MOOC student learning prediction method and device

The MOOC students' dropout prediction model constructed through convolution module, attention mechanism memory network and supervised learning integrated network solves the problem of poor prediction results in the existing technology, and realizes accurate prediction of MOOC students' dropout behavior.

CN118964948APending Publication Date: 2024-11-15GUANGDONG UNIV OF TECH

Patent Information

Application Number
CN202411183243.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-27
Publication Date
2024-11-15

AI Technical Summary

Technical Problem

The existing MOOC student dropout prediction technology uses simple questionnaires and simple analysis methods, resulting in poor accuracy of prediction results and the inability to predict and analyze dropout problems in depth.

Method used

The convolution module, attention mechanism memory network and supervised learning integrated network are used to construct a MOOC student dropout prediction model. Local features are extracted through the convolution module, attention mechanism memory network is used to select feature, and the supervised learning integrated network is used to predict dropouts. Dynamically adjust the importance of features to improve prediction accuracy.

Benefits of technology

It improves the accuracy of MOOC students' dropout prediction, and can dynamically adjust the importance of different characteristics, thereby enhancing the accuracy of model prediction results and achieving accurate prediction of MOOC students' dropout behavior.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118964948A_ABST
    Figure CN118964948A_ABST
Patent Text Reader

Abstract

The invention discloses a MOOC student learning prediction method and device, and the method comprises the steps: obtaining to-be-predicted student behavior sequence data, and inputting the to-be-predicted student behavior sequence data into a preset MOOC student learning prediction model; the preset MOOC student learning prediction model comprises a convolution module, an attention mechanism memory network and a supervised learning integrated network; local feature extraction is carried out on the to-be-predicted student behavior sequence data through a convolution module, and local target features are output; performing feature selection on the local target features by adopting an attention mechanism memory network to generate behavior feature variables; inputting the behavior characteristic variables into a supervised learning integrated network for learning prediction, and outputting a learning prediction result of the MOOC students; the technical problem of poor prediction result precision caused by an existing MOOC student learning prediction technology is solved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of online education dropout prediction, and in particular to a method and device for predicting dropout of MOOC students. Background Art

[0002] With the popularity of Massive Open Online Courses (MOOCs), they provide students with unprecedented learning opportunities. At the same time, MOOCs, as an effective way to protect students' right to education and promote lifelong learning, have become a beneficial supplement to traditional education.

[0003] Although MOOCs are gradually becoming an integral part of the higher education learning process, there are still many issues to be addressed surrounding one of the major unsolved issues of MOOCs: student dropouts. Compared to traditional face-to-face teaching, MOOCs have a higher dropout rate due to the lack of constraints and face-to-face communication between teachers and students, so developing an effective prediction scheme can predict when students will drop out.

[0004] Most of the existing MOOC student dropout prediction technologies use simple questionnaires and simple analysis methods to analyze some superficial MOOC dropout phenomena, which cannot predict and analyze the dropout problem in depth, resulting in poor prediction results. Summary of the invention

[0005] The present invention provides a method and device for predicting the dropout of MOOC students, which are used to solve the technical problem that the existing MOOC student dropout prediction technology leads to poor prediction result accuracy.

[0006] A first aspect of the present invention provides a method for predicting dropout of MOOC students, comprising:

[0007] Obtaining student behavior sequence data to be predicted, and inputting the student behavior sequence data to be predicted into a preset MOOC student dropout prediction model; the preset MOOC student dropout prediction model includes a convolution module, an attention mechanism memory network, and a supervised learning integrated network;

[0008] Extracting local features of the student behavior sequence data to be predicted through the convolution module and outputting local target features;

[0009] Using the attention mechanism memory network to perform feature selection on the local target features to generate behavior feature variables;

[0010] The behavioral characteristic variables are input into the supervised learning integrated network for dropout prediction, and the prediction results of MOOC students' dropout are output.

[0011] Optionally, the convolution module includes a convolution layer and a pooling layer; the step of extracting local features of the student behavior sequence data to be predicted by the convolution module and outputting local target features includes:

[0012] Using multiple elements in the student behavior sequence data to be predicted, constructing multiple behavior sub-matrices;

[0013] Performing a convolution operation on each of the behavioral sub-matrices through the convolution layer, and outputting a feature sub-matrix corresponding to each of the behavioral sub-matrices;

[0014] Using each of the feature sub-matrices to construct a target feature matrix;

[0015] The target feature matrix is ​​input into the pooling layer for maximum pooling, and local target features are output.

[0016] Optionally, the attention mechanism memory network includes a bidirectional long short-term memory module and a fully connected layer; the step of using the attention mechanism memory network to perform feature selection on the local target features to generate behavior feature variables includes:

[0017] Using the bidirectional long short-term memory module to extract state features of the local target features, and outputting a target cell hidden state vector;

[0018] Performing a linear transformation on the target cell hidden state vector through a fully connected layer, and outputting a query matrix, a key-value matrix, and a value matrix;

[0019] The attention mechanism is used to determine the behavior feature variables according to the query matrix, the key-value matrix and the value matrix.

