A skill label generation method, device and equipment of a knowledge tracking model

Abstract

The application discloses a skill label generation method, device and equipment of a knowledge tracking model, comprising the following steps: constructing a knowledge tracking model, inputting interactive exercises and answer results into the knowledge tracking model; based on a skill label matrix, the comprehensive codes corresponding to the interactive exercises and the answer results are respectively obtained in the form of matrix projection, and the monotone self-attention mechanism processing is respectively performed to obtain the cross codes corresponding to the interactive exercises and the answer results, and then the evaluation result vector of the student knowledge state is obtained; based on the evaluation result vector of the student knowledge state and the comprehensive code of the current interactive exercise, the prediction result of whether the answer result is correct or not is obtained; the loss of the prediction result relative to the true answer result is obtained, the skill label matrix and the knowledge tracking model are trained; the skill label matrix after training is subjected to binaryzation processing, and the skill label result is stored. The application realizes high-performance and full-automatic labeling of the skills associated with the knowledge tracking exercises.

Description

A method, apparatus, and device for generating skill tags for a knowledge tracking model. Technical Field

[0001] This invention belongs to the field of Internet technology, and in particular relates to a method, apparatus and device for generating skill tags for a knowledge tracking model. Background Technology

[0002] With the rapid development of online learning platforms over the past few decades, it has become increasingly important to build models of students' mastery of knowledge concepts based on their past interactions. This is also the main idea behind knowledge tracing.

[0003] However, most existing knowledge tracing models use skill tags (concept tags, or knowledge components) to represent interactive exercises. This requires experts to manually label each exercise in the question bank with skills to indicate the skills needed to solve the problem, thus limiting the application of these models in certain scenarios. As the question banks of online education systems continue to expand, these knowledge tracing models are becoming increasingly difficult to apply to modern large-scale online education platforms.

[0004] Therefore, there is an urgent need to propose a skill tag generation method, device, and equipment for a knowledge tracking model that can realize fully automatic knowledge tracking with skill-related annotation of exercises. Summary of the Invention

[0005] The purpose of this invention is to provide a method, apparatus, and device for generating skill tags for a knowledge tracking model, so as to solve the problems existing in the prior art.

[0006] To achieve the above objectives, this invention provides a method for generating skill tags for a knowledge tracking model, comprising the following steps:

[0007] Construct a knowledge tracing model and input the interactive exercises and solutions in the time series into the knowledge tracing model;

[0008] Based on the skill tag matrix, skill coding matrix, skill diversity vector, and exercise difficulty vector, the comprehensive codes corresponding to the interactive exercises and their solutions are obtained by matrix projection.

[0009] The comprehensive encoding is processed by a monotonic self-attention mechanism to obtain the cross-encoding corresponding to the interactive exercise and the solution result;

[0010] Based on the cross-coding of past interactive exercises and solutions, an assessment vector of students' knowledge status is obtained.

[0011] Based on the comprehensive encoding of the student's knowledge status assessment result vector and the current interactive exercise, a prediction result of whether the answer is correct is obtained;

[0012] The loss of the predicted result relative to the actual solution result is obtained based on the joint loss function, and the skill tag matrix and knowledge tracking model are trained accordingly.

[0013] Based on the trained knowledge tracking model, all exercises are labeled with skills, the trained skill label matrix is ​​binarized, and the skill labeling results of all exercises are stored.

[0014] Optionally, the process of obtaining the comprehensive code corresponding to the interactive exercise and the solution result includes: obtaining the skill tag vector corresponding to the interactive exercise and the solution result respectively; obtaining the corresponding skill code vector based on the skill tag vector; and performing layer normalization processing by introducing the difficulty parameter of the interactive exercise based on the skill code vector to obtain the comprehensive code corresponding to the interactive exercise and the solution result.

[0015] Optionally, the process of obtaining the skill tag vectors corresponding to the interactive exercises and solutions includes: performing one-hot encoding on the interactive exercises and solutions to obtain the corresponding encoding results; and based on the encoding results, using a skill tag matrix, obtaining the skill tag vectors corresponding to the interactive exercises and solutions in the form of matrix projection.

