An interpretable knowledge tracing method, an online learning platform, and an electronic device

Abstract

The present invention discloses an interpretable knowledge tracing method, an online learning platform and an electronic device, belonging to the technical fields of data mining and interpretable artificial intelligence. The method steps include: problem definition and data preprocessing, problem representation, knowledge acquisition, knowledge application, knowledge update and model training. Vectorize the preprocessed data, fuse the knowledge states of the learner at different times through a multi-layer perceptron, construct an incremental index and minimize the loss value, dynamically update the knowledge state of the learner, and generate a personalized teaching strategy. The present invention deconstructs the process of the learner solving a specific problem, models the knowledge acquisition degree of the learner before solving the problem, calculates the knowledge application degree of the learner when solving the problem, and measures the knowledge update degree of the learner after solving the problem, realizing interpretability for the problem-solving process, and helping educators and learners to master their own teaching and learning status in a timely manner.

Description

Technical Field

[0001] The present invention relates to a knowledge tracing method, an online learning platform, and an electronic device, and particularly to an interpretable knowledge tracing method, an online learning platform, and an electronic device, belonging to the technical fields of data mining and interpretable artificial intelligence. Background Art

[0002] There is a knowledge reversal phenomenon in existing educational technologies: teaching procedures that are effective for beginners are ineffective for experts. Therefore, as educators, they must be able to continuously evaluate and monitor learners' knowledge mastery during the learning process to dynamically determine appropriate teaching resources. Especially in the online learning scenario, knowledge tracing is an influential research field in distance education, which can automatically track students' knowledge levels at different stages and has been widely used in adaptive learning systems.

[0003] In recent years, some research scholars have begun to pay attention to the interpretability of knowledge tracing methods. They enhance the interpretability of the model by introducing item response theory, adopting attention mechanisms, or designing educationally meaningful learnable parameters, so as to better explain students' learning performance. However, these methods mainly focus on performance prediction and ignore the learning process, resulting in unreasonable evaluation of knowledge mastery patterns. From a cognitive perspective, learners' knowledge mastery curves are relatively stable and will not fluctuate violently in a short period of time.

[0004] Practice is a typical problem-solving process. Students enhance their understanding of problems through in-depth thinking, continuously explore solutions, and finally achieve knowledge consolidation. The entire problem-solving process involves the acquisition, application, internalization, and forgetting of memory knowledge in the brain. However, existing knowledge tracing methods ignore the process of learners solving problems and equate the memorized knowledge with the acquired or applied knowledge, resulting in violent fluctuations in knowledge states. According to problem-solving theory, knowledge acquisition involves learners organizing implicit and explicit knowledge related to problems, while knowledge application involves applying this knowledge to seek solutions. Remembering is not equal to applying, and being familiar with discrete knowledge concepts does not mean that students can effectively organize them when answering questions or convert abstract knowledge into specific knowledge or skills - as the saying goes, knowing the way and walking the way are two different things, which reflects this phenomenon.

[0005] To improve the irrationality of knowledge state assessment, enhance the interpretability of the model, and improve the accuracy of prediction, it is a new idea to deconstruct the learning process of students from the perspective of how humans solve problems. Exploring an interpretable knowledge tracing method for the problem-solving process, by using deep learning technology, deeply analyzing each stage of learners' problem-solving, and analyzing in detail how learners understand problems, acquire knowledge, apply knowledge, internalize knowledge, and forget knowledge, so as to reproduce their learning process, can diagnose and attribute the knowledge state of learners more accurately, which has important research significance and wide application value. Summary of the Invention

[0006] The purpose of the present invention is to provide an interpretable knowledge tracing method, an online learning platform, and an electronic device, which not only provide a knowledge tracing method, but also enhance the interpretability of the knowledge tracing method for the phenomenon of ignoring the problem-solving process of learners in the prior art, and solve the deficiencies existing in the prior art.

[0007] The present invention provides the following solutions:

[0008] An interpretable knowledge tracing method for the problem-solving process, applied to an interpretable knowledge tracing system for the problem-solving process, includes:

[0009] Before solving the problem: obtain the exercise records of the learner, perform data preprocessing on the exercise records to obtain corresponding preprocessed data;

[0010] Perform vectorization processing on the preprocessed data to obtain a problem embedding representation vector P n , combined with the knowledge concepts, difficulty, and discrimination in the exercise records, to obtain an enhanced problem embedding representation vector

[0011] When solving the problem: fuse the knowledge state H of the learner at the previous moment through a multi-layer perceptron n-1 and the enhanced problem embedding representation vector to model the personalized problem space of the learner, where the personalized problem space includes the initial state SP n of the problem, the target state EP n and the adaptation operator OP n , and use an activation function and a gating network to calculate the knowledge acquisition degree vector

[0012] Use a multi-layer perceptron and an activation function to generate a corresponding solution from the adaptation operator of the personalized problem space to obtain a knowledge application degree vector Calculate the probability that the learner correctly answers the question through the four-parameter item response theory, and perform attribution processing on the learner's response;

[0013] Based on the knowledge acquisition degree vector and the knowledge application degree vector an incremental index is constructed to quantify the knowledge internalization degree KI of the learner n , and the forgetting effect of the learner over time is modeled, and the knowledge internalization degree KI n and the forgetting effect act on the knowledge state H at the previous moment n-1 to obtain the updated knowledge state H of the learner at the current moment n ;

[0014] After solving the problem: By using the cross-entropy loss function, calculate the loss value between the probability of the learner correctly answering the question and their actual response, and minimize the loss value by optimizing the model;

[0015] Repeat the above process iteratively to dynamically update the knowledge state of the learner and generate corresponding personalized teaching strategies for the learner.

[0016] Furthermore, obtaining the practice records of the learner, performing data preprocessing on the practice records to obtain corresponding preprocessed data, further including:

[0017] The practice records include the answering time ts n , the answered question p n , the knowledge concept k examined by the question n and the answering result r n , and the data preprocessing includes cleaning, removing duplicates, and deleting missing values.

[0018] Furthermore, performing vectorization processing on the preprocessed data to obtain the question embedding representation vector P n , and combining the knowledge concept, difficulty, and discrimination degree in the practice record to obtain an enhanced question embedding representation vector Further including:

[0019] The enhanced question embedding representation vector takes the knowledge concept embedding vector K n as the main body, and takes the question difficulty b n and the discrimination degree α n as the variation coefficients, and the enhanced question embedding representation vector satisfies the following formula:

[0020]

[0021] where: K n is the knowledge concept embedding vector, is the variation coefficient, φ and is a trainable weight, P n is the problem embedding representation vector.