[0020] Optionally, the bidirectional long short-term memory module includes a forget gate, an input gate, an output gate and an activation function layer; the step of using the bidirectional long short-term memory module to extract state features of the local target features and outputting the target cell hidden state vector includes:

[0021] Performing forgetting state feature extraction on the local target feature through the forgetting gate, and outputting a forgetting state vector;

[0022] Using the input gate to extract input state features of the local target features to generate an input state vector;

[0023] Inputting the local target feature into the output gate to extract the output state feature and generate an output state vector;

[0024] Using the activation function layer to perform nonlinear mapping on the local target features and output a nonlinear state vector;

[0025] Calculating an initial cell hidden state vector according to the nonlinear state vector, the forgetting state vector and the input state vector;

[0026] Determining an intermediate cell hidden state vector based on the initial cell hidden state vector and the output state vector;

[0027] Performing inverse processing on the intermediate cell hidden state vector, and outputting an inverse cell hidden state vector;

[0028] The reverse cell hidden state vector and the intermediate cell hidden state vector are concatenated to output a target cell hidden state vector.

[0029] Optionally, the step of using the attention mechanism to determine the behavior feature variables according to the query matrix, the key-value matrix and the value matrix includes:

[0030] Performing a similarity operation on the query matrix and the key value matrix, and outputting an attention distribution weight;

[0031] The attention distribution weight and the value matrix are weightedly summed to determine the behavior characteristic variable.

[0032] Optionally, it also includes:

[0033] Obtaining behavior sequence data of students to be trained, inputting the behavior sequence data of students to be trained into an initial MOOC student dropout prediction model, and outputting a dropout prediction result of students to be trained;

[0034] Using a back propagation algorithm, the model gradient is calculated according to the prediction results of the MOOC students to be trained to drop out;

[0035] The model gradient is used to update the model parameters of the initial MOOC student dropout prediction model, determine the intermediate MOOC student dropout prediction model, and count the number of model updates in real time;

[0036] Determining whether the model update times reaches a preset update threshold;

[0037] If so, the intermediate MOOC student dropout prediction model is used as the trained preset MOOC student dropout prediction model.

[0038] A second aspect of the present invention provides a device for predicting dropout of MOOC students, comprising:

[0039] An acquisition module is used to acquire the student behavior sequence data to be predicted, and input the student behavior sequence data to be predicted into a preset MOOC student dropout prediction model; the preset MOOC student dropout prediction model includes a convolution module, an attention mechanism memory network and a supervised learning integrated network;

[0040] An output module, used for extracting local features of the student behavior sequence data to be predicted through the convolution module, and outputting local target features;

[0041] A generation module, used for performing feature selection on the local target features by using the attention mechanism memory network to generate behavior feature variables;

[0042] The prediction module is used to input the behavioral characteristic variables into the supervised learning integrated network to perform dropout prediction and output the MOOC student dropout prediction results.

[0043] A third aspect of the present invention provides a computer device, comprising a memory and a processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the processor executes the steps of the method for predicting dropout of MOOC students as described in any one of the above items.

[0044] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed, implements the steps of the method for predicting dropout of MOOC students as described in any one of the above items.

[0045] A fifth aspect of the present invention provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions, wherein when the program instructions are executed by a computer, the computer executes the steps of the method for predicting dropout of MOOC students as described in any one of the above items.

[0046] It can be seen from the above technical solutions that the present invention has the following advantages:

[0047] The above technical scheme of the present invention provides a method for predicting the dropout of MOOC students. First, the behavior sequence data of the students to be predicted is obtained, and the behavior sequence data of the students to be predicted is input into a preset MOOC student dropout prediction model; the preset MOOC student dropout prediction model includes a convolution module, an attention mechanism memory network and a supervised learning integrated network; then, the local feature extraction of the student behavior sequence data to be predicted is performed through the convolution module, and the local target feature is output; the local target feature is selected by the attention mechanism memory network to generate behavior feature variables; finally, the behavior feature variables are input into the supervised learning integrated network for dropout prediction, and the MOOC student dropout prediction result is output; based on the above scheme, the local feature extraction of the student behavior sequence data to be predicted is performed through the convolution module, and the generated local target feature is selected in combination with the attention mechanism memory network, and finally the dropout prediction is performed based on the supervised learning integrated network, and the process of outputting the MOOC student dropout prediction result can dynamically adjust the importance of different features, thereby enhancing the accuracy of the model prediction result. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0049] Figure 1 A flowchart of the steps of a method for predicting dropout of MOOC students provided by an embodiment of the present invention;

[0050] Figure 2 A schematic diagram of the structure of a convolution module provided in an embodiment of the present invention;

[0051] Figure 3 A schematic diagram of the structure of the Leaf-wise growth strategy provided by an embodiment of the present invention;

[0052] Figure 4 A flowchart of another method for predicting dropout of MOOC students provided by an embodiment of the present invention;

[0053] Figure 5 A structural diagram of the BiLSTM model provided in an embodiment of the present invention;

[0054] Figure 6 A schematic diagram of the structure of the attention mechanism framework provided by an embodiment of the present invention;

[0055] Figure 7 A structural block diagram of a device for predicting dropout of MOOC students provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0056] The embodiment of the present invention provides a method and device for predicting the dropout of MOOC students, which are used to solve the technical problem that the existing MOOC student dropout prediction technology leads to poor prediction result accuracy.

[0057] In order to make the purpose, features and advantages of the present invention more obvious and easy to understand, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described below are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0058] See also Figure 1 , Figure 1 A flowchart of the steps of a method for predicting dropout of MOOC students provided by an embodiment of the present invention.

[0059] The present invention provides a method for predicting dropout of MOOC students, comprising:

[0060] Step 101: Obtain the student behavior sequence data to be predicted, and input the student behavior sequence data to be predicted into a preset MOOC student dropout prediction model; the preset MOOC student dropout prediction model includes a convolution module, an attention mechanism memory network, and a supervised learning integrated network.

