Student learning effect prediction-based learning scheme recommendation method and system

Abstract

The invention relates to the technical field of electrical digital data processing, and provides a learning scheme recommendation method and system based on student learning effect prediction, and the method comprises the steps: obtaining a learning behavior log of a to-be-predicted user, carrying out the preprocessing, and constructing a learning behavior matrix; performing feature extraction on the learning behavior matrix based on the improved convolutional neural network to generate a learning behavior feature vector; the improved convolutional neural network is that an additional convolutional layer is arranged on a convolutional layer of an original convolutional neural network, the additional convolutional layer learns data distribution of a learning behavior matrix to obtain a position offset, and an original convolutional kernel sampling position is dynamically adjusted; extracting a time sequence feature vector from the learning behavior feature vector; and performing full-connection operation processing based on the full-connection network to obtain a prediction result, and performing learning scheme recommendation. Accurate prediction of the learning effect of the students on the online learning platform is realized, and a better learning scheme can be generated to realize recommendation of the learning scheme.

Description

Technical Field

[0001] The present invention relates to the technical field related to electronic digital data processing, and in particular to a method and system for recommending learning plans based on prediction of student learning outcomes. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] With the rapid development of Internet technology, online learning platforms have become an important way to acquire knowledge. However, due to the large number of learning resources and complex content on the platform, students often find it difficult to quickly locate the learning content that best suits their needs, resulting in low learning efficiency. At the same time, there are large differences in students' learning abilities, learning interests, and knowledge mastery, and a unified course recommendation method cannot meet personalized learning needs. In order to improve the learning experience and optimize the utilization of learning resources, an intelligent system based on learning plan recommendation has emerged. By analyzing students' learning behavior and knowledge mastery, it recommends the best learning path for them, thereby improving learning outcomes.

[0004] At present, learning recommendation solutions are mainly based on content recommendation and data mining technology, combining students' historical learning records and behavior data for personalized recommendations. Among them, the most effective method is recommendation based on learning prediction, that is, by analyzing students' learning behavior data, predicting their future learning outcomes, and adjusting recommended content based on the prediction results. This type of method can effectively identify students' learning shortcomings and implement targeted learning interventions, thereby improving learning efficiency and learning success rate.

[0005] Although the recommendation method based on learning prediction has good personalized recommendation effect, the current learning prediction method still has the following problems: (1) The distribution of learning behavior data is complex and the feature extraction capability is limited; students’ learning behavior data presents complex personalized characteristics and has different distribution characteristics. However, existing feature extraction methods mainly rely on convolutional neural networks (CNNs). The traditional convolution kernel sampling position is fixed and the receptive field range is limited. It is unable to fully learn the different distribution characteristics of students’ personalized learning behavior data, resulting in insufficient learning feature extraction.

[0006] (2) Ignoring the temporal characteristics of learning behavior: The learning process of students is continuous and temporal. Factors such as learning progress and learning frequency will affect their learning outcomes. However, most existing learning prediction models use static feature extraction methods, which affects the accuracy of prediction.

[0007] (3) The model has poor generalization ability and limited prediction accuracy. Most of the current mainstream learning prediction methods use a single deep learning model, such as a multi-layer perceptron (MLP) or a convolutional neural network (CNN). MLP is prone to overfitting when processing high-dimensional features, while the performance of traditional CNNs decreases when extracting long sequences of learning behavior data, resulting in limited prediction accuracy. It can be seen that the generalization ability of a single deep learning model is poor, and it is difficult to adapt to the learning behavior patterns of different types of students, which affects the robustness of the recommendation system. Summary of the invention

[0008] In order to solve the above problems, the present invention proposes a learning plan recommendation method and system based on student learning outcome prediction, which fully mines the students' personalized learning behavior characteristics and time series characteristics, integrates the student learning outcome prediction model of multiple deep learning methods, and realizes the accurate prediction of the learning outcomes of students on the online learning platform, thereby generating a better learning plan and realizing the recommendation of learning plans.