[0022] Furthermore, the previous knowledge state H of the learner is fused through a multi-layer perceptron n-1 and the enhanced problem embedding representation vector to model the personalized problem space of the learner. The personalized problem space includes the initial state SP of the problem n , the target state EP n and the adaptation operator OP n , and the knowledge acquisition degree is calculated using an activation function and a gating network Further includes:

[0023] The previous knowledge state H of the learner is fused through a multi-layer perceptron n-1 and the enhanced problem embedding representation vector to model the personalized problem space of the learner, obtaining the initial state SP of the problem n , the target state EP n and the adaptation operator OP n ;

[0024] The tanh activation function is used to obtain the knowledge state candidate value KA from the initial state SP n and the previous knowledge state H n-1 , specifically: n Specifically:

[0025] The initial state SP in the personalized problem space is modeled by integrating the prior knowledge H of the learner n-1 and the enhanced problem embedding representation , and the target state EP n in the personalized problem space. The difference between the target state EP n and the initial state SP n is used to represent the adaptation operator OP n , and the adaptation operator OP n satisfies the following formula: n That is:

[0026]

[0027] OP n = EP n - SP n

[0028] Where: represents vector concatenation, is a learnable matrix, and |K| is the total number of knowledge concepts;

[0029] The knowledge state candidate value KA nInput into the initial state SP n and the target state EP n through a gating network composed of a sigmoid activation function to obtain the final knowledge acquisition level n Specifically:

[0030] Use the tanh activation function to generate a candidate value KA for the knowledge state of knowledge acquisition n and retain the necessary part of the candidate value KA for the knowledge state n through a problem - information - oriented gating network

[0031] Furthermore, generate a corresponding solution from the adaptation operator OP n in the personalized problem space using a multi - layer perceptron and an activation function to obtain a knowledge application degree vector Calculate the probability that the learner answers the question correctly through the four - parameter item response theory and perform attribution processing on the learner's response, further including:

[0032] Use a multi - layer perceptron and the tanh activation function to generate possible solutions from the adaptation operator OP n in the personalized problem space, and determine the learner's final solution according to the problem target vector EP n in the personalized problem space and the knowledge acquisition degree vector

[0033] Simulate the learner's response through the four - parameter item response theory and calculate the probability that the learner answers the question correctly

[0034] Furthermore, construct an incremental index based on the knowledge acquisition degree vector and the knowledge application degree vector to quantify the learner's knowledge internalization degree KI n and model the forgetting effect of the learner over time, and apply the knowledge internalization degree KI n and the forgetting effect

[0035] to the knowledge state H at the previous moment n-1 to obtain the updated knowledge state H of the learner at the current moment n Specifically, it further includes:

[0036] When quantifying the learner's knowledge internalization degree KI n to ensure that the incremental index is non - negative, use (tanh(x)+1) / 2 as a non - negative constraint operator, and the knowledge internalization degree KI n satisfies the following formula:

[0036]

[0037] Among them: represents vector concatenation, SP n is the initial state, EP n is the target state, OP n is the adaptation operator, b KI is the bias term;

[0038] The personalized knowledge increment of the learner is calculated using the following formula

[0039]

[0040] Among them: is the degree of knowledge acquisition, is the vector of the degree of knowledge application, R n is the answer result, b ΓKI is the bias term, KI n is the degree of knowledge internalization, H n-1 is the knowledge state at the previous moment.

[0041] Furthermore, it also includes:

[0042] The degree of knowledge reduction of the learner during the answering interval is calculated according to the following formula:

[0043]

[0044] Among them: The time interval is it n = ts n+1 - ts n represents the time difference between two answers, in minutes, IT n is it n 's embedded representation vector, is the embedded representation vector of the next enhanced question, is the knowledge state at the current moment, b ΓKF is the bias term, ts n+1 is the time of the next answer, ts n is the time of this answer.

[0045] Furthermore, by using the cross-entropy loss function, calculating the loss value between the probability of the learner answering questions correctly and their actual response, and minimizing the loss value by optimizing the model, it further includes:

[0046] The cross-entropy loss function satisfies the following formula:

[0047]

[0048] Among them: is the cross - entropy loss function, r n is the answer result, is the predicted probability.

[0049] An online learning platform for implementing the interpretable knowledge tracing method for the problem - solving process described above, including:

[0050] The original data pre - processing module is used to obtain the exercise records of learners, perform data pre - processing on the exercise records, and obtain the corresponding pre - processed data;

[0051] The problem representation module performs vectorization processing on the pre - processed data to obtain the problem embedding representation vector P n , combines the knowledge concepts, difficulty, and discrimination in the exercise records to obtain the enhanced problem embedding representation vector

[0052] The knowledge acquisition module fuses the knowledge state H of the learner at the previous moment through a multi - layer perceptron n-1 and the enhanced problem embedding representation vector to model the personalized problem space of the learner. The personalized problem space includes the initial state SP of the problem n , the target state EP n and the adaptation operator OP n , and uses an activation function and a gating network to calculate the knowledge acquisition degree vector

[0053] The knowledge application module uses a multi - layer perceptron and an activation function to generate corresponding solutions from the adaptation operators in the personalized problem space, and obtains the knowledge application degree vector Calculates the probability that the learner correctly answers the question through the four - parameter item response theory, and conducts attribution processing on the learner's response;

[0054] The knowledge update module constructs an incremental index based on the knowledge acquisition degree vector and the knowledge application degree vector to quantify the knowledge internalization degree KI of the learner n , and models the forgetting effect of the learner over time , and applies the knowledge internalization degree KI n and the forgetting effect to the knowledge state H at the previous moment n-1 , to obtain the updated knowledge state H of the learner at the current moment n ;

[0055] The model training module calculates the loss value between the probability that the learner correctly answers the question and the actual response by using the cross - entropy loss function, and minimizes the loss value by optimizing the model;

[0056] Repeat and iterate the above process to dynamically update the knowledge state of the learner and generate corresponding personalized teaching strategies for the learner.

[0057] An electronic device includes: a processor, a communication interface, a memory, and a communication bus. Among them, the processor, the communication interface, and the memory complete communication with each other through the communication bus; a computer program is stored in the memory. When the computer program is executed by the processor, the processor is caused to execute the steps of the method.

[0058] The present invention has the following advantages compared with the prior art:

[0059] In the learning process, the present invention takes problem-solving as the core, and through means of artificial intelligence such as problem representation, knowledge acquisition, knowledge application, knowledge update, and model training, deconstructs the process of the learner solving a specific problem, quantifies and optimizes this process, models the degree of knowledge acquisition of the learner before solving the problem, calculates the degree of knowledge application of the learner when solving the problem, measures the degree of knowledge update of the learner after solving the problem, and realizes the interpretability oriented to the problem-solving process during the process of knowledge tracking, which helps educators and learners to master their own teaching and learning status in the first time.

[0060] The present invention uses a multi-layer perceptron to learn the difficulty and discrimination of a problem from the problem embedding, takes the knowledge concept embedding as the main body, and constructs a problem representation with the difficulty and discrimination as the coefficient of variation, realizing the accurate and comprehensive representation of the problem.

