)) Read the learner's mastery of relevant knowledge points;

[0250] The learners' mastery of relevant knowledge points is as follows:

[0251] Step 2.4: Take the weighted sum of the learners’ mastery of relevant knowledge points to obtain the overall mastery status of the knowledge points contained in the learning resource;

[0252]

[0253] Step 2.5: Combine the overall mastery status of the knowledge points contained in the learning resource with the embedding matrix vector representation k of the learning resource sequence t The connections are then activated through the Tanh function to obtain a feature vector containing the learner's overall mastery of the potential knowledge points.

[0254] F t =Tanh(W1 T [r t ,k t ]+b1)

[0255] As a preferred embodiment of the present invention, learners' personalized error parameters and guess parameters from feature vectors. Then, based on cognitive diagnosis theory, the error and guess parameters based on the cognitive diagnosis model are integrated to obtain a new feature vector. This allows prediction of learners' performance on specific learning resources. This increases the model's interpretability both in terms of parameter interpretability and in terms of prediction result interpretability. Specifically, the following methods are used:

[0256] Step 3.1: Learn learner-based personalized error parameters and guess parameters from the feature vector;

[0257] s′ t =tanh(W1F t +b1)

[0258] g′ t =tanh(W2F t +b2)

[0259] Step 3.2: Based on cognitive diagnosis theory, the learner-based personalized error parameter s' and guess parameter g' are integrated with the error parameter s and guess parameter g based on the cognitive diagnosis model to obtain a new feature vector.

[0260]

[0261]

[0262] Step 3.3: Based on the obtained new feature vector, activate it through the sigmoid function to predict the learner's answer performance on the specific learning resource;

[0263]

[0264] Step 3.4: Incorporating the error and guess parameters estimated by the cognitive diagnosis model DINA and learned by the neural network, while ensuring the prediction performance of the deep knowledge tracking model, referring to cognitive diagnosis theory, the model prediction results are explained from the two aspects of learners' answering errors and guesses.

[0265] As a preferred embodiment of the present invention, based on the embedding matrix vector representation of the interaction sequence between the learner and the learning resource, and taking into account the learner's current knowledge state, a knowledge growth vector is obtained. Then, considering the learner's personalized forgetting rate, the learner's forgetting degree and memory reinforcement degree of the knowledge points are obtained, thereby comprehensively updating the learner's knowledge state after the interaction with the learning resource, and further increasing the model's interpretability from the perspective of knowledge state update interpretability. Specifically, the following are included:

[0266] Step 4.1: Considering the current knowledge state of the learner, the new feature vector is obtained and the embedding matrix vector representation of the interaction sequence between the learner and the learning resource is V = (v1, v2, ..., v t ) to fuse, thereby obtaining the learner’s knowledge growth vector after answering the learning resources;

[0267] v′ t =[v t ,F t ]

[0268] Step 4.2: Get the knowledge growth vectors of all knowledge points contained in the learning resource and calculate them according to the value matrix KC v =(KC v (1),...,KC v (i),...,KC v (N)) value slots are divided, thereby distributing the knowledge growth vectors of all knowledge points contained in the learning resource in all memory fragments.

[0269] The knowledge growth vector of all knowledge points contained in the acquired learning resources is w t v′ t .

[0270] The knowledge growth vector after segmentation is (w t v′ t (1),...,w t v′ t (i),...,wt v′ t (N)).

[0271] Step 4.3: Use each value slot distribution segment as input to the LSTM model to obtain the learner's personalized forgetting rate;

[0272] The learner-specific forgetting rate obtained in a certain value slot is LSTM(w t (i)v′ t ).

[0273] Step 4.4: Based on the learner's personalized forgetting rate, obtain the learner's forgetting degree of the knowledge point after completing the learning resource, providing a certain degree of explanation for the learner's knowledge status update;

[0274]

[0275] Step 4.4 further includes:

[0276] Step 4.4.1: Input a fragment of a value slot distribution into the LSTM and determine the discarded information;

[0277] f t (i)=σ(W f ·[h t-1 ,w(i)v']+b f )

[0278] Step 4.4.2: Determine the information to be updated;

[0279] m t (i)=σ(W m ·[h t-1 ,w(i)v']+b m )

[0280] Step 4.4.3: Determine the information to be updated;

[0281]

[0282] Step 4.4.4: Update the status based on the discarded information and the information to be updated;

[0283]

[0284] Step 4.4.5: Output the learner's personalized forgetting rate.