[0009] In order to achieve the above object, the present invention adopts the following technical solution: One or more embodiments provide a method for recommending a learning plan based on prediction of student learning outcomes, comprising the following steps: Obtain the learning behavior log of the user to be predicted, perform preprocessing and construct it into a learning behavior matrix; Based on the improved convolutional neural network, the learning behavior matrix is ​​feature extracted to generate the learning behavior feature vector; The improved convolutional neural network is as follows: an additional convolutional layer is set on the convolutional layer of the original convolutional neural network, the additional convolutional layer learns the data distribution of the learning behavior matrix to obtain the position offset, and dynamically adjusts the original convolution kernel sampling position; The learning behavior feature vector is used with a bidirectional long short-term memory network to extract the time sequence information in the learning behavior feature and generate a time sequence feature vector; A fully connected network with a two-layer neuron structure is constructed. The time series feature vector is fully connected to obtain the main feature components, and the feature vector is reduced in dimension to obtain the learning effect prediction result. A recommended learning plan is generated based on the learning effect.

[0010] Furthermore, the bidirectional long short-term memory network adds a reverse long short-term memory network on the basis of the forward long short-term memory network. The forward long short-term memory network is used to capture the forward dependency of learning behavior characteristics over time, and the reverse long short-term memory network models the reverse dependency of learning behavior characteristics. One or more embodiments provide a learning plan recommendation system based on student learning effectiveness prediction, including: A learning behavior matrix construction module is configured to obtain the learning behavior log of the user to be predicted, perform preprocessing and construct the learning behavior matrix; A learning behavior feature extraction module is configured to dynamically adjust the sampling position of the constructed learning behavior matrix based on the improved convolutional neural network to perform feature extraction and generate a learning behavior feature vector; A time series feature extraction module is configured to use a bidirectional long short-term memory network to extract time series information from the learning behavior feature vector in a bidirectional manner, and generate a time series feature vector of the user to be predicted; The prediction and recommendation module is configured to use a fully connected network to perform a fully connected operation on the time series feature vector, and reduce the dimension of the feature vector to obtain a learning effect prediction result, and generate a recommended learning plan based on the learning effect.

[0011] Compared with the prior art, the present invention has the following beneficial effects: The present invention combines deep learning and intelligent recommendation algorithms to predict students' learning outcomes and provide personalized learning plan recommendations for students based on the prediction results. The improved convolutional neural network (CNN) adaptive convolution extracts personalized learning behavior features, the bidirectional long short-term memory network (BiLSTM) extracts time series information, and the fully connected network (FC) is used to predict the results, making the learning plan recommendation more targeted and intelligent. Ultimately, this method can significantly improve the learning resource matching degree of the online learning platform, improve students' learning outcomes, and achieve more accurate and efficient personalized learning recommendations.

[0012] The advantages of the present invention and additional advantages will be described in detail in the following specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] The accompanying drawings constituting a part of the present invention are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their description are used to explain the present invention but do not constitute a limitation of the present invention.

[0014] Figure 1 is a method flow chart of a learning plan recommendation method according to Embodiment 1 of the present invention; Figure 2 is a structural diagram of a learning effectiveness prediction model according to Embodiment 1 of the present invention; Figure 3 Schematic diagram of the training process of the learning achievement prediction model of Example 1 of the present invention; Figure 4 It is a block diagram of a learning plan recommendation system according to Embodiment 2 of the present invention. DETAILED DESCRIPTION

[0015] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0016] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0017] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof. It should be noted that, in the absence of conflict, the various embodiments of the present invention and the features in the embodiments can be combined with each other. The embodiments will be described in detail below in conjunction with the accompanying drawings.

[0018] Example 1 In the technical solutions disclosed in one or more embodiments, Figures 1 to 3 As shown in FIG. 1 , the learning plan recommendation method based on the prediction of student learning outcomes includes the following steps: Step 1: Obtain the learning behavior log of the user to be predicted, preprocess it and construct it into a learning behavior matrix; Step 2: Extract features from the learning behavior matrix based on the improved convolutional neural network to generate a learning behavior feature vector; The improved convolutional neural network is as follows: an additional convolutional layer is set on the convolutional layer of the original convolutional neural network, the additional convolutional layer learns the data distribution of the learning behavior matrix to obtain the position offset, and dynamically adjusts the original convolution kernel sampling position; Step 3: Use a bidirectional long short-term memory network to extract the time series information in the learning behavior characteristics and generate a time series feature vector. Step 4: Build a fully connected network with a two-layer neural structure, perform a fully connected operation on the time series feature vector to obtain the main feature components, reduce the dimension of the feature vector, obtain the learning effect prediction result, and generate a recommended learning plan based on the learning effect.

