Deep knowledge tracking interpretation method and device

Abstract

The invention provides a deep knowledge tracking interpretation method and device, and relates to the field of intelligent learning guide analysis. The method comprises the steps of obtaining a historical answer record of a target student and a to-be-predicted question, and constructing a learning interaction sequence corresponding to the target student according to the historical answer record and an answer time sequence; inputting the learning interaction sequence into a knowledge tracking model to output an answer probability prediction value that the target student correctly answers the to-be-predicted question; based on a preset strategy, performing pre-screening operation on the learning interaction sequence to obtain candidate subsequences; calculating the causal contribution degree of the candidate subsequences to the answer probability predicted value in combination with an educational constraint method, wherein the educational constraint method comprises a cognitive load reduction method and a time decay weighting method; and according to the causal contribution degree, adopting a multi-target particle swarm optimization algorithm to generate a target interpretation subsequence, and performing tracking interpretation on the answer probability prediction value through the target interpretation subsequence. The problem that a DLKT model cannot visually present which key interaction pairs is solved, and the probability prediction of correctly answering a certain question by a student is mainly driven.

Description

Technical Field

[0001] This application relates to the field of intelligent learning guidance analysis, and particularly to a deep knowledge tracing and explanation method and device. Background Art

[0002] Knowledge tracing technology is widely used in the field of education to analyze students' mastery of specific knowledge points during the learning process. The development of deep learning technology enables KT models to fit complex non-linear relationships. The deep knowledge tracing model (DLKT) based on deep learning technology uses neural networks and can well model students' historical learning behaviors and predict future answering situations. However, the internal states and prediction processes of these models are often black boxes, making it difficult to explain why the model makes a certain prediction. For the results output by the model, users cannot verify their logical rationality. If the model has implicit biases (such as underestimating the abilities of certain types of students), it is difficult to trace the reasons. Therefore, interpretability is a necessary condition for the deep knowledge tracing model to be applied in the education scenario.

[0003] The existing interpretability methods for DLKT models can be roughly divided into two types: one is to calculate the contribution of a single interaction pair in the learning interaction sequence, thus ignoring the joint effect between interaction pairs; the other is to maximize the overall causal effect of the learning interaction sequence, thus ignoring internal factor redundancy and including a large number of irrelevant, pseudo-related or redundant factors. In fact, there are complex interactions between interaction pairs. For example, two interaction pairs containing similar information will result in redundancy. Redundant factors cannot bring explanatory effects but may instead bring misleading information, that is, the DLKT model cannot intuitively show which key interaction pairs mainly drive the probability prediction of a student's correct answer to a certain question.

[0004] Therefore, there is an urgent need for a deep knowledge tracing and explanation method and device. Summary of the Invention

[0005] This application provides a deep knowledge tracing and explanation method and device, which solves the problem that the DLKT model cannot intuitively show which key interaction pairs mainly drive the probability prediction of a student's correct answer to a certain question.

[0006] In the first aspect of the present application, a deep knowledge tracing and explanation method and device are provided. The method includes: obtaining the historical answer records of the target student and the question to be predicted, and constructing a learning interaction sequence corresponding to the target student according to the historical answer records in the order of answering time; inputting the learning interaction sequence into a knowledge tracing model to output a predicted value of the probability that the target student correctly answers the question to be predicted; based on a preset strategy, performing a pre-screening operation on the learning interaction sequence to obtain candidate subsequences; calculating the causal contribution degree of the candidate subsequences to the predicted value of the answering probability by combining an educational constraint method, and the educational constraint method includes a cognitive load reduction method and a time decay weighting method; according to the causal contribution degree, using a multi-objective particle swarm optimization algorithm to generate a target explanatory subsequence, and performing a tracing explanation on the predicted value of the answering probability through the target explanatory subsequence.

[0007] Optionally, based on knowledge in the field of education, performing a pre-screening operation on the learning interaction sequence to obtain candidate subsequences, specifically including: using particle swarm coding to assign corresponding particles to each subsequence in the learning interaction sequence; obtaining a target candidate subsequence, where the target candidate subsequence is any one subsequence among multiple learning interaction sequences; obtaining the position vector of the particle corresponding to the target candidate subsequence, and confirming whether the values of each dimension of the position vector meet the preset screening conditions; if the values of each dimension of the position vector meet the preset screening conditions, then taking the target candidate subsequence as the candidate subsequence.

[0008] Optionally, the preset strategy includes a hidden state observation generation strategy and a gradient prior generation strategy.

[0009] Optionally, calculating the causal contribution degree of the candidate subsequences to the predicted value of the answering probability by combining an educational constraint method, specifically including: calculating the prediction ability score corresponding to the explanatory subsequence, and the prediction ability score is used to evaluate the prediction influence degree of the interaction information contained in the explanatory subsequence on the predicted value of the answering probability; calculating the sufficiency score and the necessity score corresponding to the explanatory subsequence, and the sufficiency score and the necessity score are used to evaluate the internal information redundancy degree in the explanatory subsequence; based on the prediction ability score, the sufficiency score, and the necessity score, and calculating the causal contribution degree of the explanatory subsequence to the predicted value of the answering probability based on a time exponential decay function.