[0285] g t (i)=σ(W g ·[h t-1 ,w(i)v']+b g )

[0286] h t (i) = g t (i)·tanh(U t )

[0287] Step 4.5: Obtain the learner's memory enhancement level after completing the learning resource;

[0288]

[0289] Step 4.6: Comprehensively update the learner's knowledge status after interacting with the learning resource based on the learner's degree of forgetting and memory reinforcement after completing the learning resource.

[0290]

[0291] Step 4.7: Define the loss function and use Adma optimizer to update the weights.

[0292] The loss function is the binary cross entropy between the true output and the predicted output, as follows:

[0293]

[0294] Step 4.8: Add regularization terms l1 and l2 to the loss function to make the prediction results of the knowledge cognition level mining model more stable.

[0295] L'=L+λ1l1+λ2l2

[0296] Among them, λ1 and λ2 are the coefficients corresponding to the regularization terms.

[0297] As a preferred embodiment of the present invention, the main development environment during the experimental process of the present invention is Windows 10, AMD Ryzen Threadripper 2950X 16-Core Processor, 32RAM, GTX 1080Ti graphics, TensorFlow 1.9, Python 3.5. The experimental hyperparameters during training are shown in Table 3.

[0298] Table 3 Experimental parameters during training

[0299]

[0300] We use AUC and RMSE as metrics to measure model prediction performance. AUC is a reliable metric often used in classification problems. Correct responses are considered positive examples, while incorrect responses are considered negative examples. Higher AUC values ​​indicate more accurate predictions. RMSE measures the deviation between the actual results and the predicted results. Lower RMSE values ​​indicate better prediction performance.

[0301] We compared this method with traditional knowledge cognitive level mining methods. To make a fair comparison, all of these methods were tuned to have the best performance. The AUC and RMSE comparison results of an interpretable knowledge cognitive level mining method that integrates learning cognitive elements and traditional knowledge cognitive level mining methods on virtual datasets and real C language datasets are shown in Table 4. The model training process on virtual datasets and real C language datasets is shown in Figure 4. Figure 4 shown.

[0302] Table 4 Comparison of experimental results on virtual datasets and real C language datasets

[0303]

[0304] The experimental results show that the interpretable knowledge cognitive level mining method that integrates learning cognitive elements proposed in the present invention is more effective than the traditional knowledge cognitive level mining method. On the virtual data set, the AUC of the traditional knowledge cognitive level mining method that considers the current knowledge status of the learner is improved by about 0.8% compared with the traditional knowledge cognitive level mining method, and the RMSE is reduced by about 0.003. The experiment on the real C language data set also shows that the AUC of the traditional knowledge cognitive level mining method that considers the current knowledge status of the learner is higher than the traditional knowledge cognitive level mining method, and the RMSE is lower. This shows that the traditional knowledge cognitive level mining method that considers the current knowledge status of the learner effectively improves the prediction performance of the model by considering the current knowledge status of the learner. The AUC of this method is improved by about 2% compared with the traditional knowledge cognitive level mining method, which shows that the interpretable knowledge cognitive level mining method that integrates learning cognitive elements proposed in this chapter is effective and improves the accuracy of the model in predicting learner performance. Furthermore, we found that incorporating error and guesswork cognitive elements into traditional knowledge cognitive level mining methods, and incorporating error, guesswork, and forgetting rate cognitive elements into traditional knowledge cognitive level mining methods, improved AUC values ​​and decreased RMSE values ​​compared to traditional knowledge cognitive level mining methods. This suggests that mining and incorporating learner cognitive elements from different perspectives has an impact on tracking learners' knowledge state changes during the learning process and predicting their performance. Overall, this method has the best experimental results.

[0305] 3. Evidence of the Effects of the Embodiments: The embodiments of the present invention have achieved some positive effects during development or use, and indeed have significant advantages over the prior art. The following content describes them with reference to data, charts, etc. from the experimental process.

[0306] During the simulation experiments, the main development environment was Windows 10, an AMD Ryzen Threadripper 2950X 16-Core processor, 32 RAM, GTX 1080Ti graphics, TensorFlow 1.9, and Python 3.5. The experimental hyperparameters during training are shown in Table 5.

