Knowledge tracking method based on one-dimensional residual convolutional auto-encoding and related device

Abstract

A knowledge tracking method based on one-dimensional residual convolutional auto-encoding and related device, comprising: obtaining an original training data set, and preprocessing the obtained original training data set; extracting a plurality of features affecting the answering result from the original training data set, and then merging the preprocessed data to obtain a new input sequence; building a knowledge tracking network model; inputting the original training data set and the sorted features as the input of the model into the entire network model for training; predicting the objects on each sample one by one on the validation set and the test set after training the knowledge tracking model, until all samples are predicted, and the knowledge tracking test is completed. The present application realizes the tracking of the knowledge state of students, can predict the current knowledge state of students according to the historical answering records and a plurality of behavior characteristics of students, and improves the accuracy of knowledge tracking.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of natural language processing, and particularly relates to a knowledge tracking method based on one-dimensional residual convolution auto-encoding and a related device. BACKGROUND

[0002] Knowledge tracking is a technology for modeling a student's knowledge mastery according to the student's past answering conditions, thereby obtaining a current knowledge state representation of the student, so as to accurately predict the student's mastery degree of each knowledge concept and the student's performance in future learning behaviors.

[0003] There are mainly two kinds of knowledge tracking algorithms: a traditional knowledge tracking method and a knowledge tracking method based on deep learning. The traditional knowledge tracking method assumes that once a student masters a knowledge point, the student will not forget the knowledge point, and the student will perform well when doing exercises belonging to the knowledge point in the future. However, in fact, the student may forget the previously mastered knowledge point after a long time. The knowledge tracking method based on deep learning mostly captures the student's recent performance to predict the student's answering results, and can more utilize the student's recent performance. Although the knowledge tracking method based on deep learning has made great progress in tracking the change of the student's knowledge state, it still lacks behavior characteristics such as the number of attempts and the number of prompt requests of the student in the answering process, which affect the student's answering results, thereby leading to inaccurate prediction results. SUMMARY

[0004] The application aims to provide a knowledge tracking method based on one-dimensional residual convolution auto-encoding and a related device, so as to solve the problem of inaccurate prediction results caused by the lack of behavior characteristics affecting the student's answering results.

[0005] To achieve the above-mentioned purpose, the application adopts the following technical solutions:

[0006] In a first aspect, the application provides a knowledge tracking method based on one-dimensional residual convolution auto-encoding, comprising:

[0007] Obtaining an original training data set and preprocessing the obtained original training data set;

[0008] Extracting a plurality of characteristics affecting the answering results from the original training data set, and then merging the preprocessed data to obtain a new input sequence;

[0009] Building a knowledge tracking network model;

[0010] Taking the original training data set and the sorted characteristics as inputs of the model, and inputting them into the entire network model for training;

[0011] The trained knowledge tracking model is used to predict the objects on each sample one by one on the validation set and the test set until all samples are predicted, and the test of knowledge tracking is completed.

[0012] Further, the original training data set is obtained as the ASSIST2009 and ASSISTchall public data sets; the preprocessing operation includes: converting non-binary correctness into binary correctness, deleting rows with missing key values in the data set, and obtaining the attempt sequence of each student according to the student grouping.

[0013] Further, after extracting a plurality of features affecting the answer result from the original training data set, the features affecting the answer result are sorted by random forest, the top three features after sorting are fused with the input sequence, and the input sequence fusion adopts the concat operation.

[0014] Further, the knowledge tracking network model is built:

[0015] The knowledge tracking model includes a knowledge state extraction module, an attention module and a prediction module, the knowledge state extraction module is used to extract the mastery degree of each knowledge point, the attention module is used to identify and strengthen the different mastery degrees of the knowledge points, and the prediction module is used to obtain the probability of the correct answer to the next question.