[0010] Optionally, calculating the prediction ability score corresponding to the explanatory subsequence, specifically including: calculating the prediction ability score through the following formula: ; where is the prediction ability score corresponding to the th explanatory subsequence, is the th explanatory subsequence, represents an intervention operation, Indicates an operation that allows the knowledge tracking model to make predictions based only on and Indicates the output result obtained by allowing the knowledge tracking model to make predictions based only on The original answer probability prediction value is used to determine and and whether they are in the same decision direction, and the decision direction includes correct prediction and wrong prediction.

[0011] Optionally, calculate the sufficiency score and necessity score corresponding to the explanatory subsequence, specifically including: calculating the sufficiency score through the following formula: ; wherein is the sufficiency score, represents the prediction ability score corresponding to the th explanatory subsequence, represents the prediction ability score corresponding to the remaining explanatory subsequence after removing ; calculate the necessity score through the following formula: ; wherein is the necessity score, represents any one interaction data point in the explanatory subsequence, represents the minimum necessity score of, represents the sum of the necessity scores of all interaction data points with a necessity score less than 0.

[0012] Optionally, according to the causal contribution degree, use the multi-objective particle swarm optimization algorithm to generate the target explanatory subsequence, and perform tracking and explanation on the answer probability prediction value through the target explanatory subsequence, specifically including: dynamically stratifying the candidate subsequences according to the causal contribution degree to generate a hierarchical explanatory subsequence set, and the hierarchical explanatory subsequence set includes the first set layer, the second set layer, the third set layer, and the fourth set layer; updating the first set layer, the second set layer, and the third set layer through elite learning to obtain the first target explanatory subsequence; updating the fourth set layer through reverse learning to obtain the second target explanatory subsequence; using the first target explanatory subsequence and the second target explanatory subsequence as the target explanatory subsequence, and performing tracking and explanation on the answer probability prediction value through the target explanatory subsequence.

[0013] In the second aspect of the present application, a deep knowledge tracking explanation device is provided. The device includes an acquisition module and a processing module, wherein An acquisition module, configured to acquire the historical answering records and the questions to be predicted of a target student, and construct a learning interaction sequence corresponding to the target student according to the historical answering records in the order of answering time.

[0014] A processing module, configured to input the learning interaction sequence into a knowledge tracing model to output a predicted value of the answering probability that the target student correctly answers the question to be predicted; based on a preset strategy, perform a pre-screening operation on the learning interaction sequence to obtain candidate subsequences; calculate the causal contribution degrees of the candidate subsequences to the predicted value of the answering probability by combining an educational constraint method, where the educational constraint method includes a cognitive load reduction method and a time decay weighting method; according to the causal contribution degrees, use a multi-objective particle swarm optimization algorithm to generate a target explanatory subsequence, and perform a tracing explanation on the predicted value of the answering probability through the target explanatory subsequence.

[0015] In a third aspect of the present application, an electronic device is provided, including a processor, a memory, a user interface, and a network interface. The memory is used to store instructions, the user interface and the network interface are used to communicate with other devices, and the processor is used to execute the instructions stored in the memory so that the electronic device executes the method as described in any one of the above.

[0016] In a fourth aspect of the present application, a computer-readable storage medium is provided. The computer-readable storage medium stores a computer program, and the computer program is executed by a processor to perform the method as described in any one of the above.

[0017] One or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages: 1. Acquire the historical answering records and the questions to be predicted of a target student, and construct a learning interaction sequence corresponding to the target student according to multiple historical answering records; input the learning interaction sequence into a knowledge tracing model to output a predicted value of the answering probability that the target student answers the question to be predicted; according to the learning interaction sequence, generate multiple explanatory subsequences based on particle swarm coding; calculate multiple causal contribution degrees of the multiple explanatory subsequences to the predicted value of the answering probability; obtain a target causal contribution degree that meets a preset ranking condition among the multiple causal contribution degrees, and perform a tracing explanation on the predicted value of the answering probability based on particle swarm iteration and the explanatory subsequence corresponding to the target causal contribution degree. Thus, by generating explanatory subsequences through particle swarm coding and combining causal inference to calculate the causal contribution degrees of each subsequence to the predicted value of the answering probability, it can effectively solve the problem that the DLKT model cannot intuitively present which interactions mainly drive the probability prediction of a student's correct answer to a certain question, and further make the predicted value of the answering probability more transparent and provide a clear basis for personalized teaching.

[0018] 2. Initialize the candidate subsequences using a number of preset strategies, including but not limited to design problem-related strategies, prior-related strategies, response-related strategies, and distance-first strategies, etc., so as to improve the quality of the initial population, ensure that the initial particle swarm can cover more representative candidate sub-interaction sequences, and improve the effectiveness of the search space and the optimization efficiency.