[0019] In this embodiment, a convolutional neural network that dynamically adjusts the sampling position is used, that is, an adaptive convolution kernel is introduced, which can optimize local sampling for different students' learning behavior data and improve the ability to extract personalized features. This can more accurately capture students' behavior patterns in different course cycles, thereby improving the feature expression ability of the prediction model; a bidirectional LSTM is used to extract time series features from both the forward and backward directions, so that the model can fully capture the long-term dependencies of learning behaviors and improve the accuracy of learning outcome prediction. A fully connected network is used to reduce the dimensionality of the extracted time series features, and a Softmax or Sigmoid function is used to generate learning outcome prediction results, which are used to further guide the recommendation of learning plans so that the recommended learning resources are highly matched with the students' current learning status.

[0020] This embodiment combines deep learning and intelligent recommendation algorithms, integrates multiple deep learning methods, predicts student learning outcomes, and provides personalized learning plan recommendations for students based on the prediction results. The improved convolutional neural network (CNN) adaptive convolution extracts personalized learning behavior features, the bidirectional long short-term memory network (BiLSTM) extracts time series information, and the fully connected network (FC) predicts the results, making the learning plan recommendation more targeted and intelligent. Ultimately, this method can significantly improve the matching degree of learning resources on the online learning platform, improve student learning outcomes, and achieve more accurate and efficient personalized learning recommendations.

[0021] In the online learning platform, students have completed the courses and obtained accurate learning results, and can obtain each student's learning behavior log; In step 1, the students’ learning behavior logs are preprocessed. Data preprocessing includes missing value processing and format conversion, specifically: Step 11: Fill in the missing values ​​in the student learning behavior log: If the grade attribute is null, fill it with 0; if the date attribute is null, fill it with the learning effectiveness evaluation date; due to the reasons such as students not submitting or not participating in learning, there are problems such as missing data. Such samples will affect the model's learning of features. By analyzing the feature definition, the missing values ​​are filled; the filling strategy can avoid abnormal data distribution, make the model learning more stable, and the extracted features more representative.

[0022] Step 12: Convert the text data of the student learning behavior log into digital data: convert the varchar data into digital data, including replacing binary features with 0 and 1, converting the time format data, and marking the converted timestamp starting from 1.

[0023] Step 13: Construct a learning behavior matrix, count the number of occurrences of each learning behavior of students in the course cycle, and obtain the learning behavior matrix, which can be expressed as: ; in, Indicates the course period, It represents the number of learning behavior categories, the rows of the matrix represent the various types of learning behaviors of students, and the columns of the matrix represent the course cycle.

[0024] Specifically, learning behavior categories can include: Video viewing: such as the number of views and viewing duration of online course videos; Submitting homework: the number of times the homework is submitted, the completion rate, the accuracy rate, etc.; Participation in discussions: such as the number of times you speak and reply in the course; Tests / Exams: participation in online tests, scores, etc.; Reading textbooks / courseware: the number of times you browsed PDF, e-books and other learning materials; Interactive behaviors: such as likes, favorites, and highlighting; Course access frequency: number of times you log into the learning platform, duration of visits, etc.

[0025] In the above steps, the learning behavior data of students on the online learning platform are counted, and the learning behavior logs with different data types and complex data are processed into a learning behavior matrix with only digital format and clear and concise form, which enhances the model's adaptability to students' learning behavior data and improves the ability to represent students' related characteristics.