[0016] Optionally, the process of obtaining the skill encoding vectors corresponding to the interactive exercises and solutions includes: based on the skill tag vectors corresponding to the interactive exercises and solutions, using a skill encoding matrix and a skill diversity vector, obtaining the skill encoding hybrid vectors corresponding to the interactive exercises and solutions in the form of matrix projection; processing the skill encoding hybrid vectors based on the ReLU activation function, and then performing scaling processing based on the corresponding skill tag vectors to obtain the skill encoding vectors corresponding to the interactive exercises and solutions.

[0017] Optionally, the process of training the skill tag matrix and knowledge tracking model includes: obtaining the loss value of the predicted result relative to the actual solution result based on the joint loss function; calculating the gradient optimization of minimizing the joint loss function based on the loss value, thereby optimizing the parameters of the skill tag matrix and knowledge tracking model; traversing each time series of the dataset in the knowledge tracking model, iteratively optimizing the skill tag matrix and knowledge tracking model parameters until the loss no longer decreases, thereby obtaining the trained skill tag matrix and knowledge tracking model.

[0018] Optionally, the trained skill label matrix can be binarized using the following formula:

[0019]

[0020] Where η is the hyperparameter used in the binarization process, Q is the initial skill tag matrix, Q' is the binarized skill tag matrix Q', i represents the i-th exercise, j represents the j-th skill, 1 indicates required, and 0 indicates not required.

[0021] The present invention also provides a skill tag generation device for a knowledge tracking model, comprising:

[0022] The basic coding module is used to obtain the interactive exercises and solutions in the current time series, and to obtain the comprehensive codes corresponding to the interactive exercises and solutions based on the skill tag matrix, skill coding matrix, skill diversity vector and exercise difficulty vector.

[0023] A cross-coding module, connected to the basic encoding module, is used to process the comprehensive encoding separately using a monotonic self-attention mechanism to obtain the cross-coding corresponding to the interactive exercise and the solution result.

[0024] The knowledge status assessment module, connected to the cross-coding module, is used to obtain the assessment result vector of the student's knowledge status based on the cross-coding corresponding to the past interactive exercises and solution results.

[0025] The solution result prediction module is connected to the knowledge state assessment module and is used to obtain a prediction result of whether the solution result is correct or not based on the comprehensive encoding of the student's knowledge state assessment result vector and the current interactive exercise.

[0026] The training module, connected to the basic encoding module, is used to train the skill tag matrix and knowledge tracking model based on the difference between the prediction results and the actual answer results;

[0027] The binarization module, connected to the training module, is used to perform skill labeling on all exercises based on the trained knowledge tracking model, perform binarization on the trained skill label matrix, and store the skill labeling results for all exercises.

[0028] Optionally, the skill tag generation device for the knowledge tracking model further includes a transfer module, which is connected to the binarization module and is used to obtain the model parameters and skill tag matrix of the first knowledge tracking model, and transfer the model parameters and skill tag matrix of the first knowledge tracking model to the initial values ​​of the second knowledge tracking model, wherein the training samples of the second knowledge tracking model include training samples of the corresponding second type of exercises.

[0029] The present invention also provides a skill tag generation device for a knowledge tracking model, comprising:

[0030] At least one processor, a memory, and a computer program stored in the memory; the computer program can be executed by the processor to implement a skill tag generation method for a knowledge tracking model.

[0031] The technical effects of this invention are as follows:

[0032] This invention provides a method, apparatus, device, and storage medium for generating skill tags in a knowledge tracking model. It establishes a monotonic attention mechanism knowledge tracking model, encodes learners' past interactions by integrating an initialized skill tag matrix and features such as question difficulty, calculates the student's knowledge state vector using the monotonic attention mechanism, and then predicts the correctness of the student's answer to the current question through a solution prediction module. By comparing the answer results in the knowledge tracking dataset, the loss function calculation result is obtained, thereby optimizing the aforementioned skill tag matrix and other model parameters. Finally, the skill tag matrix is ​​binarized to obtain the skill labeling result that best matches the student's sequential feedback.

[0033] To address the shortcomings of existing knowledge tracing algorithms, this invention effectively mines deep-seated information in time series, creatively applies an initialized skill tag matrix to a knowledge tracing model based on a monotonic attention mechanism, and designs an interactive coding method for knowledge tracing based on the skill tag matrix and the knowledge tracing model, achieving high-performance, fully automatic annotation of skills associated with knowledge tracing exercises.