[0019] 3. Calculate the prediction ability score to evaluate the influence degree of the interaction information contained in the explanatory subsequence on the prediction value of the response probability. At the same time, calculate the sufficiency score and necessity score corresponding to the explanatory subsequence to evaluate the internal information redundancy degree of the explanatory subsequence, and calculate the causal contribution degree of the explanatory subsequence to the prediction value of the response probability based on the prediction ability score, the sufficiency score, and the necessity score, so as to accurately identify and screen out the key interaction data points that have the greatest influence on the prediction value of the response probability of the knowledge tracking model, and improve the interpretability of the prediction. Description of the Drawings

[0020] Figure 1 is a schematic flowchart of a deep knowledge tracking explanation method and device provided by an embodiment of the present application; Figure 2 is a schematic diagram of a method for constructing an attribution quality function provided by an embodiment of the present application; Figure 3 is a schematic diagram of an iterative update method for learning interaction sequences provided by an embodiment of the present application; Figure 4 is a schematic diagram of the modules of a deep knowledge tracking explanation device provided by an embodiment of the present application; Figure 5 is a schematic diagram of the structure of an electronic device provided by an embodiment of the present application.

[0021] Description of the reference numerals: 41, acquisition module; 42, processing module; 501, processor; 502, communication bus; 503, user interface; 504, network interface; 505, memory. Detailed Embodiments

[0022] In order to enable those skilled in the art to better understand the technical solutions in this specification, the following will clearly and completely describe the technical solutions in the embodiments of this specification with reference to the accompanying drawings in the embodiments of this specification. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments.

[0023] The terms used in the following embodiments of this application are only for the purpose of describing specific embodiments and are not intended to limit this application. As used in the specification of this application, the singular forms "a", "an", "the", "above-mentioned", "said", and "this" are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used in this application refers to and includes any or all possible combinations of one or more of the listed items.

[0024] Hereinafter, the terms "first" and "second" are only used for descriptive purposes and should not be construed as implying or suggesting relative importance or implicitly indicating the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of such features. In the description of the embodiments of this application, unless otherwise specified, the meaning of "a plurality" is two or more.

[0025] In order to enable those skilled in the art to better understand the technical solutions of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings.

[0026] Please refer to Figure 1 , which shows a schematic flowchart of a deep knowledge tracing and explanation method and device provided by an embodiment of this application. The flowchart mainly includes the following steps: S101 to S105.

[0027] Step S101, obtain the historical answer records of the target student and the questions to be predicted, and construct a learning interaction sequence corresponding to the target student according to the historical answer records in the order of answer time.

[0028] Specifically, collect the historical answer records of multiple students. Taking the target student as an example for illustration, the target student can be any one of the multiple students. The historical answer records include the collection of questions answered by the target student during the historical learning process, as well as the correct or incorrect information of the target student's answer to each question. The collection of questions can include various different types of questions, such as mathematics type, physics type, chemistry type, and each question can involve different knowledge points. Construct the learning interaction sequence of the target student according to the historical answer records, that is, the questions he answered in the past , and the corresponding answer results to . Then the interaction sequence of the target student can be: , where , indicates whether the target student answered correctly .

[0029] Step S102: Input the learning interaction sequence into the knowledge tracing model to output the predicted value of the probability that the target student answers the question to be predicted correctly.

[0030] Specifically, assume the question to be predicted is , then input the learning interaction sequence into the knowledge tracing model, so as to output the predicted value of the probability that the target student answers the question to be predicted. The predicted value of the answering probability is the probability prediction that the target student answers correctly at the next moment on . At this time, the predicted value of the answering probability is . The explanatory subsequence of the model regarding is denoted as , specifically: ; Among them, is the -th explanatory subsequence in . represents the attribution quality function. Therefore, in the knowledge tracing model, the predicted value of the probability that the target student answers the question to be predicted is obtained by maximizing the attribution quality function.

[0031] Step S103: Based on a preset strategy, perform a pre-screening operation on the learning interaction sequence to obtain candidate subsequences.

[0032] Specifically, construct a particle swarm and correspond the particles in the particle swarm to the subsequences in the learning interaction sequence. One particle corresponds to one subsequence. Encode the particles in the particle swarm. During the encoding process, use binary or continuous numerical encoding methods to represent the selection situation of data points in the interaction sequence. For example, map each data point in the interaction sequence to a dimension of the particle position vector. 1 indicates that the data point is selected into the explanatory subsequence, and 0 indicates that it is not selected. The subsequence corresponding to the selected particle is used as the explanatory subsequence.

[0033] In a possible implementation manner, step S103 further includes: The preset strategy includes a hidden state observation generation strategy and a gradient prior generation strategy.

[0034] Specifically, the preset strategies include, but are not limited to: strategies based on hidden state observation, strategies based on gradient prior, strategies based on experience, etc. Among them, the strategy based on hidden state observation: Regardless of the model, for each time step, output the hidden knowledge state / correct answer probability of the student on the question to be predicted, and observe the change of the probability over time steps. For the time steps where the rising and falling trends of the change correspond to the positive and negative values of the predicted value to be explained, retain the interaction pairs corresponding to these time steps; The strategy based on gradient prior: Use the Gradient-based method to screen out the K interaction pairs with the greatest correlation with the predicted value of the answer probability to form an initial candidate subsequence; The strategy based on experience: Retain the interaction pairs in the input interaction sequence where the correct / incorrect answer situation is consistent with the predicted answer situation of the model output to be explained, and ignore those that are inconsistent. Further, ignore those with a large time step distance / ignore those that are conceptually irrelevant, respectively forming initial candidate subsequences. The random strategy has the advantage of maintaining diversity and can prevent falling into local optima to a certain extent.