[0307] Table 5 Experimental parameters during training

[0308] Experimental parameters Virtual Dataset Real C language dataset batch_size 32 32 learning_rate 0.05 0.05 epoch 300 300 memory size 5 5 sequence length 50 36 dropout 0.2 0.2 hidden size(LSTM) 100 100

[0309] The simulation experiment uses AUC and RMSE as metrics to measure the model's predictive performance. AUC is a reliable metric often used in classification problems. Correct responses are considered positive examples, while incorrect responses are considered negative examples. Higher AUC values ​​indicate more accurate predictions. RMSE measures the deviation between actual and predicted results; smaller RMSE values ​​indicate better predictive performance.

[0310] The embodiments of the present invention were compared with traditional knowledge cognitive level mining methods. To ensure a fair comparison, all of these methods were adjusted to have the best performance. The AUC and RMSE comparison results of an interpretable knowledge cognitive level mining method that integrates learning cognitive elements and traditional knowledge cognitive level mining methods on virtual datasets and real C language datasets are shown in Table 3. The model training process on virtual datasets and real C language datasets is shown in Figure 3. Figure 3 、 Figure 4 shown.

[0311] Table 6 Comparison of experimental results on virtual datasets and real C language datasets

[0312]

[0313] The experimental results show that the interpretable knowledge cognitive level mining method that integrates learning cognitive elements proposed in the embodiment of the present invention is more effective than the traditional knowledge cognitive level mining method. On the virtual data set, the AUC of the traditional knowledge cognitive level mining method that considers the current knowledge status of the learner is improved by about 0.8% compared with the traditional knowledge cognitive level mining method, and the RMSE is reduced by about 0.003. The experiment on the real C language data set also shows that the AUC of the traditional knowledge cognitive level mining method that considers the current knowledge status of the learner is higher than the traditional knowledge cognitive level mining method, and the RMSE is lower. This shows that the model that considers the current knowledge status of the learner in the traditional knowledge cognitive level mining method effectively improves the prediction performance of the model by considering the current knowledge status of the learner. The AUC of the embodiment of the present invention is improved by about 2% compared with the traditional knowledge cognitive level mining method, which shows that the interpretable knowledge cognitive level mining method that integrates learning cognitive elements proposed in the embodiment of the present invention is effective and improves the accuracy of the model in predicting learner performance. In addition, the present invention also found that the integration of error and guessing cognitive elements into traditional knowledge cognitive level mining methods and the integration of error, guessing, and forgetting rate cognitive elements into traditional knowledge cognitive level mining methods improved the AUC value and decreased the RMSE value compared with traditional knowledge cognitive level mining methods. This shows that mining and integrating learner cognitive elements from different perspectives has an impact on tracking learners' knowledge state changes during the learning process and predicting their performance. In short, the interpretable knowledge cognitive level mining method that integrates learning cognitive elements provided by the present invention has the best experimental results.

[0314] It should be noted that the embodiments of the present invention can be implemented by hardware, software, or a combination of software and hardware. The hardware portion can be implemented using dedicated logic; the software portion can be stored in a memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated design hardware. Those skilled in the art will appreciate that the above-mentioned devices and methods can be implemented using computer-executable instructions and / or contained in processor control code, for example, such as a carrier medium such as a disk, CD or DVD-ROM, a programmable memory such as a read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The devices and modules of the present invention can be implemented by hardware circuits such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc., can also be implemented by software executed by various types of processors, or can be implemented by a combination of the above-mentioned hardware circuits and software, such as firmware.

[0315] The above description is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with this technical field within the technical scope disclosed by the present invention and within the spirit and principles of the present invention should be covered by the scope of protection of the present invention.