[0034] This invention can achieve performance metrics similar to or even better than the latest models in knowledge tracking tasks without skill tags. Furthermore, the skill tags it provides are more consistent with student feedback compared to expert annotations and other model annotations. Attached Figure Description

[0035] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0036] Figure 1 is a flowchart illustrating a skill tag generation method for a knowledge tracking model according to an embodiment of the present invention;

[0037] Figure 2 is a diagram of the knowledge tracking module based on the improved multi-head self-attention mechanism in an embodiment of the present invention;

[0038] Figure 3 is a schematic diagram of the structure of a skill tag generation device for a knowledge tracking model in an embodiment of the present invention;

[0039] Figure 4 is a schematic diagram of the structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0040] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0041] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0042] Example 1

[0043] As shown in Figures 1-2, this embodiment provides a method, apparatus, and device for generating skill tags for a knowledge tracing model. Figure 1 is a flowchart illustrating the skill tag generation method for a knowledge tracing model provided in this embodiment, applied to a time series containing exercise tags and solution results. The method includes:

[0044] S1: Construct a knowledge tracking model based on the monotonic attention mechanism, which includes a basic encoding module, a cross-encoding module, a knowledge state evaluation module, a solution result prediction module, and a binarization module, and initialize model parameters including the skill tag matrix.

[0045] S2: Obtain the current time series and determine the skill tag vector corresponding to the i-th exercise in the skill tag matrix Q.

[0046] The user's interaction at time t can be represented as h. t =(q t ,r t ),in, This represents the index of the exercise answered by the user, where M represents the number of distinct exercises in the dataset; r t ∈{0,1} represents the true answer, where 1 indicates that the user answered the question correctly and 0 indicates that the answer was incorrect.

[0047] In the basic encoding module, the exercises and solutions are first one-hot encoded. For the exercises, one-hot encoding is performed directly to obtain a δ(q) array of length M. t For the solution result, the one-hot encoding of the exercise is offset according to the result to obtain a one-hot encoding a of length 2M. t Specifically, in Let M be the zero vector.

[0048] Furthermore, regarding exercise q t Based on the initialized skill tag matrix Extract the skill tag vector for this exercise using matrix projection: Where N represents the preset number of skills.

[0049] The one-hot encoding of the solution result is a t After uniformly dividing it into left and right vectors, extract the skill tag vectors separately according to the same logic, and then reassemble the two results into a 2M vector, which is the skill tag vector of the solution result. The formula is as follows:

[0050]

[0051]

[0052] In this invention, the skill tag matrix This is derived from a weight matrix processed by the sigmoid(σ) activation function, as shown in the following formula:

[0053] Q=σ(W Q )

[0054] S3: Based on the skill tag vector, obtain the skill encoding vector corresponding to each exercise and answer result.

[0055] Based on the skill tag vector, and according to the skill encoding matrix E and the skill diversity vector d, the skill encoding hybrid vector of the current exercise is obtained in the form of matrix projection. After processing with the ReLU activation function, and then scaling the vector according to the skill label vector, the skill encoding vector of the exercise can be obtained, as shown in the following formula:

[0056]

[0057] in, H is the preset skill code length; Skill tag vector The value at index i in the middle.

[0058] On the other hand, for the skill tag vector of the solution result, it is still evenly divided and the above operation is performed separately, and finally the two parts of the result are reassembled to obtain the skill encoding vector of the solution result, as shown in the following formula:

[0059]

[0060] S4: Based on the skill encoding vector, introduce the difficulty of the exercises and perform layer normalization to obtain the comprehensive encoding of the exercises and the solution results.

[0061] Based on the skill coding vector, a difficulty parameter for the exercises is introduced, and layer normalization is performed to obtain the comprehensive codes for both the exercises and their solutions. The formula for calculating the comprehensive code of the exercises is as follows:

[0062]

[0063] On the other hand, for the skill tag vector of the solution result, it is still evenly divided into left and right part vectors, and the above operation is performed on each part. Finally, the two parts are reassembled to obtain the comprehensive code of the solution result, as shown in the following formula:

[0064]

[0065] S5: Input the combined encoding of past interactions and current interaction exercises and solutions into the cross-coding module, and obtain the cross-coding of each position that integrates past interaction information through a monotonic self-attention mechanism.

