A multi-task based deep knowledge tracing method and apparatus
By adding two auxiliary tasks to the deep knowledge tracing model to predict the accuracy of current knowledge points and historical answers, the problem of complex modeling and poor performance of existing methods is solved, and a more accurate assessment of students' learning status in real-world scenarios is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- BEIJING CENTURY TAL EDUCATION TECH CO LTD
- Filing Date
- 2023-01-06
- Publication Date
- 2026-05-08
AI Technical Summary
Existing multi-task-based deep knowledge tracing methods suffer from poor model performance due to complex modeling, making them difficult to deploy and use in real-world scenarios. Furthermore, they lack the ability to improve students' future problem-solving performance without introducing additional information.
Based on the existing deep knowledge tracking model, two auxiliary tasks are added: one to predict the knowledge point corresponding to the question at the current moment, and the other to predict the student's overall historical answer accuracy. By jointly optimizing these tasks and the basic task, a small number of parameters are added to improve the evaluation effect.
Without introducing additional information, the model improved the assessment of students' future problem-solving performance by adding auxiliary tasks, simplified the model structure, and improved prediction accuracy.
Smart Images
Figure CN116090632B_ABST
Abstract
Description
Technical Field
[0001] This specification relates to the field of computer software technology, and in particular to a method, apparatus, electronic device, and storage medium for deep knowledge tracing based on multi-task. Background Technology
[0002] In educational settings, timely monitoring of all students' learning progress is crucial for designing effective teaching plans and rapidly improving student abilities. Knowledge tracing is a method that uses computer algorithms to automatically assess a student's learning ability based on their historical learning history. It can rapidly evaluate the learning progress of all students in real time, enabling personalized tutoring and the development of individualized learning paths. Multi-task-based deep knowledge tracing methods primarily rely on students' historical problem-solving records to infer their performance on future questions and to obtain their mastery of various knowledge points. Most current methods replace students' answers to questions with their answers to the corresponding knowledge points, then use machine learning / deep learning methods to predict students' future performance on those knowledge points. Other methods use additional information, such as the question text, student response time, number of attempts, or learn the question difficulty through learning. A few methods directly model the questions. Current multi-task-based deep knowledge tracing methods mainly include those based on psychological statistical models, traditional machine learning, and deep learning. However, these methods often suffer from complex modeling, resulting in poor model performance or models that are too large to be deployed in real-world scenarios. Therefore, how to improve the assessment effect of students' future test-taking performance by adding a small number of parameters without introducing additional information is a technical problem that needs to be solved. Summary of the Invention
[0003] The purpose of the embodiments in this specification is to address the above-mentioned problems by providing a multi-task-based deep knowledge tracing method, apparatus, electronic device, and storage medium.
[0004] To solve the above-mentioned technical problems, the embodiments in this specification are implemented as follows:
[0005] Firstly, a multi-task-based deep knowledge tracing method is proposed, including:
[0006] Get the problem representations at the current time t and all previous times [q] 0, …,q t Knowledge point representation [c] 0, …,c t ] and the correct / incorrect answer [r 0, …,r t ];
[0007] Input the problem representation [q]0, …,q t The knowledge point representation [c] 0, …,c t ] and the correctness or incorrectness of the answer [r 0, …,r t The knowledge is then processed by the trained knowledge tracking model; and,
[0008] In response to the processing of the knowledge tracking model, the correctness of the answer to the knowledge point at the next time step t+1 is obtained;
[0009] The knowledge tracing model is based on a specified knowledge tracing network and includes a first auxiliary task component for predicting the knowledge point corresponding to the question at the current time t and a basic knowledge tracing task component for predicting the correctness of the answer to the knowledge point at the next time t+1. The first auxiliary task component and the basic knowledge tracing task component jointly participate in the training of the knowledge tracing model.
[0010] Optionally, the first auxiliary task component includes a question encoding unit, a relation learning network unit, and a question labeling predictor; and the process by which the first auxiliary task component predicts the knowledge point corresponding to the current question includes:
[0011] Input the problem representation [q] 0, …,q t ] and the knowledge point representation [c 0, …,c t After passing through the title encoding unit, the relation learning network unit outputs the relation representation z. t ;
[0012] Input the relation representation z t The question labeling predictor outputs a predicted representation of the knowledge points.