Claims

1. A method for mining the cognitive level of interpretable knowledge that integrates learning cognitive elements, characterized by: The method for mining the explainable knowledge cognitive level of the integrated learning cognitive elements includes: Step 1: Based on the interaction sequence between learners and learning resources, a learning resource-response matrix and a learning resource-knowledge point matrix are constructed to estimate the error and guess parameters based on the cognitive diagnosis model. Step 2: Create a key matrix to store the potential knowledge points of the learning resource answered by the learner and a value matrix to store the mastery status of the potential knowledge points; obtain the knowledge points related to the learning resource answered by the learner from the key matrix; read the learner's mastery status on the relevant knowledge points from the value matrix to obtain a feature vector containing the learner's overall mastery status of the potential knowledge points; Step 3: learning the learner's personalized error parameters and guess parameters from the feature vector containing the learner's overall mastery of the potential knowledge points, integrating the error and guess parameters based on the cognitive diagnostic model to obtain a new feature vector to predict the learner's answer performance on a specific learning resource; Step 4: Based on the embedding matrix vector representation of the interaction sequence between the learner and the learning resource and the learner's current knowledge state, the knowledge growth vector is obtained. Then, based on the learner's personalized forgetting rate, the learner's forgetting degree and memory reinforcement degree of the knowledge points are obtained, and the knowledge state after the learner interacts with the learning resource is comprehensively updated. The second step includes: 1) Create a key matrix to store the potential knowledge points of the learning resources answered by the learner and a value matrix to store the mastery status of the potential knowledge points; the key matrix is The value matrix is ​​KC v =(KC v (1),...,KC v (i),...,KC v (N)); 2) The embedding matrix vector of the learning resource sequence is represented by k t and value matrix Perform inner product and activate it through sigmoid function to obtain the knowledge points related to the learning resources answered by the learner and the weights of the knowledge points related to the learning resources answered by the learner; The weight w of the knowledge points related to the learning resources answered by the learner t The calculation formula is: t =σ(k t KC k ); 3) Combine the acquired knowledge points related to the learning resources answered by the learners and the weights w of the knowledge points related to the learning resources answered by the learners t , from the value matrix KC v =(KC v (1),...,KC v (i),...,KC v (N)) Read the learner's mastery of relevant knowledge points 4) Perform weighted summation on the learner's mastery of relevant knowledge points to obtain the overall mastery status of the knowledge points contained in the learning resource; The overall mastery status of the knowledge points contained in the learning resource t The calculation formula is as follows: 5) The overall mastery status of the knowledge points contained in the learning resources and the embedding matrix vector representation k of the learning resource sequence t Connect and activate through Tanh function to obtain the feature vector containing the learner's overall mastery of the potential knowledge points; The feature vector F containing the learner's overall mastery of the potential knowledge points t It is expressed as follows: F t =Fish(W1 T [r t ,k t ]+b1)。 2. The method for mining the cognitive level of interpretable knowledge integrating learning cognitive elements as claimed in claim 1, characterized in that: The step one comprises: (1) Based on the interaction sequence between learners and learning resources, we make an embedding matrix vector representation of the learning resource sequence and an embedding matrix vector representation of the interaction sequence between learners and learning resources: (1.1) Set the embedding matrix A of the learning resource and transform the learning resource sequence (q1,q2,...,q t ) is embedded with the embedding matrix A of the learning resource to obtain the embedding matrix vector representation K=(k1,k2,...,k t ); (1.2) Set the embedding matrix B of the interaction sequence between learners and learning resources, and transform the interaction sequence between learners and learning resources X = (x1, x2, ..., x t ) is embedded with the embedding matrix B of the interaction sequence between the learner and the learning resource to obtain the embedding matrix vector representation V = (v1, v2, ..., v t ); (2) Based on the interaction sequence between learners and learning resources, a learning resource-answer response matrix and a learning resource-knowledge point matrix are constructed; the learning resource-answer response matrix is ​​a matrix of learning resources answered by behavioral learners, and is a matrix of learners' answer responses on learning resources; the learning resource-knowledge point matrix is ​​a matrix of learning resources answered by behavioral learners, and is a matrix of knowledge points corresponding to learning resources; (3) Based on the learning resource-response matrix and the learning resource-knowledge point matrix, a cognitive diagnosis model is introduced to estimate the error and guess parameters based on the cognitive diagnosis model: (3.1) Initialize the learner's knowledge state α, and then combine it with the learning resource-knowledge point matrix Q to obtain the learner's ideal answer situation on the learning resource. The learner's ideal answer situation η on the learning resource is calculated as follows: Where N represents the number of knowledge points contained in the learning resource; (3.2) Based on the learning resource-answer response matrix R and the ideal answer matrix η, we formally define the error parameter s and the guess parameter g, and obtain the probability model of the learning resource-answer response matrix R as follows: s=P(R=0|η=1) g=P(R=1|η=0) P(α)=P(R=1|α)=g (1-η) ·(1-s) η (3.3) Introducing the cognitive diagnosis model DINA, we obtain the total likelihood function of the DINA model: (3.4) Use the maximum likelihood function to estimate the learner's knowledge state, use the EM algorithm to estimate the error parameter and guess parameter, and obtain the error parameter s and guess parameter g based on the cognitive diagnosis model.