[0016] Further, the knowledge state extraction module is built:

[0017] The knowledge state extraction module is composed of GRU and one-dimensional residual convolutional auto-encoding, GRU is used to extract global knowledge state, and its structure includes update gate and reset gate; the update gate is used to control the degree of state information at the previous moment being brought into the current state, and the greater the value of the update gate represents the more state information at the previous moment is brought in; the one-dimensional residual convolutional auto-encoding is used to extract local knowledge state, wherein the one-dimensional residual convolutional auto-encoding includes 1D convolution layer, maximum pooling layer, up-sampling layer, 1D deconvolution layer and residual connection in turn; the input sequence is first input into the convolution layer with a convolution kernel size of 1x3 and a step of 2 and the maximum pooling layer with a convolution kernel size of 1x2 and a step of 2, then input into the up-sampling layer with a convolution kernel size of 1x2 and a step of 2 and the deconvolution layer with a convolution kernel size of 1x3 and a step of 2, and finally the input and output of the one-dimensional residual convolutional auto-encoding are integrated together through the residual connection.

[0018] Further, the attention module and the prediction module:

[0019] The attention module is to calculate the similarity between the knowledge states first, then get the corresponding probability distribution between each knowledge state through Softmax normalization, and finally get the weighted sum of the knowledge state and the attention weight;

[0020] The prediction module obtains a final prediction result from a linear layer and a Sigmoid activation function, the linear layer is composed of two fully connected layers, and the activation function of the linear layer is a ReLU function;

[0021] P = Sigmoid(W2(RELU(W1h t +b1))+b2)

[0022] Wherein W1 and W2 are weight values of the linear layer, b1 and b2 are bias values of the linear layer, h t ' is the knowledge state after attention recognition and reinforcement.

[0023] Further, the network model is trained.

[0024] First, the knowledge tracking dataset is divided into a training set, a validation set and a test set in a ratio of 8:1:1, the training set and the validation set are input into the network for training, after obtaining the trained parameters, the trained knowledge tracking model is tested, and the trained knowledge tracking model is used to predict the objects on each sample one by one on the test set until all samples are predicted, and the knowledge tracking test is completed.

[0025] During training, the number of model iterations is 100, and the learning rate is 0.001; the loss function L is propagated in reverse direction to constantly optimize the weight parameters of each module in the model, and the training of the model is completed.

[0026] Secondly, the present application provides a knowledge tracking system based on one-dimensional residual convolution auto-encoding, comprising:

[0027] A data preprocessing module is configured to acquire an original training dataset and preprocess the acquired original training dataset.

[0028] A feature extraction and sorting module is configured to extract a plurality of features affecting the answer result from the original training dataset, and then combine the preprocessed data to obtain a new input sequence.

[0029] A model building module is configured to build a knowledge tracking network model.

[0030] A training module is configured to input the original training dataset and the sorted features as inputs of the model into the entire network model for training.

[0031] A test module is configured to use the trained knowledge tracking model to predict the objects on each sample one by one on the validation set and the test set until all samples are predicted, and the knowledge tracking test is completed.

[0032] In a third aspect, the present application provides a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the knowledge tracking method based on one-dimensional residual convolutional auto-encoding when executing the computer program.

[0033] In a fourth aspect, the present application provides a computer-readable storage medium storing a computer program, wherein the computer program implements the steps of the knowledge tracking method based on one-dimensional residual convolutional auto-encoding when executed by a processor.

[0034] Compared with the prior art, the present application has the following technical effects:

[0035] Firstly, the present application acquires the ASSIST2009 and ASSISTchall datasets and pre-processes the acquired datasets, sorts the features affecting the student answer results by using the random forest algorithm, and then merges the top three features after sorting with the original input sequence as a new input. Then, one-dimensional residual convolutional auto-encoding is used to extract the local knowledge state of the student, important problem information is retrieved in the encoding path of the residual convolutional auto-encoding, and then the local knowledge state of the student is extracted in the decoding path. At the same time, GRU is used to replace RNN to extract the global knowledge state of the student, the change of the knowledge state of the student is tracked through the hidden state of the GRU, thereby solving the long-term dependence problem of the model. Then, the local knowledge state extracted by the one-dimensional residual convolutional auto-encoding is merged with the global knowledge state extracted by the GRU, thereby obtaining a more comprehensive knowledge state. Finally, the attention mechanism is used to identify and strengthen the different mastery degrees of the student on the knowledge points, and the probability of the student correctly answering the next question can be obtained according to the output of the knowledge state and the attention weight of the student. The present application realizes the tracking task of the knowledge state of the student, can predict the future performance of the student according to the answer record and behavior characteristics of the student, and improves the accuracy of knowledge tracking.

[0036] Compared with other methods, the method of the present application can consider the features that significantly affect the answer results of the student, integrate them into the original input sequence, fully extract their features, and make predictions.

[0037] The method of the present application has obvious improvement on the ASSIST2009 and ASSISTchall datasets, and can be popularized and applied in the problem recommendation system. BRIEF DESCRIPTION OF DRAWINGS

[0038] Figure 1 is the overall flowchart of the present application.

[0039] Figure 2 is the overall structure schematic diagram of the present application.

[0040] Figure 3 (a) is the ranking result of the feature ranking ASSIST2009 dataset of the present application, Figure 3 (b) is the ranking result of the ASSISTchall dataset.

[0041] Figure 4 is the structure of one-dimensional convolutional auto-encoding.

[0042] Figure 5 is a structural diagram of the attention module of the present application.

[0043] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and serve to explain the principles of the present disclosure. The above-mentioned drawings have shown specific embodiments of the present disclosure, which will be described in more detail hereinafter. These drawings and the written description are not intended to restrict the scope of the concept of the present disclosure in any way, but to illustrate the concept of the present disclosure to those skilled in the art by reference to specific embodiments. DETAILED DESCRIPTION

[0044] The exemplary embodiments will be described in detail herein with reference to the accompanying drawings. The embodiments described in the following exemplary embodiments do not represent all the embodiments consistent with the present disclosure. Rather, they are merely examples of devices and methods consistent with some aspects of the present disclosure as detailed in the appended claims.

[0045] The present application obtains a training data set, pre-processes the obtained data set, and incorporates a plurality of features affecting student answer results into an input sequence as a new input. First, one-dimensional residual convolutional auto-encoding is used to extract the local knowledge state of the student, important problem information is retrieved in the encoding path of the residual convolutional auto-encoding, and then the local knowledge state of the student is extracted in the decoding path. At the same time, GRU is used to replace RNN to extract the global knowledge state of the student, and the change of the knowledge state of the student is tracked through the hidden state of the GRU. Then, the local knowledge state extracted by the one-dimensional residual convolutional auto-encoding is merged with the global knowledge state extracted by the GRU, so as to obtain a more comprehensive knowledge state. Finally, the attention mechanism is used to identify and strengthen the different mastery degrees of the student on the knowledge points, and the probability of the student correctly answering the next question can be obtained according to the weighted sum of the knowledge state of the student and the attention weight. The present application realizes the tracking task of the knowledge state of the student, can predict the future performance of the student according to the answer record and behavior characteristics of the student, and improves the accuracy of knowledge tracking.

[0046] An embodiment of the knowledge tracking method of the present application has the overall flow as shown in Figure 1 The overall framework of the model is as shown in Figure 2 The steps are as follows:

[0047] (1) Data Preprocessing

[0048] Download ASSIST2009 and ASSISTchall public datasets. ASSIST2009 dataset is collected on the educational platform ASSISTment during 2009. ASSIST2009 dataset contains two different interaction mode datasets: skill-builder and no-skill-builder. This work uses skill-builder dataset in ASSIST2009, in which each interaction is labeled with a unique skill id, which makes the preprocessing easier. Skill-builder dataset contains 4151 students, 110 skills, and 325637 interaction records. ASSISTchall is the dataset of ASSISTments Data Mining Challenge 2017, which contains 942816 interaction records from 1709 students, with 102 skills. The preprocessing mainly divides the dataset into three lines, the first line represents the length of the interaction sequence, the second line represents the id of the question, and the third line represents the correct answer or the wrong answer of each question. The interaction sequence of a student is X t = {x1, x2,..., x t}.