[0013] Obtain the first loss function
[0014] Optionally, the knowledge tracking model further includes a second auxiliary task component for predicting the student's overall historical answer accuracy. The second auxiliary task component, together with the first auxiliary task component and the basic knowledge tracking task component, participates in the training of the knowledge tracking model.
[0015] Optionally, the knowledge state h required for prediction by the second auxiliary task component and the basic knowledge tracking task component is obtained. t ,include:
[0016] Input the correct or incorrect answer [r] 0, …,r t ] and the knowledge point representation [c0, …,c t [To the answer encoding unit, output the relational answer encoding representation x] t ;
[0017] Calculate the answer encoding representation x t The relation representation z t and the knowledge point representation c t joint characterization m t And input the joint representation m t The knowledge tracking network outputs the knowledge state h. t Wherein, the knowledge state h t The knowledge state h in the specified knowledge tracing network at the previous time step t-1 and the joint characterization m t Sure.
[0018] Optionally, the second auxiliary task component includes an answer encoding unit and a personalized prior knowledge predictor; and the process by which the second auxiliary task component predicts the student's historical overall answer accuracy includes:
[0019] Obtain the overall historical answer accuracy y within a specified time range. t The overall historical accuracy rate y t This is the ratio of the number of correct answers in history to the total number of answers in history.
[0020] Input the knowledge state h t The personalized prior knowledge predictor outputs a historical overall accuracy prediction representation of the student's answers.
[0021] Obtain the second loss function
[0022] Optionally, the basic knowledge tracking task component includes a knowledge tracking predictor; and the process by which the basic knowledge tracking task component predicts the correctness of the answer to a knowledge point at the next time step includes:
[0023] Input the knowledge state h t The knowledge tracking predictor outputs a prediction representation of the correctness of the answer to the knowledge point at the next time step.
[0024] Obtain the third loss function
[0025] Optionally, the knowledge tracing model is trained jointly by the first auxiliary task component, the second auxiliary task component, and the basic knowledge tracing task component; or, the knowledge tracing model is trained jointly by the first auxiliary task component and the basic knowledge tracing task component.
[0026] Optionally, the objective loss function for training the model is: Where β1 and β2 are hyperparameters used to adjust the weights of the first loss function and the second loss function; or, the target loss function for model training is... Wherein, β1 is a hyperparameter used to adjust the weights of the first loss function.
[0027] Optionally, the extreme value of the target loss function is calculated to obtain the optimal values of the parameters during the model training process.
[0028] Optionally, the knowledge-tracking network based on the specified knowledge includes recurrent neural networks, memory networks, graph networks, and / or attention networks.
[0029] Secondly, a multi-task-based deep knowledge tracing device is proposed, including:
[0030] The first module is capable of obtaining the problem representations [q0,…,q] for the current time t and all previous times. t Knowledge point representation [c0,…,c] t ] and the correct / incorrect answer results [r0,…,r t ];
[0031] The second module can input the problem representation [q0,…,q t The knowledge point representation [c] 0, …,c t ] and the correctness or incorrectness of the answer [r0,…,r t The knowledge is then processed by the trained knowledge tracking model; and,
[0032] The third module is able to respond to the processing of the knowledge tracking model and obtain the correctness of the answer at the next time step t+1.
[0033] The knowledge tracing model is based on a specified knowledge tracing network and includes a first auxiliary task component for predicting the knowledge point corresponding to the question at the current time t and a basic knowledge tracing task component for predicting the correctness of the answer to the knowledge point at the next time t+1. The first auxiliary task component and the basic knowledge tracing task component jointly participate in the training of the knowledge tracing model.
[0034] Optionally, the knowledge tracking model further includes a second auxiliary task component for predicting the student's overall historical answer accuracy. The second auxiliary task component, together with the first auxiliary task component and the basic knowledge tracking task component, participates in the training of the knowledge tracking model.
[0035] Thirdly, an electronic device is proposed, comprising: a processor; and
[0036] A memory is configured to store computer-executable instructions, which, when executed, cause the processor to perform the multitasking-based deep knowledge tracing method described in the first aspect.