3. The method for mining the cognitive level of interpretable knowledge integrating learning cognitive elements as claimed in claim 1, characterized in that: The step three includes: First, learn the learner-based personalized error parameters and guess parameters from the feature vector: S' t =tanh(W1F t +b1) g′ t =tanh(W2F t +b2) Secondly, the learner-based personalized error parameter s' and guess parameter g' as well as the error parameter s and guess parameter g based on the cognitive diagnosis model are incorporated to obtain a new feature vector: Finally, the new feature vector is activated through the sigmoid function to predict the learner's answer performance on a specific learning resource; The predicted learner's answer performance P on a specific learning resource t The formula is as follows: P t =σ(W3 T F t ”+b3)。 4. The method for mining the cognitive level of interpretable knowledge integrating learning cognitive elements as claimed in claim 1, characterized in that: The fourth step includes: (1) Combining the learner’s current knowledge state, the new feature vector and the embedding matrix vector representation of the interaction sequence between the learner and the learning resource are obtained. V = (v1, v2, ..., v t ) to obtain the learner’s knowledge growth vector after answering the learning resources: v′ t =[v t ,F t ]; (2) Obtain the knowledge growth vectors of all knowledge points contained in the learning resource, and distribute the knowledge growth vectors of all knowledge points contained in the learning resource in all memory fragments: The knowledge growth vectors of all knowledge points contained in the learning resource are obtained as w t v′ t , the knowledge growth vector is calculated according to the value matrix KC v =(KC v (1),...,KC v (i),...,KC v (N)) value slot is split to obtain the knowledge growth vector (w t v′ t (1),...,w t v′ t (i),...,w t v′ t (N)); (3) Take each value slot distribution fragment as the input of the LSTM model to obtain the learner's personalized forgetting rate LSTM (w t (i)v′ t ); (4) Based on the learner's personalized forgetting rate, obtain the learner's forgetting degree of the knowledge point after completing the learning resource: (5) Obtain the memory enhancement degree α after the learner finishes answering the learning resources t :α t =Tanh(W3 T v′ t +b3); (6) Comprehensively update the knowledge status of learners after interacting with learning resources based on the degree of forgetting and memory reinforcement after learners finish answering the learning resources: (7) Define the loss function and use the Adma optimizer to update the weights; the loss function is as follows: (8) Add regularization terms l1 and l2 to the loss function: L' = L + λ1l1 + λ2l2; where λ1 and λ2 are the coefficients corresponding to the regularization terms.

5. The method for mining the cognitive level of interpretable knowledge integrating learning cognitive elements as claimed in claim 4, characterized in that: In step (4), based on the learner's personalized forgetting rate, obtaining the learner's forgetting degree of the knowledge point after completing the learning resource includes: (4.1) Input any fragment of the value slot distribution into the LSTM and determine the discarded information: f t (i)=σ(W f ·[h t-1 ,w(i)v']+b f ); where W represents the weight coefficient; b represents the corresponding bias vector; (4.2) Determine the information to be updated: m t (i)=σ(W m ·[h t-1 ,w(i)v']+b m ); (4.3) Determine the information that needs to be updated: (4.4) Update the status based on the discarded information and the information to be updated: (4.5) Obtain the learner’s personalized forgetting rate: Among them, h t (i) Indicates the degree of forgetting acquired by the memory fragment of a certain value slot.