[0061] It should be noted that the student behavior sequence data to be predicted is the learner behavior data used for model prediction, including video viewing time, number of discussions participated in, homework submission status, etc.; the preset MOOC student dropout prediction model proposed in the present invention is composed of a convolution module, an attention mechanism memory network and a supervised learning integrated network, and the attention mechanism is introduced. The network structure of the convolution module is a convolutional neural network (CNN, Cable News Network), the attention mechanism memory network is composed of a bidirectional long short-term memory module and a fully connected layer, and the attention mechanism is introduced. The network structure of the bidirectional long short-term memory module is a bidirectional long short-term memory network (BiLSTM, Bidirectional Long Short-Term Memory), and the supervised learning integrated network is a LightGBM model (Light Gradient Boosting Machine).

[0062] In this embodiment, the student behavior sequence data to be predicted is obtained, and the student behavior sequence data to be predicted is input into a preset MOOC student dropout prediction model; the preset MOOC student dropout prediction model includes a convolution module, an attention mechanism memory network and a supervised learning integrated network.

[0063] Step 102: extract local features from the student behavior sequence data to be predicted through a convolution module and output local target features.

[0064] The convolution module includes convolutional layers and pooling layers.

[0065] Please note that Figure 2 In the process of extracting features from learner behavior data (student behavior sequence data to be predicted) using the convolution module, namely the convolutional neural network (CNN), the model structure includes convolutional layers for feature extraction and pooling layers for reducing feature dimensions and sampling key information, which helps to capture the sequence characteristics and local features in learner behavior data, such as learning patterns, interest changes, and participation fluctuations.

[0066] Furthermore, in the process of feature extraction, that is, when processing multi-feature student behavior sequence data to be predicted in a convolutional neural network (CNN), the convolutional layer is used to automatically extract local features that are important for the prediction target (such as learning outcomes, interest preferences, participation, etc.). Here, it is assumed that there are s n×m-dimensional student behavior sequence data to be predicted as input, where s is the number of learners, n is the number of time steps (for example, the number of time units such as learning days, weeks, months, etc.), and m is the number of features at each time step (such as the length of time watched, the number of exercises completed, the number of homework submitted, etc.). For each learner's behavior data x s ∈R n×m , use multiple convolution kernels (or filters) to extract features. Specifically, after obtaining the student behavior sequence data to be predicted, multiple elements in the student behavior sequence data to be predicted are used to construct multiple behavior sub-matrices; convolution operations are performed on each behavior sub-matrix through a convolution layer, and a feature sub-matrix corresponding to each behavior sub-matrix is ​​output; the processing process of the feature sub-matrix can be expressed as:

[0067] ;

[0068] in, is the i-th feature sub-matrix; w is the convolution kernel; c is the bias term; is the activation function; It is the behavior submatrix composed of the i-th to i+g-1-th elements in the student behavior sequence data to be predicted; g is the convolution kernel size.

[0069] Furthermore, each feature sub-matrix is ​​used to construct a target feature matrix; the target feature matrix is ​​input into the pooling layer for maximum pooling, and the local target features are output; wherein the target feature matrix can be expressed as ,in, is the first eigenmatrix, is the second feature sub-matrix, is the n-g+1th feature submatrix.

[0070] It is worth mentioning that the main purpose of the pooling layer is to downsample (or subsample) the target feature matrix P output by the convolutional layer to reduce the spatial dimension of the data (i.e., height and width) while retaining important features. In this process, the pooling layer calculates a certain statistic of the local area as a representative of the area, and this statistic is usually the local optimal solution of the area. Finally, the local optimal solution (local target feature) obtained by the pooling layer through the MaxPooling technology (maximum pooling) is used as the input of the attention mechanism memory network; the maximum pooling process can be expressed as:

[0071] ;

[0072] in, is the local target feature at time t; P is the target feature matrix; is the first eigenmatrix, is the second feature sub-matrix, is the n-g+1th feature submatrix; is the maximum pooling.

[0073] In this embodiment, local features are extracted from the student behavior sequence data to be predicted through a convolution module, and local target features are output.

[0074] Step 103: Use the attention mechanism memory network to perform feature selection on local target features and generate behavioral feature variables.

[0075] The attention mechanism memory network consists of a bidirectional long short-term memory module and a fully connected layer.

[0076] Specifically, a bidirectional long short-term memory module is used to extract state features of local target features and output the hidden state vector of the target cell. The hidden state vector of the target cell is linearly transformed through a fully connected layer to output the query matrix, key-value matrix and value matrix. The attention mechanism is used to determine the behavioral feature variables based on the query matrix, key-value matrix and value matrix.

[0077] In this embodiment, an attention mechanism memory network is used to perform feature selection on local target features to generate behavioral feature variables.

[0078] Step 104: Input the behavioral feature variables into the supervised learning integrated network for dropout prediction, and output the MOOC student dropout prediction results.