[0037] Fourthly, a computer-readable storage medium is proposed that stores one or more programs, which, when executed by an electronic device including multiple applications, cause the electronic device to perform the multi-task-based deep knowledge tracing method described in the first aspect.
[0038] This manual can achieve at least the following technical effects:
[0039] This invention adds two auxiliary tasks to the basic task of predicting the correctness of students' answers to questions. These tasks simultaneously predict the knowledge points corresponding to the questions at the current moment and the student's overall historical accuracy at each time step. By optimizing these two auxiliary tasks along with the basic task, the evaluation effect of students' future problem-solving performance can be improved by adding a small number of parameters without introducing additional information. Attached Figure Description
[0040] To more clearly illustrate the technical solutions in the embodiments or prior art of this specification, the drawings used in the description of the embodiments or prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0041] Figure 1 This is one of the schematic diagrams of a multi-task-based deep knowledge tracing method provided in the embodiments of this specification.
[0042] Figure 2 This is the second schematic diagram of a multi-task-based deep knowledge tracing method provided in the embodiments of this specification.
[0043] Figure 3 This is the third schematic diagram of a multi-task-based deep knowledge tracing method provided in the embodiments of this specification.
[0044] Figure 4This is the fourth schematic diagram of a multi-task-based deep knowledge tracing method provided in the embodiments of this specification.
[0045] Figure 5 This is the fifth schematic diagram of a multi-task-based deep knowledge tracing method provided in the embodiments of this specification.
[0046] Figure 6 This is the sixth schematic diagram of a multi-task-based deep knowledge tracing method provided in the embodiments of this specification.
[0047] Figure 7 This is the seventh schematic diagram of a multi-task-based deep knowledge tracing method provided in the embodiments of this specification.
[0048] Figure 8 This is a schematic diagram of a multi-task-based deep knowledge tracking device provided in an embodiment of this specification.
[0049] Figure 9 This is a schematic diagram of the structure of an electronic device provided as an embodiment of this specification. Detailed Implementation
[0050] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this specification.
[0051] The following section provides a detailed description of a knowledge tracking scheme covered in this manual through specific examples.
[0052] The purpose of this invention is to overcome the limitations of current multi-task-based deep knowledge tracing methods, such as complex modeling, poor model performance, and difficulty in deploying them in real-world scenarios. It aims to improve the assessment of students' future problem-solving performance by adding a small number of parameters without introducing additional information. Knowledge tracing, simply put, uses algorithms to track and simulate the learning status of students on online education platforms, achieving the personalized learning effect of traditional offline education. Currently, the mainstream research methods include Bayesian knowledge tracing (BKT) and deep learning-based knowledge tracing (DKT). BKT is the earliest knowledge tracing model, a Hidden Markov Model (HMM) with latent variables. It builds a separate HMM model for each knowledge point to predict students' mastery of specific knowledge points, meaning the model does not consider the correlation between knowledge points. DKT models, especially deep knowledge tracing networks based on Long Short-Term Memory (LSTM) or Recurrent Neural Networks (RNN), not only have better predictive capabilities than BKT models but also do not require expert annotation of the knowledge points in the exercises. However, recurrent neural networks store all memories in a hidden vector, making it difficult to accurately remember sequences exceeding hundreds of time steps.
[0053] Example 1
[0054] like Figure 1 The diagram shown is the original framework of Deep Knowledge Tracking (DKT), a deep learning-based, multi-task-based deep knowledge tracing method using a Long Short-Term Memory (LSTM) network as an example. It can be seen that the Response Encoder module encodes the knowledge point ID (c) at time t. t And the correctness or incorrectness of the answer at time t. t The latent vector x obtained after encoding into a dense vector t After inputting into the Long Short-Term Memory (LSTM) network, the KT Predictor predicts the correctness of the answer at time t+1. The technical approach of this invention is... Figure 1Building upon the original method, this invention employs multi-task optimization to enhance knowledge tracing performance. Specifically, in addition to the basic task of predicting the correctness of students' answers, two auxiliary tasks are added: predicting the knowledge point corresponding to the question at the current moment and predicting the student's overall historical accuracy at each time step. By optimizing these two auxiliary tasks along with the basic task, the evaluation of students' future performance can be improved with a small number of parameters, without introducing additional information. It should be noted that this embodiment of the invention is a further optimization of the deep learning-based knowledge tracing (DKT) method. However, any method that improves knowledge tracing performance by adding auxiliary tasks to the basic task of predicting the correctness of questions satisfies the technical features of this invention and is also within the scope of protection of this invention.