[0061] The present invention combines the four-parameter item response theory of educational psychology and probability statistics theory, can calculate the probability of the learner answering the test questions correctly, attributes the performance of the learner to subjective problem factors (difficulty, discrimination), objective learner knowledge factors (knowledge application), and behavioral factors (guessing and slipping), extracts cognitive features from the exercise records of the learner during the teaching and learning process, measures the degree of knowledge internalization, models the degree of forgetting according to the answering interval, achieves the purpose of dynamically updating the knowledge state, generates corresponding personalized teaching strategies for the learner, and makes the learning process of the learner more targeted and personalized.

[0062] The present invention can scientifically and comprehensively diagnose the knowledge mastery of learners, and quantify the performance of learners in two aspects: knowledge acquisition and knowledge application. In personalized education, by quantifying the knowledge state of learners, it can dynamically identify the types of learning activities that learners need to focus on at different time points. If it is detected that a learner has a high level of knowledge acquisition but weak knowledge application ability, the present invention can, through precise analysis, determine that the learner already has an understanding of a certain knowledge point but has difficulties in actual application. Based on this quantified result, more application exercises can be recommended for the learner instead of simply reviewing the knowledge points.

[0063] The personalized guidance strategy based on data of the present invention can effectively help students improve their problem-solving ability, ensure that learning activities are more targeted, and improve teaching efficiency, accurately predict future performance, and explain the reasons behind the prediction, helping teachers and students understand the key factors affecting learning effects, thereby enhancing the pertinence and effectiveness of personalized education. Brief Description of the Drawings

[0064] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for use in the description of the specific embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0065] Figure 1 It is a flowchart of an interpretable knowledge tracing method for the problem-solving process.

[0066] Figure 1A It is a flowchart of an optimized technical solution for step S3.

[0067] Figure 2 It is an architecture diagram of an interpretable knowledge tracing system for the problem-solving process.

[0068] Figure 3 It is an example diagram of three stages of learners solving problems and knowledge transformation.

[0069] Figure 4 It is a model diagram of the interpretable knowledge tracing method in a specific application scenario of an embodiment of the present application.

[0070] Figure 5 It is a schematic structural diagram of an electronic device. Detailed Embodiments

[0071] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0072] The purpose of the multiple embodiments of the present application listed below is to address the deficiencies of existing knowledge tracing methods and provide an interpretable knowledge tracing method for the problem-solving process. By comprehensively utilizing data mining techniques, deep learning, and natural language processing techniques, the knowledge state transformation of learners is modeled in three stages of problem-solving: before, during, and after problem-solving, and the process of how students transform theoretical knowledge into practical skills is explored in depth.

[0073] In the process of the interpretable knowledge tracing method provided by the embodiments of the present application, the performance of learners is decomposed into three major factors: objective problem factors (question difficulty, discrimination), learners' knowledge factors (knowledge acquisition and knowledge application), and learners' behavior factors (guessing and slipping). These factors are combined to help the personalized learning system comprehensively and scientifically diagnose learners' knowledge mastery, and generate two core indicators, namely "knowledge acquisition value" and "knowledge application value", through quantification, which accurately reflects learners' learning status.

[0074] In the personalized learning system, the methods and systems provided by the embodiments of the present application will, according to these quantitative indicators and other behavioral data (such as guessing and slipping situations), evaluate learners' learning needs at different time points in real time, and determine whether they need to focus on strengthening knowledge acquisition or emphasizing knowledge application training. For example, assume that a student's knowledge acquisition value in the'mathematical calculus' module is 80%, while the knowledge application value is only 50%, and the system also detects that the student has more guessing behaviors in answering questions, indicating that the student has a solid grasp of theoretical knowledge but encounters difficulties in practical applications, and there are also situations of insufficient understanding of problems or blind guessing. Through these analyses, the system determines that the student currently needs more targeted application training. Based on this analysis result, the methods and systems provided by the embodiments of the present application will automatically push more challenging practice questions for the student from the preset exercise question bank. These exercises not only include more difficult calculus problems but also questions with stronger discrimination to help the student identify problems at different levels. At the same time, the system will reduce the simple review of basic knowledge and instead guide the student to further improve the practical application ability of knowledge through application and high-discrimination exercises.

[0075] In addition, the methods and systems provided in the embodiments of the present application can also combine the learning habits, answering time, error distribution, and slipping situation of students to further optimize the pushed content. For example, if the system finds that a certain student often "slips" during the problem-solving process (that is, clearly has the ability to solve problems but makes mistakes in some steps), the system will recommend some exercises that help consolidate the problem-solving steps and prevent common mistakes. The methods and systems provided in the embodiments of the present application are dynamically adjusted based on quantitative data and behavioral analysis, ensuring that personalized learning is more targeted, helping teachers and students understand the key obstacles in the learning process, and providing specific strategies for teachers to guide students, thereby improving teaching efficiency.

[0076] As Figure 1 shown in the interpretable knowledge tracing method for the problem-solving process, which is applied to an interpretable knowledge tracing system for the problem-solving process, including:

[0077] Before solving the problem:

[0078] Step S1, problem definition and data preprocessing: Obtain the exercise records of the learner, perform data preprocessing on the exercise records to obtain corresponding preprocessed data;

[0079] Step S2, problem representation: Perform vectorization processing on the preprocessed data to obtain a problem embedding representation vector P n , and combine the knowledge concepts, difficulty, and discrimination in the exercise records to obtain an enhanced problem embedding representation vector

[0080] When solving the problem:

[0081] Step S3, knowledge acquisition: Model the personalized problem space of the learner by fusing the knowledge state H of the learner at the previous moment n-1 and the enhanced problem embedding representation vector in a multi-layer perceptron. The personalized problem space includes the initial state SP n of the problem, the target state EP n and the adaptation operator OP n , and calculate the knowledge acquisition degree vector using an activation function and a gating network

[0082] Step S4, knowledge application: Use a multi-layer perceptron and an activation function to generate a corresponding solution from the adaptation operator in the personalized problem space to obtain a knowledge application degree vector Calculate the probability that the learner correctly answers the question through the four-parameter item response theory, and perform attribution processing on the learner's response;

[0083] Step S5, knowledge update: Based on the knowledge acquisition degree vector and the knowledge application degree vector Construct an incremental index for quantifying the knowledge internalization degree KI of learners n , and for the forgetting effect of learners over time perform modeling, and apply the knowledge internalization degree KI n and the forgetting effect on the knowledge state H at the previous moment n-1 to obtain the updated knowledge state H of the learner at the current moment n ; in step S5, the forgetting effect describes the phenomenon that learners gradually forget the knowledge they have mastered over time, which will affect the knowledge state of learners. In order to characterize the forgetting effect the impact on learners can be modeled in different ways. Exemplarily, by reducing the mastery probability of learners on specific knowledge points, or by introducing a decay function to simulate the gradual weakening of learners' memory.