[0066] In this step, the integrated coding vectors of the exercises and the integrated coding vectors of the results for each interaction are processed using the following monotonic self-attention mechanism:

[0067] The input matrix is ​​projected to obtain the query, key, and value, denoted as q, k, and v, where the length of the key vector k is d. k Self-attention employs the Scaled Dot-product Attention function. Taking the interaction at time τ (τ≤t) as an example, the attention coefficient α of this interaction to the interaction at time t is... t,τ for:

[0068]

[0069]

[0070] Where θ is a learnable parameter that controls the decay rate of the attention coefficient with distance. d(t,τ) is the distance index interacting between times t and τ, obtained by the following formula:

[0071]

[0072]

[0073] Where t and τ represent the positions of the two interactions, respectively, and γ t,t'The importance index represents the interaction t' (τ+1≤t'≤t) to the current interaction t. The sum of the importance indices of all interactions between t and τ to the current interaction is used to generate the final position code based on the interaction sequence interval and decay rate.

[0074] S6: Input the cross-encoding of past interactive exercises and solutions into the knowledge state assessment module. This module obtains the model's assessment vector of the learner's current knowledge state through a monotonic attention mechanism.

[0075] This module employs a monotonic attention mechanism. It uses the cross-encoding of exercises to perform matrix projection to obtain the query q and key k, and uses the cross-encoding of the solution results to perform matrix projection to obtain the value v. After processing through the same scaled dot product attention function as in step S5, it outputs an evaluation result vector representing the learner's current learning state.

[0076] S7: After concatenating the student's knowledge state vector with the comprehensive code of the exercise at time t, input the result prediction module to output the prediction result of whether the learner's answer to the current exercise is correct or not.

[0077] The prediction module consists of three sets of interconnected normalized layers, fully connected layers, and Dropout layers. The dimensions of the three fully connected layers decrease to 1 in sequence. The model finally outputs the predicted probability that the learner will answer the current exercise correctly.

[0078] S8: Calculate the loss of the predicted result relative to the true result using the joint loss function, and optimize the Q matrix and other model parameters.

[0079] The loss of the true results in the remaining knowledge tracking dataset is calculated using the following formula based on the obtained prediction results:

[0080]

[0081]

[0082]

[0083] L = L p +βL s +λL c

[0084] Where, r t It represents the actual solution results of student interaction at time t, with 1 for a correct answer and 0 for an incorrect answer. This represents the model's predicted solution at that moment; 1 indicates a correct answer and 0 indicates an incorrect answer. μ jLet be the difficulty index value of exercise j. β and λ are hyperparameters used to balance the proportions of the three losses.

[0085] Based on this loss value, the gradient optimization of minimizing the loss function is calculated, thereby optimizing the skill tag matrix Q and other parameters. The Q matrix and other model parameters are iteratively optimized for each time series in the knowledge tracking dataset until the loss no longer decreases, resulting in a trained Q matrix and knowledge tracking model.

[0086] S9: Binarize the skill label matrix Q obtained from training to obtain the final result of the model's skill labeling for all exercises.

[0087] The obtained Q matrix is ​​binarized using the following formula:

[0088]

[0089] Here, η is the hyperparameter used in the binarization process, set to 0.99, i represents the i-th exercise, j represents the j-th skill, 1 indicates required, and 0 indicates not required. After binarization, the final skill label matrix Q' is obtained, which stores the model's skill labeling results for each exercise. Elements can only be 0 or 1. Each column of this matrix represents an exercise, and each row represents a skill. The position with a value of 1 in the column represents the skill that needs to be mastered in advance to correctly answer the exercise.

[0090] This embodiment conducted extensive experiments to evaluate model performance. The hyperparameter settings are as follows: each dataset was randomly divided into three parts: 60% for training, 20% for validation, and 20% for testing. The validation set was used to verify model performance and to stop training early. All experiments used 5-fold cross-validation, and all models were evaluated by averaging the results from 5 trials on the test set. This embodiment used the Adam optimizer to train the model, setting the maximum number of training iterations to 300 and the learning rate to a range of 10. -5The training batch size was set to 200. To prevent overfitting, a Dropout layer was added to the model, with its parameter set to 0.05 during training. To verify the advantages of this embodiment in solving knowledge tracing tasks, experiments were conducted on four public datasets: ASSISTment2009, Statics2011, ASSISTment2017, and Junyi. Four state-of-the-art knowledge tracing models were also compared: Deep Knowledge Tracing (DKT), End-to-End Deep Knowledge Tracing (E2E-DKT), Dynamic Key-Value Memory Network (DKVMN), Interpretable Knowledge Tracing (IKT), and Attention-Based Knowledge Tracing (AKT-R). The AUC (Area Under Curve) metric was used to measure the model's performance; it is the area under the ROC curve and the coordinate axes. The closer the AUC value is to 1, the better the model's performance and the closer it is to reality. Table 1 shows the performance results of our proposed method compared with five state-of-the-art models on four public datasets, with the best results for each dataset highlighted in bold. The results demonstrate that the knowledge tracking model reconstructed based on the skill tags generated by this invention achieves performance similar to, or even higher than, state-of-the-art models using expert-annotated skill tags.