[0049]

[0050] where e t represents the answered exercise, a t represents whether the answer is correct or not, a t is usually {0, 1}.

[0051] (2) Feature Ranking

[0052] Rank the features that affect the student's answer results by random forest, the ranking results of ASSIST2009 and ASSISTchall datasets are shown in Figure 3 (a) and Figure 3 (b) respectively. The top three features after ranking are fused with the input sequence, that is:

[0053]

[0054]

[0055] where is the matrix concatenation operation, and O is the one-hot encoding function. In the ASSIST2009 dataset, f1 is attempt_count, f2 is first_action, and f3 is hint_total. In the ASSISTchall dataset, f1 is hint, f2 is hintTotal, and f3 is hintCount.

[0056] (3) Building a knowledge tracking network model:

[0057] The structure of the knowledge tracking network model is shown in Figure 2 The knowledge tracking model mainly includes a knowledge state extraction module, an attention module, and a prediction module. The knowledge state extraction module is mainly used to extract the mastery level of each knowledge point by the student. The attention module is used to identify and strengthen the different mastery levels of the knowledge points by the student. The prediction module is used to obtain the probability of the student correctly answering the next question.

[0058] (4) Building a knowledge state extraction module:

[0059] The knowledge state extraction module is composed of a GRU and a one-dimensional residual convolutional autoencoder. The GRU is used to extract the global knowledge state of the student. Its structure includes an update gate and a reset gate. The update gate is used to control the degree of state information from the previous time being brought into the current state. The greater the value of the update gate, the more state information from the previous time is brought in. The one-dimensional residual convolutional autoencoder is used to extract the local knowledge state of the student. The structure of the one-dimensional convolutional autoencoder is shown in Figure 4 It sequentially includes a 1D convolutional layer, a max pooling layer, an upsampling layer, a 1D deconvolutional layer, and a residual connection. The input sequence is first input into a convolutional layer with a kernel size of 1x3 and a step size of 2, and a max pooling layer with a kernel size of 1x2 and a step size of 2. Then it is input into an upsampling layer with a kernel size of 1x2 and a step size of 2, and a deconvolutional layer with a kernel size of 1x3 and a step size of 2. Finally, the input and output of the one-dimensional residual convolutional autoencoder are integrated together through a residual connection to ensure the integrity of the data.

[0060] (5) Attention module and prediction module:

[0061] The attention module first calculates the similarity between knowledge states, then obtains the corresponding probability distribution between each knowledge state through Softmax normalization, and finally obtains the weighted sum of the knowledge state and the attention weight.

[0062] The prediction module obtains the final prediction result through a linear layer and a Sigmoid activation function. The linear layer is composed of two fully connected layers, and the activation function of the linear layer is the ReLU function.

[0063] P = Sigmoid(W2(RELU(W1ht +b1) +b2) (4)

[0064] where W1 and W2 are the weights of the linear layer, b1 and b2 are the biases of the linear layer, h t is the knowledge state after attention recognition and reinforcement.

[0065] (6) Training network model

[0066] First, the knowledge tracking dataset is divided into training set, validation set and test set in the ratio of 8:1:1, and the training set and validation set are input into the network for training. After obtaining the trained parameters, the trained knowledge tracking model is tested, and the trained knowledge tracking model is used to predict the objects on each sample one by one until all samples are predicted, and the test of knowledge tracking is completed.

[0067] The number of iterations of the model during training is 100, and the learning rate is 0.001; by performing back propagation on the loss function L, the weight parameters of each module in the model are constantly optimized, and the training of the model is completed;

[0068] The specific process of model setting in the embodiment of the application is as follows:

[0069] 6.1) Optimizer. The Adam optimizer is selected to alleviate the problem of gradient disappearance or explosion in the back propagation process of the neural network, which helps to better and faster convergence of the neural network.