[0084] In step S5, apply the knowledge internalization degree KI n and the forgetting effect on the knowledge state H at the previous moment n-1 to update and obtain the knowledge state H of the learner at the current moment n , that is: combine the knowledge internalization degree KI n and the forgetting effect and apply it to the knowledge state H at the previous moment n-1 to obtain the knowledge state H at the current moment n .

[0085] The process of combining the knowledge internalization degree KI n and the forgetting effect includes the analysis of learners' answering performance, the fitting of the forgetting curve, and the evaluation of the effectiveness of intervention measures, and finally obtains the updated knowledge state H at the current moment n , and the knowledge state H at the current moment n can dynamically track the knowledge state of learners and provide real-time feedback and adjustment suggestions for teaching.

[0086] Exemplarily, in other embodiments, apply the knowledge state H at the current moment n to the knowledge tracking model (CMKT), and then use the educational concept map to model learners, solving the problem of sparse learner data. The CMKT model combines the educational relationship information and topological information in the concept map, and uses a recurrent neural network to perform the knowledge tracking task - the above application scenario of the knowledge state H at the current moment n provides a way to quantify the knowledge internalization degree of learners and model the forgetting effect, so as to update the knowledge state of learners.

[0087] After solving the problem:

[0088] Step S6, model training: By using the cross-entropy loss function, calculate the loss value between the probability that the learner correctly answers the question and their actual response, and minimize the loss value by optimizing the model;

[0089] Step S7, repeat the process of steps S1 to S6 to dynamically update the learner's knowledge state and generate corresponding personalized teaching strategies for the learner.

[0090] The perceptron (also known as the Multilayer Perceptron, MLP) to which the technical solutions of steps S1 to S7 are applied is a type of feedforward artificial neural network composed of multiple layers, including an input layer, one or more hidden layers, and an output layer. The MLP is trained using a supervised learning technique called backpropagation, usually in combination with the gradient descent algorithm. The technical solutions of steps S1 to S7 can be summarized as:

[0091] (1) Problem definition and data preprocessing; including the definition of the knowledge tracing problem, data collection and data preprocessing on the online learning platform;

[0092] (2) Problem representation: Propose a knowledge-centered problem representation and adjust the variability of the problem by learning the difficulty and discrimination of the problem;

[0093] (3) Knowledge acquisition: Before solving the problem, model the learner's personalized problem space (including the initial state, goal state, and adaptation operator of the problem), and simulate their acquisition of problem-related knowledge through a gating mechanism;

[0094] (4) Knowledge application: During the process of solving the problem, capture the degree of goal-oriented knowledge application of the learner and attribute their response to the problem using the four-parameter item response theory;

[0095] (5) Knowledge update: After solving the problem, quantify the degree of knowledge internalization of the learner using incremental metrics and model the forgetting effect over time;

[0096] (6) Model training: By using the cross-entropy loss function, calculate the loss value between the predicted probability that the learner correctly answers the question and their actual response, and minimize the loss value by optimizing the model.

[0097] (7) Iterative update: Repeat the process of (1) to (6) above to dynamically update the learner's knowledge state and generate corresponding personalized teaching strategies for the learner.

[0098] In the technical solutions provided in steps S1 to S7, the knowledge transformation of the learner is modeled by deconstructing the three stages of the learner's problem-solving, so as to predict the learner's future performance. In the specific technical solutions of steps S1 to S7, first, the historical exercise records of the learner are collected and preprocessed from the online learning platform, then the rich attributes (difficulty and discrimination) of the learning problems are learned according to the exercise records to enhance the knowledge-centered problem representation, and then the knowledge conversion in the learner's problem-solving process is simulated. Before solving the problem (i.e., the knowledge acquisition module), the learner's personalized problem perception and knowledge acquisition degree are modeled; when solving the problem (i.e., the knowledge application module), the knowledge application level is measured according to the knowledge acquisition, and the problem response is calculated through the four-parameter item response theory; after the problem is solved (i.e., the knowledge update module), the degree of the learner's knowledge internalization and forgetting is quantified. Finally, the model is trained to ensure the rationality of the knowledge state diagnosis and the accuracy of the performance prediction.

[0099] Preferably, in step S1, the obtaining of the learner's exercise records and the data preprocessing of the exercise records to obtain the corresponding preprocessed data further include:

[0100] The exercise records include the answering time ts n , the answered question p n , the knowledge concept k examined by the question n and the answering result r n , and the data preprocessing includes cleaning, duplicate removal, and deletion of missing values.

[0101] Exemplarily, in step S1, the problem definition and data preprocessing are specifically:

[0102] Step S11, problem definition: Given N exercise records of a learner, each exercise record contains the answering time ts n , the answered question p n , the knowledge concept k examined by the question n and the answering result r n . The task of knowledge tracing is to evaluate the learner's knowledge state according to the learner's exercise records and predict the probability of correctly answering the next question. For example:

[0103] Given a learner's historical learning sequence X N ={X1, X2,..., X N}, N represents the total number of student exercises. Where X n =(ts n , p n , k n , r n ), n ∈ N is the most basic exercise unit, indicating that the student, at the time of ts n , for the knowledge concept k examinedn The problem p n The response is r n , r n = 1 indicates a correct answer, r n = 0 indicates a wrong answer. The task of knowledge tracing is to evaluate the student's knowledge state based on the learner's learning history X N and predict the probability that the learner will answer the next question correctly, that is, p(r n+1 = 1|p n+1 , X n ).

[0104] Step S12, data preprocessing: Collect the learner's answering behavior data on the online learning platform, and perform operations such as cleaning, deduplication, and deleting missing values on this data to ensure data quality. For example:

[0105] Collect the learner's answering behavior data on the online learning platform, including the time when each student answers each time, the specific questions answered, the knowledge concepts examined by the questions, and the answering results. Then, perform operations such as cleaning, deduplication, and deleting missing values on this data to ensure data quality. Next, encode the features for subsequent analysis and modeling by the input model.

[0106] Preferably, in step S2, the preprocessed data is vectorized to obtain the question embedding representation vector P n , combined with the knowledge concepts, difficulty, and discrimination in the exercise record, to obtain the enhanced question embedding representation vector Further includes:

[0107] The enhanced question embedding representation vector takes the knowledge concept embedding vector K n as the main body, takes the question difficulty d n and the discrimination α n as the coefficient of variation. The enhanced question embedding representation vector satisfies the following formula:

[0108]

[0109] Where: K n is the knowledge concept embedding vector, is the coefficient of variation, φ and are trainable weights, P n is the question embedding representation vector.

[0110] In step S2, the question representation is enhanced by learning the rich attributes of the question. Therefore, the question representation includes the difficulty and discrimination attributes of the learning question and models the knowledge-centered question representation:

[0111] Step S21, learning the difficulty and discrimination of questions: Learning the difficulty and discrimination of questions from the question embedding vector through a multi-layer perceptron:

[0112]

[0113] Note: σ represents the sigmoid activation function, is a learnable vector of dimension d h and P n is the question embedding vector, and b pa is the bias. d n represents the question difficulty, and α n represents the question discrimination, both of which are scalars.