[0079] Please note that Figure 3, the supervised learning ensemble network is the LightGBM model. LightGBM (LightGradient Boosting Machine) is an efficient implementation based on the gradient boosting framework, which is specially optimized for large-scale data sets. Compared with the traditional GBDT (Gradient Boosting Decision Tree) algorithm, LightGBM has significantly improved in speed and efficiency, mainly due to its two core improvements: the histogram algorithm (Histogram-based Algorithm) and the Leaf-wise growth strategy with depth restrictions. The histogram algorithm is one of the key technologies used in LightGBM to accelerate the training process. The algorithm first discretizes continuous feature values ​​into finite integers (i.e., buckets or bins), and then constructs a histogram based on these discrete values. When traversing the data, the algorithm accumulates the discretized values ​​into the corresponding histogram index, so as to quickly count the number of samples in each bucket, the sum of gradients, and other information. In addition, the histogram algorithm also supports histogram difference acceleration, that is, the histogram of a node can be obtained by subtracting the histogram of its parent node from the histogram of its sibling node, thereby avoiding repeated calculations and further improving computational efficiency. At the same time, its Leaf-wise growth strategy with depth restriction finds the leaf with the largest split gain from all leaves each time for splitting, thereby more efficiently improving the accuracy of the model. Since only the leaves with the largest split gain are split, the training error can be reduced faster, and the complexity of the model can be controlled by limiting the depth of the tree and the number of leaves to avoid overfitting. These improvements give LightGBM a significant advantage in processing large-scale data sets and are widely used in various machine learning tasks such as classification, regression, and sorting.

[0080] Furthermore, by using supervised learning ensemble networks to predict dropouts based on behavioral feature variables, the prediction accuracy can be improved, thereby accurately predicting the dropout behavior of learners in massive open online courses (MOOCs). The final prediction is made with the efficiency and accuracy of LightGBM. LightGBM builds a strong prediction model by integrating multiple weak learners to further improve the accuracy of the prediction.

[0081] In this embodiment, the behavioral feature variables are input into the supervised learning integrated network for dropout prediction, and the prediction results of MOOC students' dropout are output.

[0082] In an embodiment of the present invention, the present invention provides a method for predicting dropout of MOOC students. First, the behavior sequence data of the students to be predicted is obtained, and the behavior sequence data of the students to be predicted is input into a preset MOOC student dropout prediction model; the preset MOOC student dropout prediction model includes a convolution module, an attention mechanism memory network and a supervised learning integrated network; then, the local feature extraction of the student behavior sequence data to be predicted is performed through the convolution module, and the local target feature is output; the local target feature is selected by the attention mechanism memory network to generate behavior feature variables; finally, the behavior feature variables are input into the supervised learning integrated network for dropout prediction, and the MOOC student dropout prediction result is output; based on the above scheme, the local feature extraction of the student behavior sequence data to be predicted is performed through the convolution module, and the generated local target feature is selected in combination with the attention mechanism memory network, and finally the dropout prediction is performed based on the supervised learning integrated network, and the process of outputting the MOOC student dropout prediction result can dynamically adjust the importance of different features, thereby enhancing the accuracy of the model prediction result.

[0083] See also Figure 4 , Figure 4 A flowchart of the steps of another method for predicting dropout of MOOC students provided by an embodiment of the present invention.

[0084] The present invention provides a method for predicting dropout of MOOC students, comprising:

[0085] Step 401: Obtain the student behavior sequence data to be predicted, and input the student behavior sequence data to be predicted into a preset MOOC student dropout prediction model; the preset MOOC student dropout prediction model includes a convolution module, an attention mechanism memory network, and a supervised learning integrated network.

[0086] In this embodiment, the student behavior sequence data to be predicted is obtained, and the student behavior sequence data to be predicted is input into a preset MOOC student dropout prediction model; the preset MOOC student dropout prediction model includes a convolution module, an attention mechanism memory network and a supervised learning integrated network.

[0087] Step 402: extract local features from the student behavior sequence data to be predicted through a convolution module, and output local target features.

[0088] The convolution module includes convolutional layers and pooling layers.

[0089] Further, step 402 may include the following sub-steps:

[0090] S21, using multiple elements in the student behavior sequence data to be predicted to construct multiple behavior sub-matrices;

[0091] S22, performing convolution operation on each behavior sub-matrix through a convolution layer, and outputting a feature sub-matrix corresponding to each behavior sub-matrix;

[0092] S23, using each feature sub-matrix to construct a target feature matrix;

[0093] S24: Input the target feature matrix into the pooling layer for maximum pooling and output local target features.

[0094] In this embodiment, local features are extracted from the student behavior sequence data to be predicted through a convolution module, and local target features are output.

[0095] Step 403: Use a bidirectional long short-term memory module to extract state features of local target features and output a hidden state vector of the target cell.

[0096] The bidirectional long short-term memory module includes a forget gate, an input gate, an output gate, and an activation function layer (tanh activation function).

[0097] Further, step 403 may include the following sub-steps:

[0098] S31, extracting forgotten state features of local target features through a forget gate, and outputting a forgotten state vector;

[0099] S32, using an input gate to extract input state features of local target features to generate an input state vector;

[0100] S33, inputting the local target feature into the output gate to extract the output state feature and generate an output state vector;

[0101] S34, using an activation function layer to perform nonlinear mapping on local target features and output a nonlinear state vector;

[0102] S35, calculating the initial cell hidden state vector according to the nonlinear state vector, the forgotten state vector and the input state vector;

[0103] S36, determining an intermediate cell hidden state vector based on the initial cell hidden state vector and the output state vector;

[0104] S37, performing inverse processing on the intermediate cell hidden state vector, and outputting the inverse cell hidden state vector;

[0105] S38. Concatenate the reverse cell hidden state vector and the intermediate cell hidden state vector, and output the target cell hidden state vector.