[0035] In a possible implementation manner, step S103 further includes: sorting according to the length of the learning interaction sequence, and obtaining multiple candidate subsequences based on the sorting result; using particle swarm coding to assign corresponding particles to the multiple candidate subsequences; obtaining a target candidate subsequence, where the target candidate subsequence is any one of the multiple candidate subsequences; obtaining the position vector of the particle corresponding to the target candidate subsequence, and confirming whether the values of each dimension of the position vector meet the preset screening conditions; if the values of each dimension of the position vector meet the preset screening conditions, then use the target candidate subsequence as the explanatory subsequence.

[0036] Specifically, suppose the original length of the learning interaction sequence is t, and there are 2^t feasible combinations of candidate subsequences. At this time, taking any one of the 2^t candidate subsequences as an example, assign a corresponding particle to it, and the dimension of the particle is the same as the length of the learning interaction sequence. When generating the explanatory subsequence based on particle swarm coding, the explanation problem can be transformed into an optimization problem in a binary solution space. Obtain the position vector of the particle corresponding to the target candidate subsequence, and denote the values of each dimension of this position vector as , if , then it meets the preset screening conditions, indicating that the candidate subsequence at the th time step is selected as a factor; if , then it does not meet the preset screening conditions, indicating that the candidate subsequence at the th time step is not selected as a factor. Repeat the above process until all explanatory subsequences that meet the preset screening conditions are selected.

[0037] Step S104, calculate the causal contribution degree of the candidate subsequence to the predicted value of the answer probability in combination with the educational constraint method, and the educational constraint method includes the cognitive load reduction method and the time decay weighting method.

[0038] Specifically, to construct the attribution quality function, please refer to Figure 2 , which presents a schematic diagram of an attribution quality function construction method provided in an embodiment of the present application, and measures the causal contribution of each explanation subsequence to the answer probability prediction value through the attribution quality function to discover the causal relationship between the input and output of the knowledge tracking model, and one explanation subsequence corresponds to one causal contribution.

[0039] Regarding the cognitive load reduction method: According to educational theory, it is necessary to ensure the credibility of the explanation and avoid generating misleading explanations, such as over-reliance on certain irrelevant redundant features, and introduce sufficiency and necessity to verify the independent explanatory power of the candidate explanation subsequence; according to cognitive load theory, human memory capacity is limited, and it is necessary to reduce redundant information, optimize the presentation of explanations, and avoid information overload. Therefore, a necessity score is designed to suppress redundant information (such as 5 wrong reasons are presented at the same time, the two most critical ones are retained, and the ones that provide less information are considered redundant) to avoid repeated explanations of similar errors. A suitable attribution quality function can help the optimization algorithm select the globally optimal explanation subsequence. The present invention mainly considers the explanation attribution quality of the selected subsequence from two aspects, namely the overall explanation sufficiency and internal information redundancy of the explanation subsequence. Among them, for the overall explanation sufficiency: by calculating the predictive ability score corresponding to the explanation subsequence, the degree of predictive influence of the interactive information contained in the explanation subsequence on the predicted value of the answer probability can be evaluated based on the predictive ability score. Intervention means in causal inference can be used to utilize Arithmetic operations directly manipulate the values of input variables to study how the model predictions will change. The calculation formula is as follows: ; in, For the The predictive ability score corresponding to the explanatory subsequence, For the Explanation subsequence, Indicates intervention operation, Indicates that the knowledge tracking model is based only on To perform prediction operations, Indicates that the knowledge tracking model is based only on The output result obtained by making predictions is is the original predicted value of the answer probability, Used for judgment and Is it in the same decision-making direction, where the decision-making direction includes correct prediction and incorrect prediction; for the internal information redundancy: by calculating the sufficiency score and necessity score corresponding to the explanatory subsequence, and then evaluating the internal information redundancy in the explanatory subsequence based on the sufficiency score and necessity score. Among them, for the sufficiency score, based on the principle of causal sufficiency, it is ensured that the explanatory part contains the factors that affect the decision-making of the knowledge tracing model, and the non-explanatory part does not contain the factors that affect the decision-making of the knowledge tracing model. In order to avoid the overjustification effect, that is, overly pursuing the coincidence of prediction scores and wrongly introducing relevant but non-causal information, while ignoring other interpretable information. Therefore, based on the principle of causal sufficiency, the calculation formula of the sufficiency score is as follows: ; where, is the sufficiency score, represents the prediction ability score corresponding to the th explanatory subsequence, represents the prediction ability score corresponding to the remaining explanatory subsequence after removing ; When finding an explanatory subsequence with the best overall explanatory effect through causal sufficiency, the problem of redundancy follows. Because there are complex interactions, dependencies, collisions among the internal factors of a subsequence, as well as synergy effects and redundancy effects. The synergy effect means that the overall interpretability will increase due to the addition of factors, while the redundancy effect means that the addition of some factors does not bring an increase in interpretability. Therefore, it is necessary to introduce a necessity score to evaluate the contribution of each factor in the explanatory subsequence. Specifically, based on causal screening, when the presence of some factors in the explanatory subsequence does not provide more information for the target variable, or only provides less information, they are called redundant. Therefore, based on the redundancy effect, the necessity of each factor in the explanatory subsequence is measured: ; For the time decay weighting method: According to the forgetting curve theory, as time goes by, the influence of past learning behaviors on the current shows an exponential decay trend, that is, it is difficult for students' more distant historical experiences to have an impact on the current; introducing the exponential decay factor of time distance, using the time domain decay function to reduce the weight of the long-term time steps, and improving the necessity measurement function, the formula is: ; where, represents the time step where the target prediction is located, represents at the time step of the original interaction sequence, is the interval between the two time steps, is the decay factor.