[0114] Step S22, question representation: The goal of practice is to learn knowledge, so the knowledge concepts examined by questions are crucial. In addition, questions examining the same knowledge concept may have different impacts on learning and performance due to differences in difficulty and discrimination. This method simultaneously considers the primary-secondary relationship and question attributes to enhance question representation:

[0115]

[0116] Note: K n is the knowledge concept embedding, is the coefficient of variation, and φ and are trainable weights.

[0117] As Figure 1A shown, preferably, in step S3, the previous moment knowledge state H of the learner is fused through a multi-layer perceptron n-1 and the enhanced question embedding representation vector to model the personalized question space of the learner. The personalized question space includes the initial state SP of the question n , the target state EP n and the adaptation operator OP n , and the knowledge acquisition degree is calculated using an activation function (tanh activation function or sigmoid activation function) and a gating network Further includes:

[0118] Step S31, modeling the personalized question space of the learner by fusing the previous moment knowledge state H of the learner n-1 and the enhanced question embedding representation vector to obtain the initial state SP of the question n , the target state EP n and the adaptation operator OP n ;

[0119] Step S32, using the tanh activation function to obtain the candidate knowledge state KA from the initial state SP n and the knowledge state H at the previous moment n-1 Specifically: n By integrating the learner's prior knowledge H

[0120] and the problem representation n-1 to model the initial state SP and the target state EP n in the personalized problem space, using the difference between the target state EP n and the initial state SP n to represent the adaptation operator OP n , and the adaptation operator OP n satisfies the following formula: n

[0121]

[0122] OP n = EP n - SP n ……(5)

[0123] where: represents vector concatenation, is a learnable matrix, and |K| is the total number of knowledge concepts;

[0124] Step S33, sending the candidate knowledge state KA n into the gated network composed of the initial state SP n , the target state EP n , and the adaptation operator OP n through the sigmoid activation function to obtain the final knowledge acquisition degree Specifically:

[0125] Using the tanh activation function to generate the candidate knowledge state KA of knowledge acquisition n , and retaining the necessary part in the candidate knowledge state KA n through the problem information-oriented gated network.

[0126] The specific content of knowledge acquisition in Step S3 can be summarized as:

[0127] Problem perception: By fusing the learner's prior knowledge and the problem representation to model the initial state and the target state in the personalized problem space, and using the difference between the target state and the initial state to represent the adaptation operator in the problem space.

[0128] The learner's solution to the problem begins with constructing an internal representation of the external problem statement, namely the "problem space" (including the initial state, the goal state, and the adaptation operator). This method models the initial state SP n-1 in the personalized problem space by integrating the learner's prior knowledge H and enhancing the problem embedding representation n and the goal state EP n , and uses the difference between the goal state EP n and the initial state SP n to represent the adaptation operator OP n :

[0129]

[0130] OP n = EP n - SP n ……(7)

[0131] Note: denotes vector concatenation, is a learnable matrix, and |K| is the total number of knowledge concepts.

[0132] Knowledge acquisition: Use the tanh activation function to generate candidate values for knowledge acquisition from the initial state of the problem space and the learner's knowledge state, and then select and retain the necessary parts of the candidate values through a problem information-oriented gating network.

[0133] Knowledge acquisition includes acquiring knowledge from the external environment and constructing new knowledge through internal thinking processes. In the knowledge tracing scenario, the external knowledge comes from the initial state of the problem (known conditions) SP n , and the internal knowledge is the learner's knowledge state H n-1 at the previous moment. The embodiments of this application use the tanh activation function to generate candidate knowledge states KA n for knowledge acquisition:

[0134]

[0135] Since the learner's memory capacity is limited, the learner will selectively focus on the most relevant and core aspects of the problem during the knowledge acquisition process for more effective planning. To address this issue, the embodiments of this application further design a gating network to select and retain the necessary parts through problem information:

[0136]

[0137] Preferably, in step S4, using the multi-layer perceptron and the activation function to obtain from the adaptation operator OP nGenerate corresponding solutions to obtain the knowledge application degree vector Calculate the probability that the learner answers the question correctly through the four-parameter item response theory, and attribute the learner's response, further including:

[0138] Use a multi-layer perceptron and the tanh activation function to generate possible solutions from the adaptation operator OP of the personalized problem space n Generate possible solutions according to the problem target vector EP in the personalized problem space n And the knowledge acquisition degree vector To determine the learner's final solution;

[0139] Simulate the learner's response through the four-parameter item response theory and calculate the probability that the learner answers the question correctly.

[0140] In step S4, the specific content of knowledge application can be summarized as:

[0141] Step S41, knowledge application: Use a multi-layer perceptron and the tanh activation function to generate possible solutions from the adaptation operator of the problem space, and further determine the learner's final solution according to the problem target and knowledge acquisition degree in the problem space;

[0142] Knowledge application involves transforming abstract knowledge into specific action plans and adjusting and improving these plans in practice. The solution to the problem is contained in the adaptation operator of the problem space. This method uses the tanh activation function to generate possible solutions from the adaptation operator OP n Generate possible solutions, and design a gate n And knowledge acquisition Led by To determine the learner's final solution:

[0143]

[0144]

[0145] Step S42, question answering: Use the four-parameter item response theory to calculate the probability that the student answers the question correctly, and attribute the learner's response to subjective question factors (difficulty d n , discrimination α n ), objective learner knowledge factors (knowledge application ), and learner behavior factors (guess G n And slip S n ):

[0146]

[0147] In step S42, the four-parameter item response theory is incorporated to simulate students' responses by fully considering subjective question factors and objective learner factors. First, the knowledge application of the learner is integrated with the prediction questions to determine the degree of knowledge application of the learner that is most relevant to the prediction questions:

[0148]

[0149] Then, the guessing G n and slipping S n probabilities of the learner on the prediction questions are modeled:

[0150]

[0151] Finally, by fully considering the question factors (difficulty, discrimination), the learner's knowledge factors (knowledge application), and the learner's behavior factors (guessing and slipping), the students' responses are simulated:

[0152]

[0153] Note: D is a constant with a value of 4 × 1.7.

[0154] The four-parameter item response theory (4PL) in step S4 is a model in psychometrics and educational measurement used to evaluate the difficulty of test questions (items) and the ability of test takers. The four-parameter item response theory is an extension of the item response theory (IRT) and can more precisely describe the responses of test takers to questions.

[0155] In IRT, the most commonly used model is the three-parameter model (3PL), which includes three parameters:

[0156] Difficulty parameter (b): Represents the difficulty level of the question.

[0157] Discrimination parameter (a): Represents the ability of the question to distinguish test takers with different ability levels.

[0158] Guessing parameter (c): Represents the probability that a test taker can correctly answer the question even with very low ability.