[0055] This embodiment utilizes multitasking to... Figure 1 Based on the original framework of predicting students' future answers, i.e., the basic task, such as... Figure 2 As shown, two auxiliary tasks are added to improve the performance of the basic task. The first auxiliary task predicts the knowledge point corresponding to the question at the current moment; the second auxiliary task predicts the student's overall historical answer accuracy at each time step. For the first auxiliary task, the input is the question representation q. t and the knowledge point representation c t After passing through the title encoding unit, the relation representation z is obtained through the relation learning network unit. t The data is then input into the question labeling and prediction system to obtain the predicted representation of the knowledge points. For the second auxiliary task, input the correctness result r of the question. t and knowledge point representation c t To the answer encoding unit, output the relational answer encoding representation x. t Then, calculate the answer encoding representation x. t Relational representation z t and knowledge point representation c t joint characterization m t And input the joint representation m t Output the knowledge state h to the knowledge tracing network. t Then, the knowledge state h t Input is fed into a personalized prior knowledge predictor, which outputs a representation of the predicted accuracy of past question answers. For the basic task, the input parameters, in addition to the original knowledge point ID at time t, are c. t And the correctness or incorrectness of the answer at time t. t During the intermediate process, corresponding parameters based on the first and second auxiliary tasks were also added, namely the input knowledge state h. tThe KT Predictor, a knowledge tracing predictor, outputs a predicted representation of the correctness of the answers to the knowledge points at the next time step. Figure 2 In this context, QuestionEncoder represents the question encoding unit used to encode the question ID and knowledge point ID into a dense vector; Response Encoder represents the answer encoding unit used to encode the knowledge point ID and answer ID into a dense vector; Relation Network represents the relation network used to predict the current knowledge point using historical question-answering records; QT Predictor, or Question TaggingPredictor, represents the question tagging predictor used to predict the current knowledge point for the current question; and IK Predictor represents the personalized prior knowledge predictor used to predict the overall historical accuracy of answering questions. Therefore, as... Figure 3 The diagram shown is a schematic representation of a multi-task-based deep knowledge tracing method according to an embodiment of the present invention.
[0056] The method includes:
[0057] S1: Obtain the problem representations for the current time t and all previous time points [q] 0, …,q t Knowledge point representation [c] 0, …,c t ] and the correct / incorrect answer [r 0, …,r t Specifically, the question representation and the knowledge point representation can be the question ID and the knowledge point ID, respectively; the correctness of the answer generally refers to whether the answer is correct or incorrect, with 1 for correct and 0 for incorrect.
[0058] S2: Input the problem representation [q] 0, …,q t The knowledge points are represented as [c0,…,c] t ] and the correctness or incorrectness of the answer [r 0, …,r t The knowledge tracking model is then processed after training; wherein the knowledge tracking model is based on a specified knowledge tracking network and includes a first auxiliary task component for predicting the knowledge point corresponding to the question at the current time t and a basic knowledge tracking task component for predicting the correctness of the answer to the knowledge point at the next time t+1; the first auxiliary task component and the basic knowledge tracking task component jointly participate in the training of the knowledge tracking model.
[0059] Optionally, the knowledge tracing network based on the specified network includes recurrent neural networks, memory networks, graph networks, and / or attention networks. Specifically, this embodiment of the invention uses a Long Short-Term Memory (LSTM) network as an example of a knowledge tracing network.
[0060] Optionally, the first auxiliary task component includes a question encoding unit, a relation learning network unit, and a question labeling predictor; and the process by which the first auxiliary task component predicts the knowledge points corresponding to the current question is as follows: Figure 4 As shown, it includes:
[0061] S41: Input the problem representation [q0,…,q t ] and the knowledge point representation [c0,…,c t After passing through the title encoding unit, the relation learning network unit outputs the relation representation z. t Specifically, z t =Transformer(q) t ,c t ).