[0106] Please note that Figure 5The bidirectional long short-term memory module is a BiLSTM model. In the BiLSTM model, the hidden state of each time step t is composed of the hidden states of the LSTM (Long Short-Term Memory) in two directions. The two directions are the positive time direction (also called forward) and the negative time direction (also called reverse). For the LSTM in the positive time direction, it processes the sequence from left to right, that is, according to the natural order of the input sequence. At time t, its hidden state is expressed as For the LSTM in the negative time direction, it processes the sequence from right to left, that is, it processes the input sequence in reverse order. At time t, its hidden state is expressed as LSTM (Long Short-Term Memory Network) selects information through its unique "gating" structure and effectively preserves long-term memory through the cell state update mechanism. In LSTM, the forget gate, input gate, and output gate are combined with the hidden state information of the previous moment. and the current input information , that is, the local target feature at time t, is calculated by the sigma activation function in the forget gate, input gate, and output gate to determine whether the cell state information is retained or discarded. In LSTM, the update of the cell state is a key step, which combines the forget gate, input gate, and the new candidate vector (initial cell hidden state vector) calculated by the tanh activation function to update the cell state.

[0107] Based on the above foundation, in BiLSTM, in order to make full use of the contextual information of the sequence (i.e., considering both past and future information), the hidden states in the positive time direction (also called forward) and the negative time direction (also called reverse) are concatenated to obtain the cell hidden state of BiLSTM at time t, i.e., the target cell hidden state vector; the processing process of the intermediate cell hidden state vector (positive sequence cell hidden state vector) can be expressed as:

[0108] ;

[0109] in, is the forgotten state vector at time t; is the first weight matrix; is the Sigmoid activation function; is the hyperbolic tangent function; is the hidden state information at time t-1, indicating the hidden state information at the previous moment; is the local target feature at time t; is the first bias vector; is the input state vector at time t; is the second weight matrix; is the second bias vector; is the third weight matrix; is the third bias vector; is the output state vector at time t; is the nonlinear state vector at time t; is the fourth weight matrix; is the fourth bias vector; is the initial cell hidden state vector at time t, representing the cell state at the update time; is the initial cell hidden state vector at time t-1, and represents the initial cell hidden state vector at the previous moment; is the intermediate cell hidden state vector at time t.

[0110] Furthermore, the processing of the target cell hidden state vector can be expressed as:

[0111] ;

[0112] in, is the hidden state vector of the target cell; is the hidden state vector of the intermediate cell; is the hidden state vector of the reverse cell; For splicing.

[0113] In this embodiment, a bidirectional long short-term memory module is used to extract state features of local target features and output a hidden state vector of the target cell.

[0114] Step 404: Perform a linear transformation on the target cell hidden state vector through a fully connected layer, and output a query matrix, a key-value matrix, and a value matrix.

[0115] In this embodiment, the target cell hidden state vector is linearly transformed through a fully connected layer to output a query matrix, a key-value matrix, and a value matrix.

[0116] Step 405: Use the attention mechanism to determine the behavior feature variables according to the query matrix, the key-value matrix and the value matrix.

[0117] Further, step 405 may include the following sub-steps:

[0118] S51, performing similarity operation on the query matrix and the key value matrix, and outputting the attention distribution weight;

[0119] S52. Perform weighted summation on the attention distribution weight and value matrix to determine the behavior characteristic variable.

[0120] It should be noted that the Attention Mechanism originated from the study of human attention. In daily life, due to the limited ability of humans to process information, we usually selectively focus on part of the information and ignore other visible information. This selective attention mechanism plays an important role in human perception, cognition and decision-making. Specifically, when it comes to the application of the attention mechanism in machine learning, its core idea is to simulate this selective attention ability of humans. When the model is learning and receiving a large amount of information, the attention mechanism can help the model effectively select and receive part of the information, thereby improving the performance and efficiency of the model.

[0121] For further information, see Figure 6 , the Attention mechanism is divided into the following steps: 1) In the Attention mechanism, first determine the query matrix (Query), key matrix (Key) and value matrix (Value); 2) Calculate the attention distribution weights based on the similarity or correlation between the query matrix and the key matrix. The processing of the attention distribution weights is usually normalized using the softmax function to ensure that the sum of all weights is 1, forming a valid probability distribution; 3) Based on the calculated attention distribution weights and the corresponding value matrix (Value), calculate the new input variable (behavior feature variable) by weighted summation.

[0122] In this embodiment, an attention mechanism is used to determine the behavior feature variables based on the query matrix, the key-value matrix and the value matrix.

[0123] Step 406: Input the behavior feature variables into the supervised learning integrated network to perform dropout prediction, and output the MOOC student dropout prediction results.

[0124] In this embodiment, the behavioral feature variables are input into the supervised learning integrated network for dropout prediction, and the prediction results of MOOC students' dropout are output.

[0125] Optionally, the training process of the MOOC student dropout prediction model is preset, including:

[0126] Obtaining the behavior sequence data of the students to be trained, inputting the behavior sequence data of the students to be trained into the initial MOOC student dropout prediction model, and outputting the dropout prediction results of the students to be trained;

[0127] The back propagation algorithm is used to calculate the model gradient based on the dropout prediction results of the MOOC students to be trained;

[0128] The model gradient is used to update the model parameters of the initial MOOC student dropout prediction model, determine the intermediate MOOC student dropout prediction model, and count the number of model updates in real time;

[0129] Determine whether the number of model updates reaches a preset update threshold;

[0130] If so, the intermediate MOOC student dropout prediction model is used as the trained preset MOOC student dropout prediction model.