[0159] The four-parameter model adds one parameter to the three-parameter model: 4. Pseudo-chance parameter (d): Represents the probability that a test taker will answer the question incorrectly even with very high ability.

[0160] The probability curve of the four-parameter model can be expressed as:

[0161] P(X = 1|θ, a, b, c, d) = c + (1 - c - d)\frac{e^{a(θ - b)}}{1 + e^{a(θ - b)}}P(X = 1|

[0162] θ, a, b, c, d) = c + (1 - c - d)\frac{e^{a(θ - b)}}{1 + e^{a(θ - b)}}……(17)

[0163] Where: P(X = 1) is the probability that the test taker answers the question correctly, θ is the ability level of the test taker, a is the discrimination parameter, b is the difficulty parameter, c is the guessing parameter, and d is the slipping parameter (pseudo - chance parameter).

[0164] In the embodiment of the present application, the four - parameter item response theory is adopted, which is particularly suitable for the situation where the ability level distribution is wide or the question difficulty difference is large. For example, the addition of the slipping parameter (pseudo - chance parameter) enables the model to better handle the following situation: the test taker may answer simple questions incorrectly due to carelessness or other non - ability factors, that is, the learner clearly has the ability to solve the problem but makes mistakes in some steps.

[0165] Preferably, in step S5, the incremental index is constructed based on the knowledge acquisition degree vector and the knowledge application degree vector n to quantify the knowledge internalization degree KI of the learner, and the forgetting effect n of the learner changing with time is modeled, and the knowledge internalization degree KI n-1 and the forgetting effect n act on the knowledge state H

[0166] at the previous moment, and the updated current - moment knowledge state H n of the learner is obtained. Further, it includes: n When quantifying the knowledge internalization degree KI

[0167]

[0168] of the learner, to ensure that the incremental index is non - negative, (tanh(x)+1) / 2 is used as the non - negative constraint operator, and the knowledge internalization degree KI denotes vector concatenation, SP n is the initial state, SP n is the target state, OP n is the adaptation operator, and b KI is the bias term;

[0169] The personalized knowledge increment of the learner is calculated by the following formula

[0170]

[0171] Wherein: is the degree of knowledge acquisition, is the degree of knowledge application vector, R n is the answer result, b ΓKI is the bias term, KI n is the degree of knowledge internalization, H n-1 is the knowledge state at the previous moment,

[0172] The degree of knowledge reduction of the learner during the answering interval is calculated according to the following formula:

[0173]

[0174] Wherein: the time interval is it n = ts n+1 - ts n represents the time difference between two answers, in minutes, IT n is the embedding representation vector of it n is the embedding representation vector of the next enhanced question, is the knowledge state at the current moment, b ΓKF is the bias term;

[0175] In step S5, the knowledge state of the learner will change after practice. The increase in knowledge mastery reflects the internalization of knowledge, while forgetting will lead to the reduction of knowledge. Therefore, knowledge update includes two modules: knowledge internalization and forgetting.

[0176] Step S51, knowledge internalization: The new knowledge of the learner comes from question practice. Therefore, all the information of the question is integrated to obtain a candidate value of knowledge increment. The constructivist learning theory shows that learning is an active construction process, and the whole practice process has a positive impact on the learner. Therefore, is used as a non - negative constraint operator to ensure that the increment is non - negative:

[0177]

[0178] Through internalization, the learner not only receives information but also needs to integrate it into the cognitive framework. For this purpose, the embodiment of the present application proposes a growth index considering knowledge acquisition, knowledge application, and answering response to determine the personalized knowledge increment of the learner:

[0179]

[0180] ​Step S52, Knowledge Forgetting: Limited by the human memory system, current knowledge may be forgotten over time. To simulate complex forgetting effects, the embodiments of the present application provide a forgetting gate to model the degree of knowledge reduction of learners during the answering interval. The specific formula is as follows:

[0181]

[0182]

[0183] Note: The time interval is it n = ts n+1 - ts n , that is, the time difference between two answers, in minutes. IT n is the embedding representation of it n In the embodiments of the present application, all time intervals greater than 1 month are set to 1 month.

[0184] Preferably, in step S6, by using the cross-entropy loss function, calculating the loss value between the probability of the learner answering the questions correctly and the actual response, and minimizing the loss value by optimizing the model, further includes:

[0185] The cross-entropy loss function satisfies the following formula:

[0186]

[0187] Where: is the cross-entropy loss function, r n is the answering result, r n is the answering result (actual response), is the predicted probability (predicted response).

[0188] In step S6, in order to train all parameters Θ, the cross-entropy logarithmic loss between the predicted response and the actual response r n is used as the objective function, and the parameters in the model are updated by using the Adam optimization algorithm. The specific steps when using the Adam optimization algorithm to update the model parameters are as follows:

[0189] Step S61, Define the model and the loss function: Define the model and the loss function. The architecture of the model includes the type of layers and the number of neurons. Select cross-entropy as the loss function. The model is specifically a deep learning model;

[0190] Step S62, Initialize the parameters: Before starting the training, initialize the parameters of the model (such as the Xavier initialization method). The parameters include weights and bias values;

[0191] Step S63, Forward Propagation: Perform the predictive output of forward propagation for each training sample through the sigmoid function, converting the linear output of the model into a probability distribution;

[0192] Step S64, Calculate Loss: Use the cross-entropy loss function to calculate the difference between the model prediction and the true label;

[0193] Step S65, Backward Propagation: Calculate the gradient of the loss function with respect to the model parameters based on the chain rule;

[0194] Step S66, Parameter Update: Update the model parameters using an adaptive learning rate optimization algorithm (such as the Adam optimization algorithm).

[0195] The optimization technical solution provided by steps S61 to S66 minimizes the difference between the model prediction and the actual data by constructing and training a deep learning model and using the cross-entropy loss function as the optimization objective. In the process of model construction, first design the model architecture, including selecting the appropriate network layer type, the number of neurons in each layer, etc., assign initial values to the weights and biases of the model, perform parameter initialization before training starts, calculate the prediction output of the model for each training sample, use the sigmoid function to convert the linear output into a probability distribution, then use the cross-entropy loss function to measure the difference between the probability distribution predicted by the model and the true label, perform loss calculation, then calculate the gradient of the loss function with respect to the model parameters through the chain rule to provide a basis for parameter update, and finally apply the Adam optimization algorithm (adaptive learning rate optimization algorithm) to update the model parameters according to the gradient, achieving the minimization of the loss function.

[0196] The optimization technical solution provided by steps S61 to S66 enables the model to learn complex patterns and features in the data through iterative training. As the parameters are continuously updated, the prediction performance of the model in the classification task is gradually improved, and the generalization ability is enhanced. Moreover, through automated gradient calculation and parameter update, human intervention is reduced, improving the efficiency of the training process. Finally, based on the adaptive learning rate optimization algorithm (Adam optimization algorithm), it adapts to different training data and tasks, improving the applicability of the model.