[0026] In step 2 and step 3, feature extraction is performed to complete feature extraction for the learning behavior matrix, including student behavior feature extraction and student time series feature extraction; Step 2 is implemented by constructing a learning feature extraction module, which includes a position-modified convolutional neural network, a pooling and a flattening module; Furthermore, the improved convolutional neural network adds a convolutional layer based on the original convolution kernel of the convolutional neural network to calculate the offset of the sampling position. By learning the distribution of learning behavior data in the learning behavior matrix, the position offset is obtained, and the sampling position is dynamically adjusted based on the offset. The learning behavior feature extraction module uses an improved convolutional neural network. An additional convolution layer is added to the original convolution kernel to analyze the data distribution characteristics and generate an offset to add to the sampling position of the original convolution layer. The convolution kernel can dynamically adjust the sampling position according to the distribution characteristics of the data, so that the model has a better receptive field to generate a learning behavior feature vector. In step 2, student behavior features are extracted from the learning behavior matrix, including the following steps: Step 201: Use an additional convolution layer to learn the distribution of data positions in the learning behavior matrix to obtain a position offset, and add the offset to the original convolution kernel sampling position to update the original convolution kernel sampling position; The method of using an additional convolution layer to learn the distribution of data positions in a learning behavior matrix, obtaining a position offset, and adding the offset to the original convolution kernel sampling position comprises the following steps: Step 2-1: Perform convolution operation on the data of the learning behavior matrix to generate the offset matrix ; The offset matrix Each sampling position in contains offset values ​​in two directions, corresponding to the offset of the convolution kernel sampling position in two orthogonal directions; Use the convolution kernel to learn the distribution of the data position of the learning behavior matrix and calculate the position offset The method is as follows: An additional convolution operation is applied to the input matrix, and the number of channels of the convolution layer is ,in is the number of sampling positions in the convolution kernel, calculated as follows:

[0027] in, is the offset matrix, each value in the matrix contains offset vector , represents the offset in the x and y directions; represents the learning behavior matrix of the input, Represents the weight parameters of the convolutional layer, which can be optimized during model training; Step 2-2: The offset matrix The offset of each sample position in , based on the learnable directional weights , Make adjustments, multiply the directional weight by the corresponding directional offset to obtain the adjusted offsets in the two directions; In order to adjust the offset contribution in two directions, learnable directional weights are introduced respectively , , the initial value is set, and it can be optimized by gradient descent during training; the two directions are weighted separately, and the formula is as follows:

[0028]

[0029] Step 2-3, perform mean normalization processing on the adjusted offset to obtain the final value of the offset; The mean normalization method is used to avoid abnormal offset distribution and maintain the stability of the offset amplitude. The calculation method is as follows:

[0030]

[0031] in, , They are , The mean of , is the standard deviation.

[0032] The final calculated offset value is .

[0033] Step 2-4: Add the calculated offset to the sampling position of the original convolution kernel to obtain the updated sampling position of the convolution kernel, and perform bilinear interpolation on the updated sampling position so that the sampling position falls at an integer position; Since the sampling position after adding the offset ( ) may fall at a non-integer position. This embodiment uses a bilinear interpolation method to ensure that the sampling position where the offset is added falls at an integer position.

[0034] Step 202: Use the original convolution kernel with the offset added to perform a convolution operation on the learning behavior matrix to obtain a feature extraction result. This step is implemented in a convolutional neural network, and the extracted features are expressed as: ; in, Represents the weight of the convolution kernel corresponding to the position, Represents the input learning behavior matrix The element value at position, Represents the output learning behavior matrix The feature value of the position, that is, the feature extracted by the convolution kernel; Represents the offset position of each point in the convolution kernel relative to the center point of the convolution kernel, Indicates the position offset.

[0035] Step 203: Perform a maximum pooling operation on the feature extraction result, i.e., the convolved learning behavior matrix, and obtain the student behavior features by flattening the matrix; this step is implemented in the pooling and flattening module; In this step, the maximum pooling operation is used to reduce the feature dimension and avoid overfitting in the model prediction process; the maximum pooling calculation method is: ; in, Represents the first Line The value of the column, Represents the input matrix Line The pooling window of the column, Represents the input convolution in the learning behavior matrix Line The value of the column is the element value in the feature matrix constructed by the features extracted in step 202.