[0091] Table 1

[0092]

[0093] On the other hand, regarding the performance of the skill tags generated in this embodiment, the skill tags generated by E2E-DKT, and the skill tags annotated by experts in the knowledge tracking task, this embodiment conducted a comparative experiment on ASSIST2017 and Statics2011. The results are shown in Table 2, where the best results on each dataset have been bolded.

[0094] Table 2

[0095]

[0096] As can be seen, the skill tags generated in this embodiment perform significantly better than the skill tags generated by E2E-DKT or annotated by experts in knowledge tracking tasks.

[0097] Based on the same idea, one or more embodiments of this example also provide apparatus and devices corresponding to the above method, as shown in Figures 3 and 4.

[0098] Secondly, as shown in Figure 3, which is a structural schematic diagram of a skill tag generation device for a knowledge tracking model provided in this embodiment, applied to a sequence of exercises for knowledge tracking, the device includes:

[0099] The basic encoding module 401 obtains the sequence of interactive exercises and the sequence of solutions, and generates a comprehensive encoding vector corresponding to the exercise and solution for each interaction in the given sequence.

[0100] The cross-coding module 403 obtains cross-coding that integrates other interactive information based on the comprehensive encoding of the input time series exercises and solution results through a monotonic self-attention mechanism.

[0101] The knowledge status assessment module 405 obtains the model's assessment result vector of the learner's current knowledge status based on the cross-coding of past interactive exercises and solution results.

[0102] The solution prediction module 407, based on the student's knowledge state vector and the comprehensive encoding of the current exercise, obtains the model's prediction probability that the learner's solution result at the current moment is correct.

[0103] Training module 409 trains the knowledge tracking model based on the difference between the predicted results and the actual solutions, generating the optimal Q matrix and the target knowledge tracking model.

[0104] Binarization module 411 binarizes the Q matrix obtained during training to obtain the final result of the model's skill labeling for all exercises.

[0105] Optionally, the basic encoding module obtains the comprehensive encoding corresponding to the interactive exercise by matrix projection based on the current skill tag matrix, skill encoding matrix, skill diversity vector, and exercise difficulty vector.

[0106] Optionally, the device further includes a transfer module 413, which determines the model parameters and Q matrix in the target model; and transfers the model parameters or Q matrix to the initial values ​​of another knowledge tracing model, wherein the training samples of the other knowledge tracing model include training samples of another exercise.

[0107] In a third aspect, as shown in Figure 4, which is a structural schematic diagram of an electronic device provided in this embodiment, the device includes:

[0108] At least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method as described in the first aspect.

[0109] In a fourth aspect, based on the same idea, this embodiment also provides a non-volatile computer storage medium corresponding to the above method, which stores computer-executable instructions. When a computer reads the computer-executable instructions in the storage medium, the instructions cause one or more processors to execute the method described in the first aspect.

[0110] In the 1990s, improvements to a technology could be clearly distinguished as either hardware improvements (e.g., improvements to the circuit structure of diodes, transistors, switches, etc.) or software improvements (improvements to the methodology). However, with technological advancements, many methodological improvements today can be considered direct improvements to the hardware circuit structure. Designers almost always obtain the corresponding hardware circuit structure by programming the improved methodology into the hardware circuit. Therefore, it cannot be said that a methodological improvement cannot be implemented using hardware physical modules. For example, a Programmable Logic Device (PLD) (such as a Field Programmable Gate Array (FPGA)) is such an integrated circuit whose logic function is determined by the user programming the device. Designers can program and "integrate" a digital system onto a PLD themselves, without needing chip manufacturers to design and manufacture dedicated integrated circuit chips. Furthermore, nowadays, instead of manually manufacturing integrated circuit chips, this programming is mostly implemented using "logic compiler" software. Similar to the software compiler used in program development, the original code before compilation must be written in a specific programming language, called a Hardware Description Language (HDL). There are many HDLs, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, and RHDL (Ruby Hardware Description Language). Currently, the most commonly used are VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should understand that by simply performing some logic programming on the method flow using one of these hardware description languages ​​and programming it into an integrated circuit, the hardware circuit implementing the logical method flow can be easily obtained.