[0131] It should be noted that the student behavior sequence data to be trained is the learner behavior data used for model training, including video viewing time, number of discussions participated in, homework submission status, etc. Specifically, historical learner behavior data (student behavior sequence data to be trained) is used as the training set, and the model parameters are optimized through the back propagation algorithm and gradient descent method. During the training process, a cross-validation strategy is used to evaluate the generalization ability of the model to effectively prevent the occurrence of overfitting. In order to verify the effectiveness and scientificity of the model, accuracy is selected as the main evaluation indicator of the model to comprehensively evaluate the prediction performance of the model; among them, the calculation formula of accuracy can be expressed as:

[0132] Acc = (TP + TN) / (TP + FP + FN + TN);

[0133] Among them, Acc is the accuracy rate; TP is the number of true positive examples, that is, the number of samples that actually dropped out and were predicted correctly; TN is the number of true negative examples, that is, the number of samples that actually did not drop out and were predicted correctly; FN is the number of false negative examples, that is, the number of samples that actually dropped out but were predicted to have no dropout; FP is the number of false positive examples, that is, the number of samples that actually did not drop out but were predicted to have dropped out.

[0134] Furthermore, if the number of model updates does not reach the preset update threshold, the intermediate MOOC student dropout prediction model is used as the new initial MOOC student dropout prediction model, and the step of inputting the student behavior sequence data to be trained into the initial MOOC student dropout prediction model and outputting the MOOC student dropout prediction result to be trained is jumped to the step until the number of model updates reaches the preset update threshold, and the intermediate MOOC student dropout prediction model determined when the number of model updates reaches the preset update threshold is used as the trained preset MOOC student dropout prediction model.

[0135] It is worth mentioning that the experimental environment of the present invention adopts Python 3.7.4 and Keras framework, and the server is configured with an x64 architecture processor, an NVIDIA GeForce RTX 3060 graphics card and 16G memory. In order to improve the learning effect of the neural network, Adam is selected as the optimizer, and the learning rate is set to 0.01. In terms of model structure, the CNN layer contains multiple convolution kernels and pooling layers; the BiLSTM layer is set to a bidirectional structure, and each layer contains a certain number of neurons; the attention mechanism adopts a multi-head attention mechanism to improve the ability to capture information. The LightGBM model is configured accordingly according to the data characteristics and task requirements. After multiple rounds of training and parameter adjustment, a fusion model with higher prediction performance (a trained preset MOOC student dropout prediction model) is finally obtained.

[0136] As a comparison of technical effects, we can refer to existing technologies. With the popularity of massive open online courses (MOOCs), they provide unprecedented learning opportunities for learners around the world. However, the high dropout rate of MOOCs has become a problem that needs to be solved urgently. This not only affects the learning effect of learners, but also poses a challenge to the effective use of educational resources. In order to improve the completion rate of MOOCs, researchers have begun to explore the use of advanced data analysis techniques to predict and intervene in potential dropout behaviors. However, how to extract effective prediction features from these complex and multi-dimensional data and build a high-precision prediction model is the core issue of current research.

[0137] Furthermore, current MOOCs dropout prediction technology faces the challenges of high model complexity, low prediction efficiency, high cost and difficulty in widespread application. Existing models are often bulky, consume large computing resources, and the prediction accuracy is limited by the limitations of data processing. In addition, the real-time performance and user experience of the prediction results need to be improved. The present invention aims to build an efficient and accurate MOOCs dropout prediction system by combining the CNN-BiLSTM and LightGBM models with attention mechanisms. The system optimizes the model structure, reduces computing costs, and improves prediction accuracy and real-time performance. At the same time, a friendly monitoring interface is designed to support remote monitoring by managers, timely intervention, and effective prevention of high dropout risks, thereby promoting the improvement of MOOCs education quality.

[0138] In view of the above problems, the present invention proposes a method for predicting the dropout of MOOC students, which aims to build an efficient MOOCs dropout prediction model (preset MOOCs student dropout prediction model) by combining the advantages of convolutional neural network (CNN), bidirectional long short-term memory network (BiLSTM) and LightGBM model, and introducing attention mechanism. Specifically, the model uses CNN to capture the local features of learning behavior data, BiLSTM to process the long-term dependencies of time series data, LightGBM as an integrated learning model to improve the prediction accuracy, and the attention mechanism is used to dynamically adjust the importance of different features, thereby enhancing the interpretability and prediction performance of the model; that is, by combining deep learning and integrated learning technology, a CNN-BiLSTM and LightGBM fusion model based on attention mechanism is constructed to achieve accurate prediction of the dropout behavior of learners of massive open online courses (MOOCs).

[0139] In summary, the CNN-BiLSTM and LightGBM models based on the attention mechanism of the present invention realize high-precision and low-cost MOOCs dropout prediction, ensuring efficient and reliable prediction. At the same time, the present invention provides educational institutions with dropout risk warnings through real-time predictions, reduces management costs, improves education quality, effectively protects user interests and efficient use of educational resources; compared with traditional MOOCs dropout prediction technology, the CNN-BiLSTM and LightGBM models based on the attention mechanism of the present invention significantly improve prediction accuracy and efficiency. Its advantage lies in deep mining of learning behavior data, combined with efficient algorithms, to achieve accurate prediction. At the same time, the model design is flexible and adaptable, providing strong decision-making support for educational institutions, effectively reducing the burden on managers, and promoting the continuous improvement of MOOCs education quality.