[0070] 6.2) Activation function. The Sigmoid activation function is selected to increase the nonlinearity of the neural network model, which can output the combination of linear functions as a probability between (0, 1).

[0071] 6.3) Loss function. The loss function used in the application is binary cross-entropy loss function.

[0072]

[0073] where p i is the predicted value, r i is the true value, and N is the total number of exercises.

[0074] (7) Test using the trained model

[0075] The trained knowledge tracing model is used to predict the objects on each sample one by one on the test set until all samples are predicted, and the test of knowledge tracing is completed.

[0076] In order to verify the beneficial effects of the present application, the inventors used the method of the embodiment of the present application and BKT, DKT, DKT+ and DKVMN (hereinafter referred to as comparative experiments) to perform comparative experiments on the ASSIST2009 dataset and the ASSISTchall dataset.

[0077] Table 1 Experimental results of the method of embodiment 1 and comparative experiments

[0078]

[0079] The experimental results show that the present application has excellent effects on the ASSIST2009 and ASSISTchall datasets, which indicates that the present application can well predict the mastery of knowledge points by students and improve the prediction performance of knowledge tracing.

[0080] In another embodiment of the present application, a knowledge tracing system based on one-dimensional residual convolution auto-encoding is provided, which can be used to implement the above-mentioned knowledge tracing method based on one-dimensional residual convolution auto-encoding. Specifically, the system comprises:

[0081] The data preprocessing module is used to obtain the original training data set and preprocess the obtained original training data set;

[0082] The feature extraction and sorting module is used to extract a plurality of features affecting the answer results from the original training data set, and then merge the preprocessed data to obtain a new input sequence;

[0083] The model building module is used to build a knowledge tracing network model;

[0084] The training module is used to input the original training data set and the sorted features as the input of the model into the entire network model for training;

[0085] The test module is used to predict the objects on each sample one by one on the validation set and the test set by using the trained knowledge tracing model until all samples are predicted, and the test of knowledge tracing is completed.

[0086] The division of the modules in the embodiments of the present application is illustrative, and is merely a logical function division, and in actual implementation, another division manner can be used, and in addition, each function module in each embodiment of the present application can be integrated in one processor, or can be a separate physical existence, or two or more modules can be integrated in one module. The integrated module can be realized in the form of hardware or in the form of a software function module.

[0087] In still another embodiment of the present application, a computer device is provided, which comprises a processor and a memory, the memory is used to store a computer program, the computer program comprises program instructions, and the processor is used to execute the program instructions stored in the computer storage medium. The processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc., which are the computing core and control core of the terminal, and are suitable for implementing one or more instructions, and are specifically suitable for loading and executing one or more instructions in the computer storage medium to implement a corresponding method flow or a corresponding function; the processor in the embodiments of the present application can be used for the operation of a knowledge tracking method.

[0088] In still another embodiment of the present application, the present application further provides a storage medium, specifically a computer readable storage medium (Memory), which is a memory device in a computer device, and is used to store programs and data. It can be understood that the computer readable storage medium here can include an internal storage medium in the computer device, and of course can also include an extended storage medium supported by the computer device. The computer readable storage medium provides a storage space, and the storage space stores an operating system of the terminal. In addition, one or more instructions suitable for being loaded and executed by the processor are also stored in the storage space, and the instructions can be one or more computer programs (including program codes). It should be noted that the computer readable storage medium here can be a high-speed RAM memory, or a non-volatile memory such as at least one disk memory. One or more instructions stored in the computer readable storage medium can be loaded and executed by the processor to implement the corresponding steps of the knowledge tracking method in the above embodiments.

[0089] Those skilled in the art will appreciate that embodiments of the present application can be readily used as a method, apparatus such as a system, or computer program product. Accordingly, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application can take the form of a computer program product on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage devices, etc.) embodying computer readable program code.