[0111] The controller can be implemented in any suitable manner. For example, it can take the form of a microprocessor or processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) executable by the (micro)processor, logic gates, switches, application-specific integrated circuits (ASICs), programmable logic controllers, and embedded microcontrollers. Examples of controllers include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicon Labs C8051F320. A memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art will also recognize that, in addition to implementing the controller in purely computer-readable program code form, the same functionality can be achieved by logically programming the method steps to make the controller take the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers. Therefore, such a controller can be considered a hardware component, and the means included therein for implementing various functions can also be considered as structures within the hardware component. Alternatively, the means for implementing various functions can be considered as both software modules implementing the method and structures within the hardware component.

[0112] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, a computer can be, for example, a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email device, game console, tablet computer, wearable device, or any combination of these devices.

[0113] For ease of description, the above devices are described in terms of function, divided into various units. Of course, in implementing this specification, the functions of each unit can be implemented in one or more software and / or hardware.

[0114] Those skilled in the art will understand that the embodiments of this specification can be provided as methods, systems, or computer program products. Therefore, the embodiments of this specification can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the embodiments of this specification can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0115] This specification is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this specification. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in one or more flowchart illustrations and / or one or more block diagrams.

[0116] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means that implement the functions specified in one or more flowcharts and / or one or more block diagrams.

[0117] These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, such that the instructions, which execute on the computer or other programmable apparatus, provide steps for implementing the functions specified in one or more flowcharts and / or one or more block diagrams.

[0118] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0119] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0120] Computer-readable media, including both permanent and non-permanent, removable and non-removable media, can store information using any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined in this embodiment, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0121] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0122] This specification can be described in the general context of computer-executable instructions that are executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform a specific task or implement a specific abstract data type. This specification can also be practiced in distributed computing environments, where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0123] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments of apparatus, devices, and non-volatile computer storage media are basically similar to the method embodiments, so the descriptions are relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0124] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0125] The above description is merely a preferred embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for generating skill tags for a knowledge tracking model, characterized in that, Includes the following steps: Construct a knowledge tracing model and input the interactive exercises and solutions in the time series into the knowledge tracing model; Based on the skill tag matrix, skill coding matrix, skill diversity vector, and exercise difficulty vector, the comprehensive codes corresponding to the interactive exercises and their solutions are obtained by matrix projection. The comprehensive encoding is processed by a monotonic self-attention mechanism to obtain the cross-encoding corresponding to the interactive exercise and the solution result; Based on the cross-coding corresponding to past interactive exercises and solutions, an evaluation result vector of the student's knowledge status is obtained; based on the comprehensive encoding of the student's knowledge status evaluation result vector and the current interactive exercise, a prediction result of whether the solution is correct or not is obtained; based on the joint loss function, the loss of the prediction result relative to the actual solution result is obtained, and the skill tag matrix and knowledge tracking model are trained. Based on the trained knowledge tracking model, all exercises are labeled with skills, the trained skill label matrix is ​​binarized, and the skill labeling results of all exercises are stored. The process of obtaining the comprehensive code corresponding to the interactive exercise and its solution results includes: obtaining the skill tag vectors corresponding to the interactive exercise and its solution results respectively; obtaining the corresponding skill encoding vector based on the skill tag vectors; and performing layer normalization processing by introducing the difficulty parameter of the interactive exercise based on the skill encoding vector to obtain the comprehensive code corresponding to the interactive exercise and its solution results. The process of obtaining the skill tag vector corresponding to the interactive exercise and its solution results includes: performing one-hot encoding on the interactive exercise and its solution results to obtain the corresponding encoding results; using the skill tag matrix, obtaining the skill tag vectors corresponding to the interactive exercise and its solution results respectively in the form of matrix projection based on the encoding results; and the process of obtaining the skill encoding vector corresponding to the interactive exercise and its solution results includes: using the skill tag vectors corresponding to the interactive exercise and its solution results, obtaining the skill encoding hybrid vector corresponding to the interactive exercise and its solution results respectively in the form of matrix projection using the skill encoding matrix and the skill diversity vector; processing the skill encoding hybrid vector based on the ReLU activation function, and then... The process of obtaining the skill encoding vector corresponding to the interactive exercise and the answer result by scaling the corresponding skill tag vector and performing scaling processing; the process of obtaining the evaluation result vector of the student's knowledge state based on the cross-encoding corresponding to the past interactive exercises and the answer result includes: using a monotonic attention mechanism, matrix projection is performed on the cross-encoding of the exercise to obtain the query and key, matrix projection is performed on the cross-encoding of the answer result to obtain the value, and then after processing by the scaling dot product attention function, the evaluation result vector representing the learner's current learning state is output; the process of obtaining the prediction result of whether the answer result is correct or not based on the evaluation result vector of the student's knowledge state and the comprehensive encoding of the current interactive exercise includes: concatenating the evaluation result vector of the student's knowledge state and the comprehensive encoding of the current interactive exercise and inputting it into the answer result prediction module, and outputting the prediction result of whether the learner answered the current exercise correctly or not; wherein, the answer result prediction module consists of three sets of connected layers: a normalized layer, a fully connected layer and a Dropout layer, wherein the dimensions of the three fully connected layers decrease to 1 in turn, and the model finally outputs the prediction probability that the learner answered the current exercise correctly.