[0140] In an embodiment of the present invention, the present invention provides a method for predicting dropout of MOOC students. First, the behavior sequence data of the students to be predicted is obtained, and the behavior sequence data of the students to be predicted is input into a preset MOOC student dropout prediction model; the preset MOOC student dropout prediction model includes a convolution module, an attention mechanism memory network and a supervised learning integrated network; then, the local feature extraction of the student behavior sequence data to be predicted is performed through the convolution module, and the local target feature is output; the local target feature is selected by the attention mechanism memory network to generate behavior feature variables; finally, the behavior feature variables are input into the supervised learning integrated network for dropout prediction, and the MOOC student dropout prediction result is output; based on the above scheme, the local feature extraction of the student behavior sequence data to be predicted is performed through the convolution module, and the generated local target feature is selected in combination with the attention mechanism memory network, and finally the dropout prediction is performed based on the supervised learning integrated network, and the process of outputting the MOOC student dropout prediction result can dynamically adjust the importance of different features, thereby enhancing the accuracy of the model prediction result.

[0141] See also Figure 7 , Figure 7 A structural block diagram of a device for predicting dropout of MOOC students provided by an embodiment of the present invention.

[0142] The present invention provides a device for predicting dropout of MOOC students, comprising:

[0143] The acquisition module 701 is used to acquire the student behavior sequence data to be predicted, and input the student behavior sequence data to be predicted into a preset MOOC student dropout prediction model; the preset MOOC student dropout prediction model includes a convolution module, an attention mechanism memory network and a supervised learning integrated network;

[0144] An output module 702 is used to extract local features from the student behavior sequence data to be predicted through a convolution module and output local target features;

[0145] A generation module 703 is used to select local target features using an attention mechanism memory network to generate behavior feature variables;

[0146] The prediction module 704 is used to input the behavioral feature variables into the supervised learning integrated network to perform dropout prediction and output the MOOC student dropout prediction results.

[0147] Furthermore, the convolution module includes a convolution layer and a pooling layer; the output module 702 is specifically used for:

[0148] Using multiple elements in the student behavior sequence data to be predicted, multiple behavior sub-matrices are constructed;

[0149] Perform convolution operation on each behavioral sub-matrix through the convolution layer, and output the feature sub-matrix corresponding to each behavioral sub-matrix;

[0150] Using each feature sub-matrix, construct the target feature matrix;

[0151] The target feature matrix is ​​input into the pooling layer for maximum pooling and the local target features are output.

[0152] Furthermore, the attention mechanism memory network includes a bidirectional long short-term memory module and a fully connected layer; the generation module 703 includes:

[0153] The first submodule is used to extract state features of local target features using a bidirectional long short-term memory module and output a hidden state vector of the target cell;

[0154] The second submodule is used to perform a linear transformation on the hidden state vector of the target cell through a fully connected layer, and output a query matrix, a key-value matrix, and a value matrix;

[0155] The third submodule is used to determine the behavior feature variables according to the query matrix, the key-value matrix and the value matrix using the attention mechanism.

[0156] Furthermore, the bidirectional long short-term memory module includes a forget gate, an input gate, an output gate and an activation function layer; the first submodule is specifically used for:

[0157] The forgotten state feature of the local target feature is extracted through the forget gate, and the forgotten state vector is output;

[0158] The input gate is used to extract the input state features of the local target features and generate the input state vector;

[0159] Input the local target features into the output gate to extract the output state features and generate the output state vector;

[0160] The activation function layer is used to perform nonlinear mapping on the local target features and output a nonlinear state vector;

[0161] Calculate the initial cell hidden state vector according to the nonlinear state vector, the forgotten state vector and the input state vector;

[0162] Determine the intermediate cell hidden state vector based on the initial cell hidden state vector and the output state vector;

[0163] Reverse the hidden state vector of the intermediate cell and output the hidden state vector of the reversed cell;

[0164] The reverse cell hidden state vector and the intermediate cell hidden state vector are concatenated to output the target cell hidden state vector.

[0165] Furthermore, the third submodule is specifically used for:

[0166] Perform similarity operations on the query matrix and the key-value matrix, and output the attention distribution weights;

[0167] The attention distribution weights and value matrices are weighted and summed to determine the behavioral characteristic variables.

[0168] In an optional embodiment, the device further comprises:

[0169] The first module is used to obtain the behavior sequence data of the students to be trained, input the behavior sequence data of the students to be trained into the initial MOOC student dropout prediction model, and output the dropout prediction results of the MOOC students to be trained;

[0170] The second module is used to calculate the model gradient based on the dropout prediction results of the MOOC students to be trained using the back-propagation algorithm;

[0171] The third module is used to update the model parameters of the initial MOOC student dropout prediction model using the model gradient, determine the intermediate MOOC student dropout prediction model, and count the number of model updates in real time;

[0172] The fourth module is used to determine whether the number of model updates reaches a preset update threshold;

[0173] The fifth module is used for, if yes, using the intermediate MOOC student dropout prediction model as the trained preset MOOC student dropout prediction model.

[0174] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described devices, modules and sub-modules can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0175] An embodiment of the present invention further provides a computer device, including a memory and a processor, wherein a computer program is stored in the memory; when the computer program is executed by the processor, the processor executes the steps of the method for predicting dropout of MOOC students as described in any of the above embodiments.

[0176] An embodiment of the present invention further provides a computer-readable storage medium having a computer program / instruction stored thereon. When the computer program / instruction is executed by a processor, the steps of the method for predicting dropout of MOOC students as described in any of the above embodiments are implemented.

[0177] An embodiment of the present invention further provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the method for predicting dropout of MOOC students as described in any of the above embodiments.