[0040] The improved The necessity score metric is as follows: ; After that, the necessity score is calculated through the following formula: ; where is the necessity score, represents any interaction data point in the explanation subsequence, represents the minimum necessity score of represents the sum of the necessity scores of all interaction data points with necessity scores less than 0, represents when removing from then the prediction ability score of ; where, is the causal explanation quality of the th explanation subsequence for the predicted answer probability value represents the causal sufficiency of the th explanation subsequence for the predicted answer probability value represents the causal necessity of the th explanation subsequence for the predicted answer probability value

[0041]

[0041] Step S105, according to the causal contribution degree, use the multi-objective particle swarm optimization algorithm to generate the target explanation subsequence, and track and explain the predicted answer probability value through the target explanation subsequence.

[0042] Specifically, use the multi-objective particle swarm optimization algorithm to optimize the subsequence combination, ensuring both explanation sufficiency and consideration of explanation necessity, and generate the final explanation subsequence. Optimize the search space through the particle swarm algorithm. Given the knowledge tracking model, explore the optimal explanation subsequence, determine the explanation subsequence corresponding to the target causal contribution degree that meets the preset ranking conditions in the learning interaction sequence, use it as the optimal explanation subsequence, and update the learning interaction sequence based on the particle swarm iteration and the explanation subsequence corresponding to the target causal contribution degree.

[0043] In a possible implementation, step S105 further includes: dynamically hierarchizing the candidate subsequences according to the causal contribution degree to generate a hierarchical set of explanatory subsequences, where the hierarchical set of explanatory subsequences includes a first set layer, a second set layer, a third set layer, and a fourth set layer; updating the first set layer, the second set layer, and the third set layer through elitist learning to obtain a first target explanatory subsequence; updating the fourth set layer through reverse learning to obtain a second target explanatory subsequence; using the first target explanatory subsequence and the second target explanatory subsequence as the target explanatory subsequence, and tracking and explaining the answer probability prediction value through the target explanatory subsequence.

[0044] Specifically, a dynamic hierarchization strategy and a constraint handling technique are introduced: First, all particles in the population are sorted according to the fitness value and initially divided into upper and lower layers. The constraint handling technique is applied to the upper N / 2 particles to differentiate into a first set layer and a second set layer, namely, the elite layer G and the better layer. Specifically, taking the particle with the longest length in the upper layer as the boundary, find the particles with fitness values better than to form the elite layer G, representing short and good particles, and the remaining particles form the better layer; the constraint handling is also applied to the lower N / 2 particles, that is, taking the particle with the shortest length as the boundary, find the particles with fitness values worse than to represent long and bad particles and form the redundant layer, that is, the fourth set layer, and the remaining particles form the worse layer, that is, the third set layer. The first set layer, the second set layer, the third set layer, and the fourth set layer are collectively referred to as the hierarchical set of explanatory subsequences, and the hierarchical relationship of the hierarchical set of explanatory subsequences is as shown in Figure 2 the figure.

[0045] Divide and conquer, adopt the elitist retention strategy to maintain the elite layer, the better layer, and the worse layer G, and introduce the Opposition-Based Learning (OBL) strategy to handle redundancy. Specifically, different from the traditional OBL (Opposition-Based Learning) method, here a reverse solution is not generated for all particles, but only for the long and bad redundant particles in the population:

[0046] where r is a random number between 0 and 1. Compare the fitness values of the current solution and the reverse solution, and select the better one to enter the next generation to enhance the diversity of the population in the search space.

[0047] The ordinary particle swarm optimization algorithm is used to update both the better layer and the worse layer. The learning objects are the particles in the elite layer and the individual historical optimum, and the update formula is as follows:

[0048] Among them, is the velocity of the th particle at the th round of update, which determines the moving direction and amplitude of the particle in the next step. is the velocity of the th particle at the th round of update. is the inertia weight, which is used to control the influence degree of the current velocity of the particle. The larger the value, the more inclined the particle is to move along the original velocity direction to explore unknown areas; the smaller the value, the more easily it is affected by new solutions and adjusts the search direction. is the particle selected from the elite layer G. is the historical optimal position of the particle itself. is the elite learning factor, which is used to control the degree of approaching of the particle to . When the numerical value is large, the particle is more likely to approach the global optimal direction. is the individual learning factor, which is used to control the degree of approaching of the particle to its own historical optimal to promote the utilization of its own successful experience. is the position of the th particle in the learning interaction sequence at the th round of update. is the position of the th particle in the learning interaction sequence at the th round of update. Please refer to Figure 3 , which shows a schematic diagram of an iterative update method for a learning interaction sequence provided by an embodiment of the present application. Figure 3 In