[0062] S42: Input the relation representation z t The question labeling predictor outputs a predicted representation of the knowledge points. Specifically,
[0063] S43: Obtain the first loss function
[0064] Optionally, another implementation of the present invention is that the knowledge tracking model further includes a second auxiliary task component for predicting the student's overall historical answer accuracy, wherein the second auxiliary task component, together with the first auxiliary task component and the basic knowledge tracking task component, participates in the training of the knowledge tracking model.
[0065] Optionally, the knowledge state h required for prediction by the second auxiliary task component and the basic knowledge tracking task component is obtained. t ,include:
[0066] S51: Input the correctness of the answer [r0,…,r t ] and the knowledge point representation [c0,…,c t [To the answer encoding unit, output the relational answer encoding representation x] t .
[0067] S52: Calculate the answer encoding representation x t The relation representation z t and the knowledge point representation c t joint characterization m t And input the joint representation m t The knowledge tracking network outputs the knowledge state h. t It should be noted here that the knowledge state h t It is the knowledge state h in the specified knowledge tracing network at the previous time step.t-1 and the joint characterization m t Determined. For example, in the case where a Long Short-Term Memory (LSTM) network is used as the knowledge tracing network, h... t =LSTM(h t-1 ,m t ).
[0068] Optionally, the second auxiliary task component includes an answer encoding unit and a personalized prior knowledge predictor; and the process by which the second auxiliary task component predicts the historical accuracy rate of answering questions, such as... Figure 6 As shown, it includes:
[0069] S61: Obtain the overall historical answer accuracy y within the specified time range. t The overall historical accuracy rate y t This is the ratio of the number of correct answers in history to the total number of answers in history.
[0070] S62: Input the knowledge state h t The personalized prior knowledge predictor outputs a predicted representation of the student's overall historical answer accuracy. Specifically,
[0071] S63: Obtain the second loss function
[0072] Optionally, the basic knowledge tracking task component includes a knowledge tracking predictor; and the process by which the basic knowledge tracking task component predicts the correctness of the question at the next moment, such as... Figure 7 As shown, it includes:
[0073] S71: Input the knowledge state h t The knowledge tracking predictor outputs a predictive representation of the correctness of the answer to the knowledge point at the next time step. Specifically,
[0074] S72: Obtain the third loss function
[0075] S3: In response to the processing of the knowledge tracking model, obtain the correctness of the knowledge point answer at the next time step t+1.
[0076] Optionally, the knowledge tracking model is trained by combining the first auxiliary task component and the basic knowledge tracking task component, or by combining the first auxiliary task component, the second auxiliary task component, and the basic knowledge tracking task component. It can be seen that, in addition to the latent vectors learned through the relation network influencing the input of the basic task component, the first and second auxiliary task components jointly optimize the model parameters through backpropagation of the loss function, thereby affecting the prediction performance of the basic task component.
[0077] Optionally, the objective function for training the model is: Wherein, β1 and β2 are hyperparameters used to adjust the weights of the first loss function and the second loss function. It should be understood that when the knowledge tracing model only includes a first auxiliary task component and a basic task component, the first auxiliary task component and the basic knowledge tracing task component jointly train the knowledge tracing model; the corresponding objective function for model training is... Wherein, β1 is a hyperparameter used to adjust the weights of the first loss function.
[0078] Optionally, the extreme value of the target loss function is calculated to obtain the optimal values of the parameters during the model training process.
[0079] This invention adds two auxiliary tasks to the basic task of predicting the correctness of students' answers to questions. These tasks simultaneously predict the knowledge points corresponding to the current question and the student's overall historical accuracy at each time step. By optimizing these two auxiliary tasks along with the basic task, the evaluation effect of students' future problem-solving performance can be improved by adding a small number of parameters without introducing additional information.
[0080] Example 2
[0081] Figure 8 This is a schematic diagram of a multi-task-based deep knowledge tracing device 800 provided as an embodiment of this specification. Please refer to... Figure 8 In one embodiment, a multi-task-based deep knowledge tracing device 800 includes:
[0082] The first module 801 is capable of obtaining the problem representations [q0,…,q] for the current time t and all previous times. t Knowledge point representation [c0,…,c] t ] and the correct / incorrect answer results [r0,…,r t ].