[0036] This embodiment aims at the problem that the receptive field range of existing feature extraction methods is limited, such as convolutional neural networks (CNN). The traditional convolution kernel sampling position is fixed, and it is impossible to fully learn the different distribution characteristics of students' personalized learning behavior data, resulting in insufficient learning feature extraction. By improving the convolutional neural network (CNN), a convolution layer is added on the basis of the traditional convolution kernel to calculate the offset of the sampling position, where the offset is calculated by learning the distribution of students' learning behavior data. It is realized that the sampling position of the convolution kernel can be flexibly changed according to the distribution characteristics of students' learning behavior data, which expands the receptive field of the convolution kernel, increases the convolution efficiency, and fully taps the characteristics of students' personalized learning behavior.

[0037] In step 3, a time series feature extraction module is constructed to implement the process of step 3, and a bidirectional long short-term memory network can be used to extract a time series feature vector; The long short-term memory network is used to extract the student's temporal features, and the student behavior feature vector after flattening the learning behavior matrix after step 2 is used as the input of the long short-term memory network. The long short-term memory network is connected through the input gate , Forget Gate and output gate Control the transmission of input features and obtain timing features by updating the cell state. The calculation method is: ; ; ; ; ; ; in, represents the data input into LSTM, Indicates the current time t The output of the input gate is Indicates the current time t The output of the forget gate is Indicates the current timet The output result of the output gate, Indicates the current time t Update status, Indicates the update status of the last moment. Indicates newly added information. Represents the output of the LSTM unit at the current moment, Represents the output of the LSTM unit at the previous moment ， 、 、 、 represents the preset weight matrix, 、 、 、 represents the preset weight parameter, represents the sigmoid activation function, represents the tanh activation function, Represents an element-by-element multiplication operation. The final output is From the forward LSTM result And the reverse LSTM results The splicing is expressed as: ; The bidirectional long short-term memory network adds a reverse long short-term memory network based on the forward long short-term memory network. The forward long short-term memory network is used to capture the forward dependency of learning behavior features over time, and the reverse long short-term memory network models the reverse dependency of learning behavior features. The time series feature vector after time series feature extraction is used as the final feature vector; In this embodiment, when processing the student learning behavior characteristics, the classification model takes into account the implicit temporal characteristics in the student learning behavior, and uses a bidirectional long short-term memory network to extract the temporal characteristics in the student learning behavior characteristics, thereby improving the model prediction performance.

[0038] In step 4, the full connection operation can be implemented by a multilayer perceptron. The final feature vector is used as the input of the multilayer perceptron. In the fully connected network of the multilayer perceptron, each neuron is connected to the neuron in the previous layer, and each connection is weighted. The calculation method of the fully connected output is: ; in, represents the fully connected network The output of a neuron, Indicates The output of the input neuron is Indicates the connection Input neuron to The weights of the output neurons, Indicates The bias of a neuron, Represents the activation function. After the feature vector is processed by the multi-layer perceptron and reduced in dimension, the final prediction result is output; Specifically, the fully connected network can adopt a three-layer fully connected network; Furthermore, for the current course, after the learning outcomes of all users are predicted through step 4, the learning outcomes prediction results of the users to be predicted are sent to the users. If the learning outcomes prediction results of the users to be predicted are poor, a warning message is sent to the users to be predicted.

[0039] Furthermore, the learning solution recommendation method may adopt a collaborative filtering algorithm, including: Step 41: Construct a user-learning resource matrix, analyze the learning trajectories of different students, and obtain the learning trajectories of similar students; User-learning resource matrix: used to represent the interaction relationship between users (students) and learning resources; Step 42: Based on the learning trajectories of similar students and the collaborative filtering algorithm (CF), the learning resources that best match the current student situation are recommended to achieve learning plan recommendations.

[0040] Similar students refer to students who have similar patterns in learning behavior, learning performance or knowledge mastery, which can be achieved by calculating the similarity of the data.