[0049] By adopting the above method, this application obtains the historical answering records of the target student and the questions to be predicted, and constructs a learning interaction sequence corresponding to the target student based on multiple historical answering records; inputs the learning interaction sequence into a knowledge tracing model to output the predicted value of the answering probability of the target student for the questions to be predicted; generates multiple explanatory subsequences based on the learning interaction sequence and particle swarm coding; calculates multiple causal contribution degrees of the multiple explanatory subsequences to the predicted value of the answering probability; obtains the target causal contribution degree that meets the preset ranking condition among the multiple causal contribution degrees, and based on particle swarm iteration and the explanatory subsequence corresponding to the target causal contribution degree, conducts a tracing explanation on the predicted value of the answering probability. Thus, by generating explanatory subsequences through particle swarm coding and combining causal inference to calculate the causal contribution degrees of each subsequence to the predicted value of the answering probability, it can effectively solve the problem that the DLKT model cannot intuitively present which interaction sequences mainly drive the probability prediction of a student's correct answering of a certain question, and further make the predicted value of the answering probability more transparent and provide a clear basis for personalized teaching.

[0050] Please refer to Figure 4 , which shows a schematic diagram of the modules of a deep knowledge tracing explanation device provided by an embodiment of this application. The device includes an acquisition module 41 and a processing module 42, where, The acquisition module 41 is configured to obtain the historical answering records of the target student and the questions to be predicted, and construct a learning interaction sequence corresponding to the target student according to the historical answering records in the order of answering time.

[0051] The processing module 42 is configured to input the learning interaction sequence into a knowledge tracing model to output the predicted value of the answering probability that the target student correctly answers the questions to be predicted; perform a pre-screening operation on the learning interaction sequence based on a preset strategy to obtain candidate subsequences; calculate the causal contribution degrees of the candidate subsequences to the predicted value of the answering probability by combining an educational constraint method, and the educational constraint method includes a cognitive load reduction method and a time decay weighting method; generate a target explanatory subsequence by using a multi-objective particle swarm optimization algorithm according to the causal contribution degrees, and conduct a tracing explanation on the predicted value of the answering probability through the target explanatory subsequence.

[0052] In a possible implementation manner, the processing module 42 is configured to perform a pre-screening operation on the learning interaction sequence based on knowledge in the field of education to obtain candidate subsequences, specifically including: using particle swarm coding to assign corresponding particles to each subsequence in the learning interaction sequence; obtaining a target candidate subsequence, where the target candidate subsequence is any one subsequence among multiple learning interaction sequences; obtaining the position vector of the particle corresponding to the target candidate subsequence, and confirming whether each dimension value of the position vector meets the preset screening condition; if each dimension value of the position vector meets the preset screening condition, then use the target candidate subsequence as the candidate subsequence.

[0053] In a possible implementation, the preset policy includes a policy generated based on hidden state observation and a policy generated based on gradient prior.

[0054] In a possible implementation, the processing module 42 is used to calculate the causal contribution degree of the candidate subsequence pair to the answer probability prediction value by combining the pedagogy constraint method, specifically including: calculating the prediction ability score corresponding to the explanatory subsequence, where the prediction ability score is used to evaluate the prediction influence degree of the interaction information contained in the explanatory subsequence on the answer probability prediction value; calculating the sufficiency score and necessity score corresponding to the explanatory subsequence, where the sufficiency score and necessity score are used to evaluate the internal information redundancy degree in the explanatory subsequence; and calculating the causal contribution degree of the explanatory subsequence to the answer probability prediction value based on the prediction ability score, sufficiency score, and necessity score, and based on the time exponential decay function.

[0055] In a possible implementation, the processing module 42 is used to calculate the prediction ability score corresponding to the explanatory subsequence, specifically including: calculating the prediction ability score through the following formula: ; where is the prediction ability score corresponding to the th explanatory subsequence, is the th explanatory subsequence, represents the intervention operation, represents the operation of making the knowledge tracking model predict only based on ; represents the output result obtained by making the knowledge tracking model predict only based on ; is the original answer probability prediction value, is used to judge whether and are in the same decision direction, and the decision direction includes correct prediction and wrong prediction.

[0056] In a possible implementation, the processing module 42 is used to calculate the sufficiency score and necessity score corresponding to the explanatory subsequence, specifically including: calculating the sufficiency score through the following formula: ; where is the sufficiency score, represents the prediction ability score corresponding to the th explanatory subsequence, represents the prediction ability score corresponding to the remaining explanatory subsequence after removing ; calculating the necessity score through the following formula: ;

[0057] wherein is the necessity score represents any one of the interaction data points in the explanatory subsequence represents the minimum necessity score of represents the sum of the necessity scores for all interaction data points with a necessity score less than 0