[0083] The second module 802 can input the problem representation [q0,…,q] t The knowledge points are represented as [c0,…,c]t ] and the correctness or incorrectness of the answer [r0,…,r t The knowledge is then processed by the trained knowledge tracking model; and,
[0084] The third module 803 is able to respond to the processing of the knowledge tracking model and obtain the correctness of the knowledge point answer at the next time step t+1.
[0085] The knowledge tracing model is based on a specified knowledge tracing network and includes a first auxiliary task component for predicting the knowledge point corresponding to the question at the current time t and a basic knowledge tracing task component for predicting the correctness of the answer to the knowledge point at the next time t+1. The first auxiliary task component and the basic knowledge tracing task component jointly participate in the training of the knowledge tracing model. And / or, the knowledge tracing model further includes a second auxiliary task component for predicting the student's historical overall answer accuracy rate, and the second auxiliary task component, together with the first auxiliary task component and the basic knowledge tracing task component, jointly participate in the training of the knowledge tracing model.
[0086] It should be understood that the multi-task-based deep knowledge tracing device 800 in the embodiments of this specification can also perform... Figures 1 to 7 A method for executing a multi-task-based deep knowledge tracing device (or apparatus) is described, and the implementation of the multi-task-based deep knowledge tracing device (or apparatus) in... Figures 1 to 7 The functionality of the example shown will not be elaborated upon here.
[0087] Example 3
[0088] Figure 9 This is a schematic diagram of the structure of an electronic device according to one embodiment of this specification. Please refer to it. Figure 9 At the hardware level, the electronic device includes a processor, and optionally also includes an internal bus, a network interface, and memory. The memory may include main memory, such as high-speed random-access memory (RAM), or non-volatile memory, such as at least one disk drive. Of course, the electronic device may also include other hardware required for other business operations.
[0089] The processor, network interface, and memory can be interconnected via an internal bus, which can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, etc. This bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 9 The symbol is represented by a single double-headed arrow, but this does not mean that there is only one bus or one type of bus.
[0090] Memory is used to store programs. Specifically, programs may include program code, which includes computer operation instructions. Memory may include main memory and non-volatile memory, and provides instructions and data to the processor.
[0091] The processor reads the corresponding computer program from non-volatile memory into main memory and then executes it, forming a shared resource access control mechanism at the logical level. The processor executes the program stored in memory and specifically performs the following operations:
[0092] Obtain the problem representations [q0,…,q] for the current time t and all previous time points. t Knowledge point representation [c0,…,c] t ] and the correct / incorrect answer results [r0,…,r t ];
[0093] Input the problem representation [q0,…,q t The knowledge points are represented as [c0,…,c] t ] and the correctness or incorrectness of the answer [r0,…,r t The knowledge is then processed by the trained knowledge tracking model; and,
[0094] In response to the processing of the knowledge tracking model, the correctness of the answer to the knowledge point at the next time step t+1 is obtained;
[0095] The knowledge tracing model is based on a specified knowledge tracing network and includes a first auxiliary task component for predicting the knowledge point corresponding to the question at the current time t and a basic knowledge tracing task component for predicting the correctness of the answer to the knowledge point at the next time t+1. The first auxiliary task component and the basic knowledge tracing task component jointly participate in the training of the knowledge tracing model. And / or, the knowledge tracing model further includes a second auxiliary task component for predicting the student's historical overall answer accuracy rate, and the second auxiliary task component, together with the first auxiliary task component and the basic knowledge tracing task component, jointly participate in the training of the knowledge tracing model.
[0096] The above is as described in this instruction manual. Figures 1 to 7 The multi-task-based deep knowledge tracing method disclosed in the illustrated embodiments can be applied to or implemented by a processor. The processor may be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method can be completed by integrated logic circuits in the processor's hardware or by instructions in software form. The processor can be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it can also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this specification. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this specification can be directly manifested as execution by a hardware decoding processor, or execution by a combination of hardware and software modules in the decoding processor. The software module can reside in a mature storage medium in the field, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory, and the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method.