[0041] The above method process can be realized by constructing a learning effectiveness prediction model, which includes a learning feature extraction module, a time series feature extraction module and a fully connected network; Furthermore, it also includes a method for training a learning effectiveness prediction model, such as Figure 3 As shown, the steps are as follows: Step S1, obtaining each student's historical learning behavior log and learning achievement label; preprocessing the learning behavior log and constructing it into a learning behavior matrix, and constructing a training set and a test set; Specifically, the learning outcome labels are labeled data, which are classified into two categories: qualified and unqualified; Step S2, based on the improved convolutional neural network, the learning behavior matrix in the training set is subjected to feature extraction by dynamically adjusting the sampling position to generate a learning behavior feature vector; This step is implemented in the same way as the above step 2, and will not be repeated here; Step S3, using a bidirectional long short-term memory network to extract the time series information in the learning behavior characteristics, and generating a time series feature vector; This step is implemented in the same way as step 3 above, and will not be repeated here; Step S4: using a fully connected network to perform a fully connected operation on the time series feature vector, and reducing the dimension of the feature vector to obtain a learning effect prediction result; Step S5, calculating the prediction loss of the learning effect prediction result, adjusting the parameters of the model, iteratively executing the above steps S2 to S5 for training until the iteration cutoff condition is met, and obtaining the trained learning effect prediction model; Based on the trained learning outcome prediction model, the test set is used for testing. After the test accuracy is met, the learning situation of the students to be predicted in the online platform is predicted to obtain the prediction results, and then the learning plan is recommended based on the prediction structure.

[0042] This embodiment adopts a neural network model and integrates multiple deep learning algorithms. By optimizing the convolution kernel, the convolution kernel can dynamically adjust the sampling position according to different data distribution characteristics, so that the model has a better receptive field and improves the model's feature extraction ability when facing different data distributions. When facing learning behavior matrices with different data distribution characteristics, it can not only extract students' global learning behavior characteristics, but also fully explore students' local personalized learning behavior characteristics; use a bidirectional long short-term memory network to extract time series features. It can effectively identify students with poor learning results and conduct early warning intervention on students, and can adjust learning plans in time to improve learning efficiency, reduce the dropout rate of online learning platforms, and make more full use of platform educational resources.

[0043] Example 2 Based on Example 1, this embodiment provides a learning plan recommendation system based on student learning achievement prediction, such as Figure 4 As shown, including: A learning behavior matrix construction module is configured to obtain the learning behavior log of the user to be predicted, perform preprocessing and construct the learning behavior matrix; A learning behavior feature extraction module is configured to extract features from the learning behavior matrix based on the improved convolutional neural network to generate a learning behavior feature vector; The improved convolutional neural network is as follows: an additional convolutional layer is set on the convolutional layer of the original convolutional neural network, the additional convolutional layer learns the data distribution of the learning behavior matrix to obtain the position offset, and dynamically adjusts the original convolution kernel sampling position; A time series feature extraction module is configured to use a bidirectional long short-term memory network to extract time series information from the learning behavior feature vector in a bidirectional manner to generate a time series feature vector; The prediction and recommendation module constructs a fully connected network with a two-layer neural structure, performs a fully connected operation on the time series feature vector to obtain the main feature components, and reduces the dimension of the feature vector to obtain the learning effect prediction result, and generates a recommended learning plan based on the learning effect.

[0044] The improved convolutional neural network adds a convolution layer on the basis of the original convolution kernel of the convolutional neural network to calculate the offset of the sampling position; the position offset is obtained by learning the distribution of learning behavior data in the learning behavior matrix, and the sampling position is dynamically adjusted based on the offset.

[0045] Furthermore, the bidirectional long short-term memory network adds a reverse long short-term memory network based on the forward long short-term memory network. The forward long short-term memory network is used to capture the forward dependency of learning behavior characteristics over time, and the reverse long short-term memory network models the reverse dependency of learning behavior characteristics.

[0046] Furthermore, the learning behavior feature extraction module extracts student behavior features from the learning behavior matrix, including the following steps: An additional convolutional layer is used to learn the distribution of data positions in the learning behavior matrix to obtain the position offset, and the offset is added to the original convolution kernel sampling position to update the original convolution kernel sampling position; The original convolution kernel with an offset added is used to perform a convolution operation on the learning behavior matrix to obtain the feature extraction result; The feature extraction results are subjected to maximum pooling operation, and the student behavior characteristics are obtained by matrix flattening.

[0047] It should be noted here that the various modules in this embodiment correspond one-to-one to the various steps in Example 1, and the specific implementation process is the same, which will not be repeated here.

[0048] Example 3 This embodiment provides an electronic device, including a memory and a processor, and computer instructions stored in the memory and executed on the processor. When the computer instructions are executed by the processor, the steps of the method for recommending a learning plan based on prediction of student learning outcomes in Embodiment 1 are completed.