[0058] In a possible implementation manner, the processing module 42 is configured to generate a target explanatory subsequence by using a multi-objective particle swarm optimization algorithm according to the causal contribution degree, and perform a tracking explanation on the answer probability prediction value through the target explanatory subsequence, specifically including: dynamically hierarchizing the candidate subsequences according to the causal contribution degree to generate a hierarchical explanatory subsequence set, where the hierarchical explanatory subsequence set includes a first set layer, a second set layer, a third set layer, and a fourth set layer; updating the first set layer, the second set layer, and the third set layer through elite learning to obtain a first target explanatory subsequence; updating the fourth set layer through reverse learning to obtain a second target explanatory subsequence; using the first target explanatory subsequence and the second target explanatory subsequence as the target explanatory subsequence, and performing a tracking explanation on the answer probability prediction value through the target explanatory subsequence

[0059] It should be noted that: when the device provided in the above embodiment realizes its functions, only the above division of each functional module is used for illustration. In practical applications, the above functions can be allocated to different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the device and method embodiments provided in the above embodiment belong to the same concept, and the specific implementation process can be seen in the method embodiment, which will not be elaborated here

[0060] This application also provides an electronic device. Referring to Figure 5 , Figure 5 is a schematic structural diagram of an electronic device provided in an embodiment of this application. The electronic device may include: at least one processor 501, at least one communication bus 502, a user interface 503, at least one network interface 504, and a memory 505

[0061] Among them, the communication bus 502 is used to realize the connection and communication between these components

[0062] Among them, the user interface 503 may include a display screen (Display) and a camera (Camera). Optionally, the user interface 503 may further include a standard wired interface and a wireless interface

[0063] Among them, the network interface 504 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface)

[0064] Among them, the processor 501 may include one or more processing cores. The processor 501 is connected to various parts within the entire server through various interfaces and lines. By running or executing instructions, programs, code sets, or instruction sets stored in the memory 505, and by calling the data stored in the memory 505, it performs various functions of the server and processes data. Optionally, the processor 501 may be implemented in at least one hardware form of digital signal processing (DSP), field-programmable gate array (FPGA), or programmable logic array (PLA). The processor 501 may integrate a combination of one or several of a central processing unit (CPU), a graphics processing unit (GPU), and a modem, etc. Among them, the CPU mainly processes the operating system, user interface, application programs, etc.; the GPU is responsible for rendering and drawing the content to be displayed on the display screen; the modem is used to process wireless communications. It can be understood that the above-mentioned modem may not be integrated into the processor 501 and may be implemented separately through a single chip.

[0065] Among them, the memory 505 may include random access memory (RAM) and may also include read-only memory. Optionally, the memory 505 includes a non-transitory computer-readable storage medium. The memory 505 can be used to store instructions, programs, code, code sets, or instruction sets. The memory 505 may include a program storage area and a data storage area. Among them, the program storage area can store instructions for implementing the operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-mentioned various method embodiments, etc.; the data storage area can store the data involved in the above-mentioned various method embodiments. The memory 505 is optionally also at least one storage device located far from the aforementioned processor 501. Refer to Figure 5 , the memory 505, as a computer storage medium, may include an operating system, a network communication module, a user interface module, and a deep knowledge tracking and interpretation application program.

[0066] In Figure 5In the electronic device shown, the user interface 503 is mainly used to provide an interface for the user to input data and obtain the data input by the user; and the processor 501 can be used to call the deep knowledge tracking and interpretation application program stored in the memory 505. When executed by one or more processors 501, the electronic device is caused to execute one or more of the methods described in the above embodiments. It should be noted that for the foregoing method embodiments, for the sake of simple description, they are all expressed as a series of action combinations. However, those skilled in the art should know that the present application is not limited by the described action sequence, because according to the present application, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to the present application.

[0067] The present application also provides a computer-readable storage medium storing instructions. When executed by one or more processors, the electronic device is caused to execute one or more of the methods described in the above embodiments.

[0068] In the above embodiments, the descriptions of the various embodiments have their own emphases. For the parts not detailed in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0069] In several implementation manners provided by the present application, it should be understood that the disclosed device can be implemented in other ways. For example, the device embodiments described above are only illustrative. For example, the division of units is only a logical function division. In actual implementation, there can be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some service interfaces. The indirect couplings or communication connections of the device or unit can be in electrical or other forms.

[0070] The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they can be located in one place, or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0071] In addition, in each embodiment of the present application, the functional units can be integrated in a processing unit, or each unit exists physically alone, or two or more units are integrated in one unit. The above integrated units can be implemented in the form of hardware or in the form of software functional units.

[0072] When the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable memory. Based on such understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a memory and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present application. The aforementioned memory includes: various media such as USB flash drives, mobile hard disks, magnetic disks, or optical discs that can store program codes.

[0073] The above are only exemplary embodiments disclosed in the present application and should not be used to limit the scope of the present application. That is, any equivalent changes and modifications made in accordance with the teachings of the present application are still within the scope covered by the present application.

[0074] The present application aims to cover any variations, uses, or adaptation changes of the present application, which follow the general principles of the present application and include common general knowledge or conventional technical means in the technical field not recorded in the present application.