[0097] Of course, in addition to the software implementation, the electronic devices in the embodiments of this specification do not exclude other implementation methods, such as logic devices or a combination of hardware and software, etc. That is to say, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.
[0098] Example 4
[0099] This specification also provides an embodiment of a computer-readable storage medium that stores one or more programs, the programs including instructions that, when executed by a portable electronic device including multiple applications, enable the portable electronic device to perform... Figures 1 to 7 The illustrated embodiment uses a multi-task-based deep knowledge tracing method, specifically for performing the following methods:
[0100] Obtain the problem representations [q0,…,q] for the current time t and all previous time points. t Knowledge point representation [c0,…,c] t ] and the correct / incorrect answer results [r0,…,r t ];
[0101] Input the problem representation [q0,…,q t The knowledge points are represented as [c0,…,c] t ] and the correctness or incorrectness of the answer [r0,…,r t The knowledge is then processed by the trained knowledge tracking model; and,
[0102] In response to the processing of the knowledge tracking model, the correctness of the answer to the knowledge point at the next time step t+1 is obtained;
[0103] The knowledge tracing model is based on a specified knowledge tracing network and includes a first auxiliary task component for predicting the knowledge point corresponding to the question at the current time t and a basic knowledge tracing task component for predicting the correctness of the answer to the knowledge point at the next time t+1. The first auxiliary task component and the basic knowledge tracing task component jointly participate in the training of the knowledge tracing model. And / or, the knowledge tracing model further includes a second auxiliary task component for predicting the student's historical overall answer accuracy rate, and the second auxiliary task component, together with the first auxiliary task component and the basic knowledge tracing task component, jointly participate in the training of the knowledge tracing model.
[0104] In summary, the above description is merely a preferred embodiment of this specification and is not intended to limit the scope of protection of this specification. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this specification should be included within the scope of protection of this specification.
[0105] 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, a laptop computer, a cellular phone, a camera phone, a smartphone, a personal digital assistant, a media player, a navigation device, an electronic data carrier device, a game console, a tablet computer, a wearable device, or any combination of these devices.
[0106] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, 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 herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.
[0107] 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.
[0108] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.
Claims
1. A deep knowledge tracing method based on multi-tasks, characterized in that, include: Get the current time and the problem representation of all previous moments. Knowledge point representation and the correct / incorrect answer ; Input the question representation The knowledge point representation and the results of the correct or incorrect answers. The knowledge tracking model, after training, will then process the information. as well as, In response to the processing of the knowledge tracing model, the next time step is obtained. Correct or incorrect answers to the knowledge points; The knowledge tracing model is based on a specified knowledge tracing network, including features for predicting the current time. The first auxiliary task component corresponding to the knowledge point in the question and the component used to predict the next moment. A basic knowledge tracking task component for answering knowledge points correctly or incorrectly; the first auxiliary task component and the basic knowledge tracking task component jointly participate in the training of the knowledge tracking model; The knowledge tracing model further includes a second auxiliary task component for predicting the student's overall historical answer accuracy. The second auxiliary task component, together with the first auxiliary task component and the basic knowledge tracing task component, participates in the training of the knowledge tracing model. Specifically, the second auxiliary task component and the basic knowledge tracking task component obtain the knowledge state required for prediction. ,include: Enter the correct or incorrect answer. and the representation of the knowledge points To the answer encoding unit, output the relational answer encoding representation. ; Calculate the answer encoding representation Relationship representation and the representation of the knowledge points joint representation and input the joint representation The knowledge tracking network outputs the knowledge status. ; wherein, the knowledge state The knowledge state in the specified knowledge tracing network at the previous time step and the joint characterization Sure.
2. The deep knowledge tracing method based on multi-tasks according to claim 1, characterized in that, The first auxiliary task component includes a question encoding unit, a relation learning network unit, and a question labeling predictor; And, the process by which the first auxiliary task component predicts the knowledge point corresponding to the current question includes: Input the question representation and the representation of the knowledge points After reaching the question encoding unit, the relation learning network unit outputs a relation representation. ; Input the relation representation The question labeling predictor outputs a predicted representation of the knowledge points. ; Obtain the first loss function .