[0090] The present application is described in reference to the drawings using a flowchart and / or a block diagram of methods, apparatus (systems) and computer program products according to embodiments of the application. It will be understood that each block of the flowchart and / or block diagram, and combinations of blocks in the flowchart and / or block diagram, 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 processing system 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 the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in one or more of the flowchart or block diagram block or blocks. Figure 1 means for carrying out each of the functionality specified in the flowchart or block diagram block or blocks.

[0091] These computer program instructions can also be stored in a computer- readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the flowchart or block diagram block or blocks. Figure 1 one or more functions specified in one or more of the flowchart or block diagram block or blocks. Figure 1 means for carrying out each of the functionality specified in the flowchart or block diagram block or blocks.

[0092] These computer program instructions can 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 flowchart or block diagram block or blocks. Figure 1 one or more functions specified in one or more of the flowchart or block diagram block or blocks. Figure 1 means for carrying out each of the functionality specified in the flowchart or block diagram block or blocks.

[0093] Finally, it should be noted that the above-mentioned embodiments are merely intended for describing and illustrating, not limiting, the technical solution of the present application. Although the present application has been described in detail with reference to the above-mentioned embodiments, those skilled in the art should understand that the specific embodiments of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the present application, and any modification or equivalent replacement without departing from the spirit and scope of the present application should be covered in the protection scope of the claims of the present application.

Claims

1. A knowledge tracing method based on one-dimensional residual convolutional autoencoders, characterized in that, include: Obtain the original training dataset and preprocess it. Multiple features that influence the answering results are extracted from the original training dataset and then merged with the preprocessed data to obtain a new input sequence. Build a knowledge tracing network model; The original training dataset and the sorted features are used as inputs to the model and fed into the entire network model for training. The trained knowledge tracking model is used to predict the objects on each sample on the validation set and the test set one by one until all samples have been predicted, thus completing the knowledge tracking test. To obtain the original training dataset, you need to obtain the ASSIST2009 and ASSISTchall public datasets. Preprocessing operations include: converting non-binary correctness to binary correctness, removing rows in the dataset that are missing key values, and obtaining the attempt sequence for each student by grouping them by student; Building a knowledge tracing network model: The knowledge tracking model includes a knowledge state extraction module, an attention module, and a prediction module. The knowledge state extraction module is used to extract the mastery level of each knowledge point, the attention module is used to identify and reinforce different mastery levels of knowledge points, and the prediction module is used to obtain the probability of correctly answering the next question. Build a knowledge state extraction module: The knowledge state extraction module consists of a GRU and a one-dimensional residual convolutional autoencoder. The GRU is used to extract the global knowledge state, and its structure includes an update gate and a reset gate. The update gate controls the degree to which the state information from the previous time step is incorporated into the current state; a larger update gate value indicates that more state information from the previous time step is incorporated. The one-dimensional residual convolutional autoencoder is used to extract the local knowledge state. The one-dimensional residual convolutional autoencoder consists of a 1D convolutional layer, a max pooling layer, an upsampling layer, a 1D deconvolutional layer, and a residual connection. The input sequence is first fed into a convolutional layer with a kernel size of 1x3 and a stride of 2 and a max pooling layer with a stride of 1x2, then into an upsampling layer with a kernel size of 1x2 and a stride of 2 and a deconvolutional layer with a stride of 1x3. Finally, the input and output of the one-dimensional residual convolutional autoencoder are integrated together through a residual connection. Attention module and prediction module: The attention module first calculates the similarity between knowledge states, then obtains the probability distribution between each knowledge state through Softmax normalization, and finally obtains the weighted sum of knowledge states and attention weights. The prediction module obtains the final prediction result by using a linear layer and a sigmoid activation function. The linear layer consists of two fully connected layers, and the activation function of the linear layer is the ReLU function. in and These are the weights of the linear layer. and These are the biases for the linear layers. This refers to the knowledge state after attention recognition and reinforcement.