Claims

1. A method for explaining deep knowledge tracing, characterized in that, The method includes: Obtaining the historical answering records of the target student and the question to be predicted, and constructing a learning interaction sequence corresponding to the target student according to the historical answering records in the order of answering time; Inputting the learning interaction sequence into a knowledge tracing model to output a predicted value of the answering probability that the target student correctly answers the question to be predicted; Based on a preset strategy, performing a pre-screening operation on the learning interaction sequence to obtain candidate subsequences; Calculating the causal contribution degree of the candidate subsequences to the predicted answering probability value by combining an educational constraint method, where the educational constraint method includes a cognitive load reduction method and a time decay weighting method; According to the causal contribution degree, using a multi-objective particle swarm optimization algorithm to generate a target explanatory subsequence, and performing a tracing explanation on the predicted answering probability value through the target explanatory subsequence.

2. The method according to claim 1, characterized in that, The performing a pre-screening operation on the learning interaction sequence based on a preset strategy to obtain candidate subsequences specifically includes: Using particle swarm encoding to assign corresponding particles to each subsequence in the learning interaction sequence; Obtaining a target candidate subsequence, where the target candidate subsequence is any one subsequence among multiple learning interaction sequences; Obtaining the position vector of the particle corresponding to the target candidate subsequence, and confirming whether the values of each dimension of the position vector satisfy a preset screening condition; If the values of each dimension of the position vector satisfy the preset screening condition, then taking the target candidate subsequence as the candidate subsequence.

3. The method according to claim 2, characterized in that The preset strategy includes a hidden state observation generation strategy and a gradient prior generation strategy.

4. The method according to claim 1, wherein The calculating the causal contribution degree of the candidate subsequences to the predicted answering probability value by combining an educational constraint method specifically includes: Calculating a prediction ability score corresponding to the explanatory subsequence, where the prediction ability score is used to evaluate the prediction influence degree of the interaction information contained in the explanatory subsequence on the predicted answering probability value; Calculating a sufficiency score and a necessity score corresponding to the explanatory subsequence, where the sufficiency score and the necessity score are used to evaluate the internal information redundancy degree in the explanatory subsequence; Based on the prediction ability score, the sufficiency score, and the necessity score, and based on a time exponential decay function, calculating the causal contribution degree of the explanatory subsequence to the predicted answering probability value.

5. The method according to claim 4, wherein The calculating the prediction ability score corresponding to the explanatory subsequence specifically includes: Calculating the prediction ability score through the following formula: ； wherein, is the prediction ability score corresponding to the th explanation subsequence, represents an intervention operation, represents an operation that makes the knowledge tracking model only based on for prediction, represents the output result obtained by making the knowledge tracking model only based on for prediction, is the original predicted answer probability value, is used to determine and are in the same decision direction, and the decision direction includes correct prediction and wrong prediction.

6. The method according to claim 5, characterized in that The calculating the sufficiency score and the necessity score corresponding to the explanatory subsequence specifically includes: Calculating the sufficiency score through the following formula: ； Among them, is the sufficiency score, represents the predictive ability score corresponding to the th explanatory subsequence, represents the predictive ability score corresponding to the remaining explanatory subsequences after removing ; Calculating the necessity score through the following formula: ； wherein is the necessity score representing any one of the interaction data points in the explanation subsequence denotes the minimum necessity score denotes the summation of the necessity scores for all the interaction data points with necessity scores less than 0 7. The method according to claim 4, characterized in that The generating a target explanatory subsequence using a multi-objective particle swarm optimization algorithm according to the causal contribution degree and performing a tracing explanation on the predicted answering probability value through the target explanatory subsequence specifically includes: Dynamically hierarchizing the candidate subsequences according to the causal contribution degree to generate a hierarchical explanatory subsequence set, where the hierarchical explanatory subsequence set includes a first set layer, a second set layer, a third set layer, and a fourth set layer; Update the first aggregation layer, the second aggregation layer, and the third aggregation layer through elite learning to obtain a first target explanation subsequence; Update the fourth aggregation layer through inverse learning to obtain a second target explanation subsequence; Use the first target explanation subsequence and the second target explanation subsequence as the target explanation subsequence, and perform a trace explanation on the answer probability prediction value through the target explanation subsequence.

8. A deep knowledge tracing and explanation device, characterized in that The device includes an acquisition module and a processing module, where the acquisition module is configured to acquire the historical answer records and the to-be-predicted questions of the target student, and construct a learning interaction sequence corresponding to the target student according to the historical answer records in the order of answer time; the processing module is configured to input the learning interaction sequence into the knowledge tracing model to output an answer probability prediction value for the target student to correctly answer the to-be-predicted question; perform a pre-screening operation on the learning interaction sequence based on a preset policy to obtain candidate subsequences; calculate the causal contribution degree of the candidate subsequences to the answer probability prediction value by combining an educational constraint method, and the educational constraint method includes a cognitive load reduction method and a time decay weighting method; generate a target explanation subsequence by using a multi-objective particle swarm optimization algorithm according to the causal contribution degree, and perform a trace explanation on the answer probability prediction value through the target explanation subsequence.

9. An electronic device, characterized in that, It includes a processor, a communication bus, a user interface, a network interface, and a memory. The memory is used to store instructions. The user interface and the network interface are used to communicate with other devices. The processor is used to execute the instructions stored in the memory to enable the electronic device to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions, and when the instructions are executed, the method according to any one of claims 1 to 7 is executed.

Images