[0197] The optimization technical solution provided by steps S61 to S66 realizes the automated learning and high-precision prediction of classification problems by constructing, training, and optimizing a neural network model, while ensuring that the (deep learning) model has good generalization ability and adaptability, providing a good optimization solution for calculating the loss value between the probability that a learner correctly answers a question and their actual response and minimizing the loss value.

[0198] For the method steps disclosed in the above embodiments, for the purpose of simple description, the method steps are expressed as a series of combinations of actions. However, those skilled in the art should be aware that the embodiments of the present invention are not limited by the described order of actions, because according to the embodiments of the present invention, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions involved are not necessarily essential for the embodiments of the present invention.

[0199] As Figure 2 shown, an interpretable knowledge tracing system for the problem-solving process is used to implement the interpretable knowledge tracing method for the problem-solving process described in the embodiments of the present application, including:

[0200] A raw data preprocessing module, configured to obtain the exercise records of the learner, perform data preprocessing on the exercise records, and obtain corresponding preprocessed data;

[0201] A problem representation module, which performs vectorization processing on the preprocessed data to obtain a problem embedding representation vector P n , and combines the knowledge concepts, difficulty, and discrimination in the exercise records to obtain an enhanced problem embedding representation vector

[0202] A knowledge acquisition module, which fuses the previous knowledge state H of the learner through a multi-layer perceptron n-1 and the enhanced problem embedding representation vector to model the personalized problem space of the learner. The personalized problem space includes the initial state SP n of the problem, the target state EP n and the adaptation operator OP n , and calculates the knowledge acquisition degree vector using an activation function and a gating network

[0203] A knowledge application module, which uses a multi-layer perceptron and an activation function to generate a corresponding solution from the adaptation operator in the personalized problem space, and obtains a knowledge application degree vector Calculates the probability that the learner correctly answers the question through the four-parameter item response theory, and performs attribution processing on the learner's response;

[0204] A knowledge update module, based on the knowledge acquisition degree vector and the knowledge application degree vector constructs an incremental index for quantifying the knowledge internalization degree KI n of the learner, and models the forgetting effect of the learner changing over time, and internalizes the knowledge degree KI n and the forgetting effect Act on the knowledge state H at the previous moment n-1 to obtain the updated knowledge state H of the learner at the current moment n ;

[0205] The model training module calculates the loss value between the probability of the learner answering questions correctly and the actual response by using the cross-entropy loss function, and minimizes the loss value by optimizing the model;

[0206] Repeat the above process iteratively to dynamically update the knowledge state of the learner and generate corresponding personalized teaching strategies for the learner.

[0207] The implementation manners of the system described above are merely illustrative. For example, each functional module, unit or subsystem in the system may or may not be physically separated, or may or may not be a physical unit, that is, it may be located in the same place or distributed to multiple different systems and their subsystems or modules. Those skilled in the art can select some or all of the functional modules, units or subsystems according to actual needs to achieve the purpose of the embodiments of the present invention. For the above situations, those of ordinary skill in the art can understand and implement them without creative efforts.

[0208] As Figure 3 and Figure 4 shown, based on the interpretable knowledge tracing method and system for the problem-solving process, the present invention provides a corresponding online learning platform. The online learning platform is provided with the interpretable knowledge tracing system for the problem-solving process and executes the interpretable knowledge tracing method for the problem-solving process.

[0209] As Figure 5 shown, based on providing the interpretable knowledge tracing method and system, the present invention further provides a corresponding electronic device and storage medium:

[0210] An electronic device, characterized in that it includes: a processor, a communication interface, a memory and a communication bus. Among them, the processor, the communication interface and the memory communicate with each other through the communication bus; a computer program is stored in the memory. When the computer program is executed by the processor, the processor executes the steps of the interpretable knowledge tracing method for the problem-solving process.

[0211] A computer-readable storage medium stores a computer program executable by an electronic device. When the computer program runs on the electronic device, the electronic device executes the steps of the interpretable knowledge tracing method for the problem-solving process.

[0212] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, storage, database, or other medium used in the various embodiments provided in this application can include non-volatile and / or volatile memories. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM has various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and Rambus dynamic RAM (RDRAM), etc.

[0213] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered that the scope described in this specification is covered.

[0214] In addition, those skilled in the art can understand that although some of the embodiments described herein include certain features included in other embodiments rather than other features, the combination of the features of different embodiments means that it is within the scope of the present invention and forms different embodiments. For example: Any one of the embodiments claimed in the claims can be used in any combination manner of the embodiments of the present invention.

[0215] In the description of this specification, the description with reference to terms such as "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in a suitable manner in any one or more embodiments or examples.

[0216] In addition, the technical solutions between the various embodiments of the present invention can be combined with each other, but it must be based on what can be achieved by those of ordinary skill in the art. When the combination of technical solutions results in contradictions or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0217] Finally, it should be noted that: the above 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 of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the various embodiments of the present invention.

Claims

1. An interpretable knowledge tracing method for the problem-solving process, applied to an interpretable knowledge tracing system for the problem-solving process, characterized in that, Including: Before problem-solving: Obtain the learner's practice records, perform data preprocessing on the practice records to obtain corresponding preprocessed data; Perform vectorization processing on the preprocessed data to obtain a problem embedding representation vector P n , and combine the knowledge concepts, difficulty, and discrimination degree in the exercise record to obtain an enhanced problem embedding representation vector When solving problems: fusing the knowledge state H of the learner at the previous moment through a multi-layer perceptron n-1 and the enhanced problem embedding representation vector to model the personalized problem space of the learner, where the personalized problem space includes the initial state SP of the problem n , the target state EP n and the adaptation operator OP n , and calculating the knowledge acquisition degree vector using an activation function and a gating network Generate a corresponding solution from the adaptation operator in the personalized problem space using a multi-layer perceptron and an activation function to obtain a knowledge application degree vector Calculate the probability that the learner answers the question correctly through the four-parameter item response theory, and attribute the learner's response Based on the knowledge acquisition degree vector and the knowledge application degree vector An incremental index is constructed to quantify the knowledge internalization degree KI of the learner n , and the forgetting effect of the learner changing over time is modeled, and the knowledge internalization degree KI n and the forgetting effect act on the knowledge state H at the previous moment n-1 to obtain the updated knowledge state H of the learner at the current moment n ; After problem-solving: By using the cross-entropy loss function, calculate the loss value between the probability of the learner correctly answering questions and their actual responses, and minimize the loss value through optimizing the model; Repeat and iterate the above process to dynamically update the learner's knowledge state and generate corresponding personalized teaching strategies for the learner.

2. The interpretable knowledge tracking method for problem-solving process according to claim 1, wherein The obtaining of the learner's practice records, performing data preprocessing on the practice records to obtain corresponding preprocessed data further includes: The exercise record includes the answering time ts n , the answered question p n , the knowledge concept k examined by the question n and the answering result r n , and the data preprocessing includes cleaning, duplicate removal, and deletion of missing values.