[0049] Example 4 This embodiment provides a computer-readable storage medium for storing computer instructions. When the computer instructions are executed by a processor, the steps in the method for recommending a learning plan based on prediction of student learning outcomes of Embodiment 1 are completed.

[0050] The electronic device proposed in the present invention can be a mobile terminal or a non-mobile terminal. The non-mobile terminal includes a desktop computer, and the mobile terminal includes a smart phone (such as an Android phone, an IOS phone, etc.), smart glasses, a smart watch, a smart bracelet, a tablet computer, a laptop computer, a personal digital assistant, and other mobile Internet devices that can perform wireless communication.

[0051] It should be understood that in the present invention, the processor may be a central processing unit CPU, and the processor may also be other general-purpose processors, digital signal processors DSP, application-specific integrated circuits ASIC, off-the-shelf programmable gate arrays FPGA or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0052] The memory may include a read-only memory and a random access memory, and provide instructions and data to the processor, and a portion of the memory may also include a non-volatile random access memory. For example, the memory may also store information about the device type.

[0053] In the implementation process, each step of the above method can be completed by an integrated logic circuit of hardware in a processor or an instruction in the form of software. The steps of the method disclosed in the present invention can be directly embodied as being executed by a hardware processor, or being executed by a combination of hardware and software modules in a processor. The software module can be located in a mature storage medium in the art such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The storage medium is located in a memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware. To avoid repetition, it is not described in detail here. A person of ordinary skill in the art can appreciate that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are executed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0054] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0055] In the several embodiments provided by the present invention, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a division of logical functions. 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, and the indirect coupling or communication connection of devices or units can be electrical, mechanical or other forms.

[0056] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program codes.

[0057] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A learning plan recommendation method based on student learning effectiveness prediction, characterized in that: The steps include: Obtain the learning behavior log of the user to be predicted, perform preprocessing and construct it into a learning behavior matrix; Based on the improved convolutional neural network, the learning behavior matrix is ​​feature extracted to generate the learning behavior feature vector; The improved convolutional neural network is as follows: an additional convolutional layer is set on the convolutional layer of the original convolutional neural network, the additional convolutional layer learns the data distribution of the learning behavior matrix to obtain the position offset, and dynamically adjusts the original convolution kernel sampling position; The learning behavior feature vector is used with a bidirectional long short-term memory network to extract the time sequence information in the learning behavior feature in a bidirectional manner and generate a time sequence feature vector; A fully connected network with a two-layer neuron structure is constructed. The time series feature vector is fully connected to obtain the main feature components, and the feature vector is reduced in dimension to obtain the learning effect prediction result. A recommended learning plan is generated based on the learning effect.

2. The method for recommending a learning plan based on prediction of student learning outcomes according to claim 1, characterized in that: The bidirectional long short-term memory network adds a reverse long short-term memory network based on the forward long short-term memory network. The forward long short-term memory network is used to capture the forward dependency of learning behavior characteristics over time, and the reverse long short-term memory network models the reverse dependency of learning behavior characteristics.

3. The method for recommending a learning plan based on prediction of student learning outcomes according to claim 1, characterized in that: Extracting student behavior features from the learning behavior matrix includes the following steps: An additional convolutional layer is used to learn the distribution of data positions in the learning behavior matrix to obtain the position offset, and the offset is added to the original convolution kernel sampling position to update the original convolution kernel sampling position; The original convolution kernel with an offset added is used to perform a convolution operation on the learning behavior matrix to obtain the feature extraction result; The feature extraction results are subjected to maximum pooling operation, and the student behavior characteristics are obtained by matrix flattening.

4. The method for recommending a learning plan based on prediction of student learning outcomes as claimed in claim 3, characterized in that: The method of using an additional convolution layer to learn the distribution of data positions in a learning behavior matrix, obtaining a position offset, and adding the offset to the original convolution kernel sampling position comprises the following steps: Performing a convolution operation on the data of the learning behavior matrix to generate an offset matrix; each sampling position in the offset matrix contains offset values ​​in two directions, corresponding to the offsets of the sampling position of the convolution kernel in two orthogonal directions; The offset of each sampling position in the offset matrix is ​​adjusted based on the learnable directional weight, and the directional weight is multiplied by the corresponding directional offset to obtain the adjusted offsets in the two directions; Perform mean normalization on the adjusted offset to obtain the final value of the offset; The calculated offset is added to the sampling position of the original convolution kernel to obtain the updated sampling position of the convolution kernel, and the updated sampling position is bilinearly interpolated so that the sampling position falls at an integer position.