2. The skill tag generation method for the knowledge tracking model according to claim 1, characterized in that, The process of training the skill tag matrix and knowledge tracking model includes: obtaining the loss value of the predicted result relative to the actual solution result based on the joint loss function; calculating the gradient optimization of minimizing the joint loss function based on the loss value, thereby optimizing the parameters of the skill tag matrix and knowledge tracking model; traversing each time series of the dataset in the knowledge tracking model, iteratively optimizing the parameters of the skill tag matrix and knowledge tracking model until the loss no longer decreases, thus obtaining the trained skill tag matrix and knowledge tracking model.

3. The skill tag generation method for the knowledge tracking model according to claim 1, characterized in that, The trained skill label matrix is ​​binarized using the following formula: Where η is the hyperparameter used in the binarization process, and Q is the initial skill tag matrix. ' It is the binarized skill tag matrix Q ' , where i represents the i-th exercise, j represents the j-th skill, 1 indicates required, and 0 indicates not required.

4. A skill tag generation device for a knowledge tracking model, characterized in that, The method for implementing any one of claims 1-3 comprises: a basic encoding module, configured to acquire interactive exercises and solutions in the current time series, and obtain comprehensive codes corresponding to the interactive exercises and solutions based on a skill tag matrix, a skill encoding matrix, a skill diversity vector, and an exercise difficulty vector; a cross-encoding module, connected to the basic encoding module, configured to perform monotonic self-attention mechanism processing on the comprehensive codes to obtain cross-encodings corresponding to the interactive exercises and solutions; and a knowledge state evaluation module, connected to the cross-encoding module, configured to obtain, based on the cross-encodings corresponding to past interactive exercises and solutions, a comprehensive encoding module for acquiring ... The system includes: a student knowledge status assessment result vector; a solution result prediction module, connected to the knowledge status assessment module, used to obtain a prediction result of whether the solution result is correct or not based on the student knowledge status assessment result vector and the comprehensive encoding of the current interactive exercise; a training module, connected to the basic encoding module, used to train the skill tag matrix and knowledge tracking model based on the difference between the prediction result and the actual solution result; and a binarization module, connected to the training module, used to perform skill labeling on all exercises based on the trained knowledge tracking model, perform binarization processing on the trained skill tag matrix, and store the skill labeling results of all exercises.

5. The skill tag generation device for the knowledge tracking model according to claim 4, characterized in that, The skill tag generation device for the knowledge tracking model further includes a transfer module, which is connected to the binarization module and is used to obtain the model parameters and skill tag matrix of the first knowledge tracking model, and transfer the model parameters and skill tag matrix of the first knowledge tracking model to the initial values ​​of the second knowledge tracking model, wherein the training samples of the second knowledge tracking model include the training samples of the corresponding second type of exercises.

6. A skill tag generation device for a knowledge tracking model, characterized in that, include: At least one processor, a memory, and a computer program stored in the memory; the computer program can be executed by the processor to implement the skill tag generation method of the knowledge tracking model as described in any one of claims 1 to 3.

Images