3. The interpretable knowledge tracing method for problem-solving process according to claim 1, wherein Performing vectorization processing on the preprocessed data to obtain a problem embedding representation vector P n , and combining the knowledge concepts, difficulty, and discrimination in the exercise records to obtain an enhanced problem embedding representation vector Further comprising: The enhanced problem embedding representation vector uses the knowledge concept embedding vector K n as the main body, and takes the problem difficulty d n and the discrimination α n as the variation coefficients. The enhanced problem embedding representation vector satisfies the following formula: Among them: K n is the knowledge concept embedding vector, is the coefficient of variation, φ and are trainable weights, P n is the problem embedding representation vector.

4. The interpretable knowledge tracing method for problem-solving process according to claim 1, wherein The knowledge state H of the learner at the previous moment is fused through a multi-layer perceptron n-1 and the enhanced problem embedding representation vector to model the personalized problem space of the learner, where the personalized problem space includes the initial state SP of the problem n , the target state EP n and the adaptation operator OP n , and the knowledge acquisition degree is calculated using an activation function and a gating network Further comprising: Fusing the learner's knowledge state H at the previous moment through a multi-layer perceptron n-1 and the enhanced problem embedding representation vector to model the learner's personalized problem space, obtaining the initial state SP of the problem n , the target state EP n and the adaptation operator OP n ; Using the tanh activation function to obtain the candidate value KA of the knowledge state from the initial state SP n and the knowledge state H at the previous moment n-1 Specifically, n it is as follows: By integrating the learner's prior knowledge H n-1 and the problem representation to model the initial state SP in the personalized problem space n and the target state EP n , using the difference between the target state EP n and the initial state SP n to represent the adaptation operator OP n , the adaptation operator OP n satisfies the following formula: OP n = EP n - SP n Wherein: represents vector concatenation, is a learnable matrix, and |K| is the total number of knowledge concepts; The candidate value KA of the knowledge state n is fed into the gating network composed of the initial state SP n , the target state EP n , and the adaptation operator OP n through the sigmoid activation function to obtain the final knowledge acquisition degree Specifically: Use the tanh activation function to generate candidate values KA of the knowledge state for knowledge acquisition n and retain the necessary part of the candidate values KA of the knowledge state through a question information-oriented gating network n in 5. The interpretable knowledge tracing method for the problem-solving process according to claim 1, characterized in that Generating a corresponding solution from the adaptation operator OP of the personalized problem space using a multi-layer perceptron and an activation function to obtain a knowledge application degree vector n and Calculating the probability that a learner answers a question correctly through the four-parameter item response theory, and performing attribution processing on the learner's response, further including: Using a multi-layer perceptron and the tanh activation function, generate possible solutions from the adaptation operator OP in the personalized problem space, and determine the learner's final solution according to the problem target vector EP in the personalized problem space n and the knowledge acquisition degree vector n ; ​ Simulate the learner's responses through the four-parameter item response theory and calculate the probability of the learner correctly answering questions.

6. The explainable knowledge tracing method for the problem-solving process according to claim 1, wherein Based on the knowledge acquisition degree vector and the knowledge application degree vector construct an incremental index for quantifying the knowledge internalization degree KI n of the learner, and model the forgetting effect of the learner changing over time. Apply the knowledge internalization degree KI n and the forgetting effect to the knowledge state H n-1 at the previous moment to obtain the updated knowledge state H n of the learner at the current moment. Further comprising: When quantifying the knowledge internalization degree KI of learners n To ensure that the incremental index is non - negative, (tanh(x)+1) / 2 is used as the non - negative constraint operator, and the knowledge internalization degree KI n satisfies the following formula: Wherein: represents vector concatenation, SP n is the initial state, EP n is the target state, OP n is the adaptation operator, b KI is the bias term; The personalized knowledge increment of the learner is calculated using the following formula Wherein: is the degree of knowledge acquisition, is the knowledge application degree vector, R n is the answer result, b ΓKI is the bias term, KI n is the degree of knowledge internalization, H n-1 is the knowledge state at the previous moment.

7. The interpretable knowledge tracing method for the problem-solving process according to claim 6, wherein Also included: Calculate the degree of knowledge reduction of the learner during the answering interval according to the following formula: where: the time interval is it n = ts n+1 - ts n , representing the time difference between two responses, in minutes, IT n is the embedded representation vector of it n and is the next enhanced question embedded representation vector, is the current knowledge state, b ΓKF is the bias term, ts n+1 is the next response time, ts n is the current response time.

8. The interpretable knowledge tracing method for problem-solving process according to claim 1, characterized in that The calculating of the loss value between the probability of the learner correctly answering questions and their actual responses by using the cross-entropy loss function and minimizing the loss value through optimizing the model further includes: The cross-entropy loss function satisfies the following formula: Wherein: is the cross-entropy loss function, r n is the answer result, is the predicted probability.

9. An online learning platform for implementing the interpretable knowledge tracing method for problem-solving processes described in any one of claims 1 to 8, characterized in that, Including: An original data preprocessing module for obtaining the learner's practice records, performing data preprocessing on the practice records to obtain corresponding preprocessed data; A problem representation module performs vectorization processing on the preprocessed data to obtain a problem embedding representation vector P n , and combines the knowledge concepts, difficulty, and discrimination degree in the exercise record to obtain an enhanced problem embedding representation vector A knowledge acquisition module fuses the learner's previous knowledge state H through a multi-layer perceptron n-1 and the enhanced problem embedding representation vector to model the learner's personalized problem space, which includes the initial state SP of the problem n , the target state EP n and the adaptation operator OP n , and calculates the knowledge acquisition degree vector using an activation function and a gating network Knowledge application module, which uses a multi-layer perceptron and an activation function to generate corresponding solutions from the adaptation operators in the personalized problem space, and obtains a knowledge application degree vector Calculate the probability that the learner answers the question correctly through the four-parameter item response theory, and attribute the learner's response Knowledge update module, based on the knowledge acquisition degree vector and the knowledge application degree vector to construct an incremental index for quantifying the knowledge internalization degree KI n of the learner, and model the forgetting effect of the learner changing over time, and apply the knowledge internalization degree KI n and the forgetting effect to the knowledge state H n-1 at the previous moment to obtain the updated knowledge state H n of the learner at the current moment; A model training module that calculates the loss value between the probability of the learner correctly answering questions and their actual responses by using the cross-entropy loss function and minimizes the loss value through optimizing the model; Repeat and iterate the above process to dynamically update the learner's knowledge state and generate corresponding personalized teaching strategies for the learner.

10. An electronic device, characterized in that, Including: A processor, a communication interface, a memory, and a communication bus. Among them, the processor, the communication interface, and the memory complete communication with each other through the communication bus; a computer program is stored in the memory. When the computer program is executed by the processor, the processor executes the steps of the method according to any one of claims 1 to 8.

Images