2. The knowledge tracing method based on one-dimensional residual convolutional autoencoder according to claim 1, characterized in that, After extracting multiple features that influence the answering results from the original training dataset, the features that influence the answering results are sorted by random forest. The top three sorted features are then fused with the input sequence using the concat operation.

3. The knowledge tracing method based on one-dimensional residual convolutional autoencoder according to claim 1, characterized in that, Training the network model: First, the knowledge tracking dataset is divided into a training set, a validation set, and a test set in a ratio of 8:1:

1. The training set and the validation set are then input into the network for training. After obtaining the trained parameters, the trained knowledge tracking model is tested by predicting the objects on each sample on the test set one by one until all samples have been predicted, thus completing the knowledge tracking test. The model was trained with 100 iterations and a learning rate of 0.

001. The weight parameters of each module in the model were continuously optimized by backpropagating the loss function L, thus completing the training of the model.

4. A knowledge tracking system based on one-dimensional residual convolutional autoencoders, characterized in that, include: The data preprocessing module is used to acquire the original training dataset and preprocess the acquired original training dataset. The feature extraction and sorting module is used to extract multiple features that affect the answer results from the original training dataset, and then merge them with the preprocessed data to obtain a new input sequence. The model building module is used to build knowledge tracing network models; The training module is used to take the original training dataset and the sorted features as inputs to the model and feed them into the entire network model for training. The testing module is used to predict the objects on each sample of the trained knowledge tracking model on the validation set and the test set one by one until all samples have been predicted, thus completing the knowledge tracking test. To obtain the original training dataset, you need to obtain the ASSIST2009 and ASSISTchall public datasets. Preprocessing operations include: converting non-binary correctness to binary correctness, removing rows in the dataset that are missing key values, and obtaining the attempt sequence for each student by grouping them by student; Building a knowledge tracing network model: The knowledge tracking model includes a knowledge state extraction module, an attention module, and a prediction module. The knowledge state extraction module is used to extract the mastery level of each knowledge point, the attention module is used to identify and reinforce different mastery levels of knowledge points, and the prediction module is used to obtain the probability of correctly answering the next question. Build a knowledge state extraction module: The knowledge state extraction module consists of a GRU and a one-dimensional residual convolutional autoencoder. The GRU is used to extract the global knowledge state, and its structure includes an update gate and a reset gate. The update gate controls the degree to which the state information from the previous time step is incorporated into the current state; a larger update gate value indicates that more state information from the previous time step is incorporated. The one-dimensional residual convolutional autoencoder is used to extract the local knowledge state. The one-dimensional residual convolutional autoencoder consists of a 1D convolutional layer, a max pooling layer, an upsampling layer, a 1D deconvolutional layer, and a residual connection. The input sequence is first fed into a convolutional layer with a kernel size of 1x3 and a stride of 2 and a max pooling layer with a stride of 1x2, then into an upsampling layer with a kernel size of 1x2 and a stride of 2 and a deconvolutional layer with a stride of 1x3. Finally, the input and output of the one-dimensional residual convolutional autoencoder are integrated together through a residual connection. Attention module and prediction module: The attention module first calculates the similarity between knowledge states, then obtains the probability distribution between each knowledge state through Softmax normalization, and finally obtains the weighted sum of knowledge states and attention weights. The prediction module obtains the final prediction result by using a linear layer and a sigmoid activation function. The linear layer consists of two fully connected layers, and the activation function of the linear layer is the ReLU function. in and These are the weights of the linear layer. and These are the biases for the linear layers. This refers to the knowledge state after attention recognition and reinforcement.

5. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the knowledge tracking method based on one-dimensional residual convolutional autoencoder as described in any one of claims 1 to 3.

6. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the knowledge tracking method based on one-dimensional residual convolutional autoencoder as described in any one of claims 1 to 3.

Images