[0178] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0179] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0180] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting dropout of MOOC students, characterized in that: include: Obtaining student behavior sequence data to be predicted, and inputting the student behavior sequence data to be predicted into a preset MOOC student dropout prediction model; the preset MOOC student dropout prediction model includes a convolution module, an attention mechanism memory network, and a supervised learning integrated network; Extracting local features of the student behavior sequence data to be predicted through the convolution module and outputting local target features; Using the attention mechanism memory network to perform feature selection on the local target features to generate behavior feature variables; The behavioral characteristic variables are input into the supervised learning integrated network for dropout prediction, and the prediction results of MOOC students' dropout are output.

2. The method for predicting dropout of MOOC students according to claim 1, characterized in that: The convolution module includes a convolution layer and a pooling layer; the step of extracting local features of the student behavior sequence data to be predicted by the convolution module and outputting local target features includes: Using multiple elements in the student behavior sequence data to be predicted, constructing multiple behavior sub-matrices; Performing a convolution operation on each of the behavioral sub-matrices through the convolution layer, and outputting a feature sub-matrix corresponding to each of the behavioral sub-matrices; Using each of the feature sub-matrices to construct a target feature matrix; The target feature matrix is ​​input into the pooling layer for maximum pooling, and local target features are output.

3. The method for predicting dropout of MOOC students according to claim 1, characterized in that: The attention mechanism memory network includes a bidirectional long short-term memory module and a fully connected layer; the step of using the attention mechanism memory network to perform feature selection on the local target features to generate behavior feature variables includes: Using the bidirectional long short-term memory module to extract state features of the local target features, and outputting a target cell hidden state vector; Performing a linear transformation on the target cell hidden state vector through a fully connected layer, and outputting a query matrix, a key-value matrix, and a value matrix; The attention mechanism is used to determine the behavior feature variables according to the query matrix, the key-value matrix and the value matrix.

4. The method for predicting dropout of MOOC students according to claim 3, characterized in that: The bidirectional long short-term memory module includes a forget gate, an input gate, an output gate and an activation function layer; the step of using the bidirectional long short-term memory module to extract state features of the local target features and outputting the target cell hidden state vector includes: Performing forgetting state feature extraction on the local target feature through the forgetting gate, and outputting a forgetting state vector; Using the input gate to extract input state features of the local target features to generate an input state vector; Inputting the local target feature into the output gate to extract the output state feature and generate an output state vector; Using the activation function layer to perform nonlinear mapping on the local target features and output a nonlinear state vector; Calculating an initial cell hidden state vector according to the nonlinear state vector, the forgetting state vector and the input state vector; Determining an intermediate cell hidden state vector based on the initial cell hidden state vector and the output state vector; Performing inverse processing on the intermediate cell hidden state vector, and outputting an inverse cell hidden state vector; The reverse cell hidden state vector and the intermediate cell hidden state vector are concatenated to output a target cell hidden state vector.

5. The method for predicting dropout of MOOC students according to claim 3, characterized in that: The step of using the attention mechanism to determine the behavior feature variables according to the query matrix, the key-value matrix and the value matrix includes: Performing a similarity operation on the query matrix and the key value matrix, and outputting an attention distribution weight; The attention distribution weight and the value matrix are weightedly summed to determine the behavior characteristic variable.

6. The method for predicting dropout of MOOC students according to claim 1, characterized in that: Also includes: Obtaining behavior sequence data of students to be trained, inputting the behavior sequence data of students to be trained into an initial MOOC student dropout prediction model, and outputting a dropout prediction result of students to be trained; Using a back propagation algorithm, the model gradient is calculated according to the prediction results of the MOOC students to be trained to drop out; The model gradient is used to update the model parameters of the initial MOOC student dropout prediction model, determine the intermediate MOOC student dropout prediction model, and count the number of model updates in real time; Determining whether the model update times reaches a preset update threshold; If yes, the intermediate MOOC student dropout prediction model is used as the trained preset MOOC student dropout prediction model.

7. A device for predicting dropout of MOOC students, characterized in that: include: An acquisition module is used to acquire the student behavior sequence data to be predicted, and input the student behavior sequence data to be predicted into a preset MOOC student dropout prediction model; the preset MOOC student dropout prediction model includes a convolution module, an attention mechanism memory network and a supervised learning integrated network; An output module, used for extracting local features of the student behavior sequence data to be predicted through the convolution module, and outputting local target features; A generation module, used for performing feature selection on the local target features by using the attention mechanism memory network to generate behavior feature variables; The prediction module is used to input the behavioral characteristic variables into the supervised learning integrated network to perform dropout prediction and output the MOOC student dropout prediction results.

8. A computer device, characterized in that: It includes a memory and a processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the processor executes the steps of the method for predicting dropout of MOOC students as described in any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed, the method for predicting the dropout of MOOC students as described in any one of claims 1 to 6 is implemented.

10. A computer program product, characterized in that The computer program product includes a computer program stored on a non-transitory computer-readable storage medium, wherein the computer program includes program instructions, wherein when the program instructions are executed by a computer, the computer is caused to execute the method for predicting dropout of MOOC students as described in any one of claims 1-6.

Citation Information

Patent Citations

  • Campus card data-based bad learner prediction method

    CN110705594A

  • CNNAE-LSTM-based MOOC prediction method and system

    CN115907123A

Cited By

  • MOOC stop prediction method and system based on large language model auxiliary graph node classification

    CN119323293A