6. A system for mining the explainable knowledge and cognitive level of integrated learning and cognitive elements, which implements the method for mining the explainable knowledge and cognitive level of integrated learning and cognitive elements as described in any one of claims 1 to 5, characterized in that: The interpretable knowledge cognitive level mining system integrating learning cognitive elements includes: The learning resource-answer response matrix and the learning resource-knowledge point matrix construction module are used to construct the learning resource-answer response matrix and the learning resource-knowledge point matrix based on the interaction sequence between learners and learning resources, and estimate the error and guess parameters based on the cognitive diagnosis model; A feature vector extraction module is configured to create a key matrix storing the potential knowledge points of the learning resource answered by the learner and a value matrix storing the mastery status of the potential knowledge points; simultaneously obtain the knowledge points related to the learning resource answered by the learner from the key matrix; read the learner's mastery status of the relevant knowledge points from the value matrix; and obtain a feature vector containing the learner's overall mastery status of the potential knowledge points; An answer performance prediction module is used to learn the learner's personalized error parameters and guess parameters from the feature vector containing the learner's overall mastery of potential knowledge points, integrate the error and guess parameters based on the cognitive diagnosis model, and obtain a new feature vector to predict the learner's answer performance on a specific learning resource; The knowledge status update module is used to obtain the knowledge growth vector based on the embedded matrix vector representation of the interaction sequence between the learner and the learning resource and the learner's current knowledge status; then, based on the learner's personalized forgetting rate, it obtains the learner's degree of forgetting of the knowledge points and the degree of memory reinforcement, and comprehensively updates the learner's knowledge status after the interaction with the learning resource.

7. A computer device, characterized in that: The computer device includes a memory and a processor, wherein the memory stores a computer program. When the computer program is executed by the processor, the processor performs the following steps of the method for mining the explainable knowledge cognitive level by integrating learning cognitive elements according to any one of claims 1 to 5: Step 1: Based on the interaction sequence between learners and learning resources, a learning resource-response matrix and a learning resource-knowledge point matrix are constructed to estimate the error and guess parameters based on the cognitive diagnosis model. Step 2: Create a key matrix to store the potential knowledge points of the learning resource answered by the learner and a value matrix to store the mastery status of the potential knowledge points; obtain the knowledge points related to the learning resource answered by the learner from the key matrix; read the learner's mastery status on the relevant knowledge points from the value matrix to obtain a feature vector containing the learner's overall mastery status of the potential knowledge points; Step 3: learning the learner's personalized error parameters and guess parameters from the feature vector containing the learner's overall mastery of the potential knowledge points, integrating the error and guess parameters based on the cognitive diagnostic model to obtain a new feature vector to predict the learner's answer performance on a specific learning resource; Step 4: Based on the embedding matrix vector representation of the interaction sequence between the learner and the learning resource, and based on the learner's current knowledge status, the knowledge growth vector is obtained; then, based on the learner's personalized forgetting rate, the learner's degree of forgetting of the knowledge points and the degree of memory reinforcement are obtained, and the knowledge status of the learner after the interaction with the learning resource is comprehensively updated.

8. A computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor executes the following steps of the method for mining the cognitive level of explainable knowledge integrating learning cognitive elements according to any one of claims 1 to 5: Step 1: Based on the interaction sequence between learners and learning resources, a learning resource-response matrix and a learning resource-knowledge point matrix are constructed to estimate the error and guess parameters based on the cognitive diagnosis model. Step 2: Create a key matrix to store the potential knowledge points of the learning resource answered by the learner and a value matrix to store the mastery status of the potential knowledge points; obtain the knowledge points related to the learning resource answered by the learner from the key matrix; read the learner's mastery status on the relevant knowledge points from the value matrix to obtain a feature vector containing the learner's overall mastery status of the potential knowledge points; Step 3: learning the learner's personalized error parameters and guess parameters from the feature vector containing the learner's overall mastery of the potential knowledge points, integrating the error and guess parameters based on the cognitive diagnostic model to obtain a new feature vector to predict the learner's answer performance on a specific learning resource; Step 4: Based on the embedding matrix vector representation of the interaction sequence between the learner and the learning resource, and based on the learner's current knowledge status, the knowledge growth vector is obtained; then, based on the learner's personalized forgetting rate, the learner's degree of forgetting of the knowledge points and the degree of memory reinforcement are obtained, and the knowledge status of the learner after the interaction with the learning resource is comprehensively updated.

9. An information data processing terminal, characterized in that: The information data processing terminal is used to implement the interpretable knowledge cognitive level mining system that integrates learning cognitive elements as described in claim 6.

Images