3. The deep knowledge tracing method based on multi-tasks according to claim 2, characterized in that, The basic knowledge tracing task components include a knowledge tracing predictor; And, the process by which the basic knowledge tracking task component predicts the correctness of the answer to the knowledge point at the next moment includes: Input the knowledge status The knowledge tracking predictor outputs a prediction representation of the correctness of the answer to the knowledge point at the next time step. ; Obtain the third loss function .
4. The deep knowledge tracing method based on multi-tasks according to claim 3, characterized in that, The second auxiliary task component includes an answer encoding unit and a personalized prior knowledge predictor; And, the process by which the second auxiliary task component predicts the student's overall historical accuracy includes: Obtain the overall historical answer accuracy within a specified time range. The overall accuracy rate of historical responses This is the ratio of the number of correct answers in history to the total number of answers in history. Input the knowledge status The personalized prior knowledge predictor outputs a historical overall accuracy prediction representation of the student's answers. ; Obtain the second loss function .
5. The deep knowledge tracing method based on multi-tasks according to claim 4, characterized in that, The knowledge tracking model is trained by the first auxiliary task component, the second auxiliary task component, and the basic knowledge tracking task component.
6. The deep knowledge tracing method based on multi-tasks according to claim 5, characterized in that, The objective loss function for training the model is: ;in, and These are hyperparameters used to adjust the weights of the first loss function and the second loss function.
7. The deep knowledge tracing method based on multi-tasks according to claim 3, characterized in that, The knowledge tracing model is trained jointly by the first auxiliary task component and the basic knowledge tracing task component.
8. The deep knowledge tracing method based on multi-tasks according to claim 7, characterized in that, The objective loss function for training the model is: ;in, is a hyperparameter used to adjust the weights of the first loss function.
9. The deep knowledge tracing method based on multi-tasks according to claim 6 or 8, characterized in that, Calculate the extreme value of the target loss function to obtain the optimal parameter values for the model training process.
10. The deep knowledge tracing method based on multi-tasks according to claim 1, characterized in that, The specified knowledge tracing network includes recurrent neural networks, memory networks, graph networks, and / or attention networks.
11. A deep knowledge tracing device based on multi-task, characterized in that, include: The first module can obtain the current time. and the problem representation of all previous moments. Knowledge point representation and the correct / incorrect answer ; The second module allows input of the problem representation. The knowledge point representation and the results of the correct or incorrect answers. The knowledge tracking model, after training, will then process the information. as well as, The third module is capable of responding to the processing of the knowledge tracking model and obtaining the next moment. The degree of mastery of the knowledge points; The knowledge tracing model is based on a specified knowledge tracing network, including features for predicting the current time. The first auxiliary task component corresponding to the knowledge point in the question and the component used to predict the next moment. A basic knowledge tracking task component for answering knowledge points correctly or incorrectly; the first auxiliary task component and the basic knowledge tracking task component jointly participate in the training of the knowledge tracking model; The knowledge tracing model further includes a second auxiliary task component for predicting the student's overall historical answer accuracy. The second auxiliary task component, together with the first auxiliary task component and the basic knowledge tracing task component, participates in the training of the knowledge tracing model. Specifically, the second auxiliary task component and the basic knowledge tracking task component obtain the knowledge state required for prediction. ,include: Enter the correct or incorrect answer. and the representation of the knowledge points To the answer encoding unit, output the relational answer encoding representation. ; Calculate the answer encoding representation Relationship representation and the representation of the knowledge points joint representation and input the joint representation The knowledge tracking network outputs the knowledge status. ; wherein, the knowledge state The knowledge state in the specified knowledge tracing network at the previous time step and the joint characterization Sure.
12. An electronic device, characterized in that, include: processor; as well as A memory configured to store computer-executable instructions, which, when executed, cause the processor to perform the multitasking-based deep knowledge tracing method according to any one of claims 1 to 10.
13. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores one or more programs that, when executed by an electronic device including multiple applications, cause the electronic device to perform the multi-task-based deep knowledge tracing method according to any one of claims 1 to 10.
Citation Information
Patent Citations
Associated knowledge tracking method, device and equipment
CN114490980A