5. The method for recommending a learning plan based on prediction of student learning outcomes according to claim 1, characterized in that: The fully connected operation is processed using a fully connected network of multi-layer perceptron.

6. The method for recommending a learning plan based on prediction of student learning outcomes according to claim 1, characterized in that: The learning plan recommendation method adopts collaborative filtering algorithm, which includes the following steps: Construct a user-learning resource matrix to analyze the learning trajectories of different students and obtain the learning trajectories of similar students; the user-learning resource matrix is ​​used to represent the interactive relationship between users and learning resources; According to the learning trajectories of similar students and based on collaborative filtering algorithms, learning resources that best match the current student situation are recommended to achieve learning plan recommendations.

7. The method for recommending a learning plan based on prediction of student learning outcomes according to claim 1, characterized in that: It also includes building a learning effectiveness prediction model, which includes a learning feature extraction module, a time series feature extraction module, and a fully connected network; The method for training the learning effectiveness prediction model is as follows: Obtain each student's historical learning behavior log and learning achievement labels; Preprocess the learning behavior log and construct it into a learning behavior matrix, and build training and test sets; Based on the improved convolutional neural network, the learning behavior matrix in the training set is used to dynamically adjust the sampling position for feature extraction to generate a learning behavior feature vector; The learning behavior feature vector is used with a bidirectional long short-term memory network to extract the time sequence information in the learning behavior feature in a bidirectional manner and generate a time sequence feature vector; A fully connected network is used to perform full connection operations on the time series feature vector, and the feature vector is reduced in dimension to obtain the learning effect prediction result; Calculate the prediction loss of the learning effect prediction result, adjust the parameters of the model, and iterate the training until the iteration cutoff condition is met to obtain the trained learning effect prediction model.

8. A learning plan recommendation system based on student learning outcome prediction, characterized in that: include: A learning behavior matrix construction module is configured to obtain the learning behavior log of the user to be predicted, perform preprocessing and construct the learning behavior matrix; A learning behavior feature extraction module is configured to extract features from the learning behavior matrix based on the improved convolutional neural network to generate a learning behavior feature vector; The improved convolutional neural network is as follows: an additional convolutional layer is set on the convolutional layer of the original convolutional neural network, the additional convolutional layer learns the data distribution of the learning behavior matrix to obtain the position offset, and dynamically adjusts the original convolution kernel sampling position; A time series feature extraction module is configured to use a bidirectional long short-term memory network to extract time series information from the learning behavior feature vector in a bidirectional manner to generate a time series feature vector; The prediction and recommendation module constructs a fully connected network with a two-layer neural structure, performs a fully connected operation on the time series feature vector to obtain the main feature components, and reduces the dimension of the feature vector to obtain the learning effect prediction result, and generates a recommended learning plan based on the learning effect.

9. The learning plan recommendation system based on student learning achievement prediction according to claim 8, characterized in that: The bidirectional long short-term memory network adds a reverse long short-term memory network based on the forward long short-term memory network. The forward long short-term memory network is used to capture the forward dependency of learning behavior characteristics over time, and the reverse long short-term memory network models the reverse dependency of learning behavior characteristics.

10. The learning plan recommendation system based on student learning achievement prediction according to claim 8, characterized in that: The learning behavior feature extraction module extracts student behavior features from the learning behavior matrix, including the following steps: An additional convolutional layer is used to learn the distribution of data positions in the learning behavior matrix to obtain the position offset, and the offset is added to the original convolution kernel sampling position to update the original convolution kernel sampling position; The original convolution kernel with an offset added is used to perform a convolution operation on the learning behavior matrix to obtain the feature extraction result; The feature extraction results are subjected to maximum pooling operation, and the student behavior characteristics are obtained by matrix flattening.

Images