A learning plan recommendation method and system based on learning effectiveness prediction

Abstract

The present invention relates to the technical field of learning program recommendation, and provides a learning program recommendation method and system based on learning effectiveness prediction, comprising: obtaining basic data of a user to be predicted and learning behavior data of a certain course, and after preprocessing, constructing and screening classification features to obtain a feature vector of the user to be predicted, inputting a trained online course student learning effectiveness prediction model, predicting the learning effectiveness of the user to be predicted on the course, sending warning information to the user to be predicted when the learning effectiveness is not passed, and obtaining the learning efficiency and learning program of a reference user, and sending the learning program of the reference user with the highest learning efficiency to the user to be predicted; and improving the generalization ability of the model for different samples by integrating multiple machine learning models, achieving accurate warning for students with failure risks and recommending the best and most suitable learning program for the warned students.

Description

Technical Field

[0001] The present invention belongs to the technical field of learning program recommendation, and in particular, relates to a learning program recommendation method and system based on learning effectiveness prediction. 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] The virtual learning environment combines remote interaction and face-to-face interaction to provide a virtual learning environment, where the learning atmosphere and sense of presence are closer to offline traditional teaching. The virtual learning environment generates a large amount of student learning behavior data, which can be used to predict student learning outcomes, track student learning outcomes in real time, predict whether students can pass course assessments, and recommend learning plans.

[0004] However, the existing learning effectiveness prediction and learning plan recommendation are carried out independently. The existing learning plan recommendation method is for all students in the virtual learning environment, which lacks pertinence. In addition, the learning plan of historical users is adopted, which is not applicable to the target users due to changes in policies, etc. The existing student learning effectiveness prediction methods can be divided into the following categories from the perspective of classification methods:

[0005] (1) Classification method based on time series data. This method converts the log data of students’ learning behavior into time series data and mines the time patterns between the data for prediction.

[0006] (2) Classification method based on machine learning. This method collects students’ learning behavior data and uses models such as support vector machine, K nearest neighbor, naive Bayes, random forest, artificial neural network, or integrated learning models based on the above models to predict learning outcomes.

[0007] (3) Classification method based on deep learning. This method uses evaluation streams and click streams to collect students’ learning behavior data in a virtual learning environment, adopts a data completion mechanism to fill in missing stream data, and uses deep learning models such as convolutional neural networks and recurrent neural networks to predict learning outcomes. It can process both static data and sequence data at the same time.

[0008] The above learning outcome prediction methods cannot identify key factors among numerous influencing factors, cannot be applied to imbalanced data sets, cannot be used for different samples, cannot accurately identify students at risk of failure, cannot provide accurate warnings, and cannot be used to accurately recommend learning plans. Summary of the invention

[0009] In order to solve the technical problems existing in the above-mentioned background technology, the present invention provides a learning plan recommendation method and system based on learning achievement prediction, which integrates multiple machine learning models. The base classifier has the characteristics of high diversity, which improves the generalization ability of the model for different samples, and realizes accurate warning of students at risk of failure and recommends the best and most suitable learning plan for the warned students.

[0010] In order to achieve the above object, the present invention adopts the following technical solution:

[0011] A first aspect of the present invention provides a method for recommending a learning plan based on learning effectiveness prediction, comprising:

[0012] Obtain the basic data of the user to be predicted and the learning behavior data of a certain course, and perform preprocessing;

[0013] Based on the preprocessed data, the classification features are constructed and screened to obtain the feature vector of the user to be predicted;

[0014] Input the feature vector of the user to be predicted into the trained online course student learning effectiveness prediction model to predict the learning effectiveness of the user to be predicted on the course;

[0015] When the learning effect is not passed, a warning message is sent to the user to be predicted, and the user whose learning effect of the course is passed is used as a reference user. The learning efficiency and learning plan of the reference user are obtained, and the learning plan of the reference user with the highest learning efficiency is sent to the user to be predicted.

[0016] Among them, the student learning effectiveness prediction model includes a pre-training layer, a parallel processing layer and a meta-learning layer; the pre-training layer is used to generate the weight of the feature vector and add the weight to the feature vector; the parallel processing layer uses multiple parallel base learners to obtain output results based on the feature vector, and merges the output results of all base learners and inputs them into the meta-learning layer.

[0017] Furthermore, the classification feature screening method is:

[0018] Combine any two different features in the feature matrix of the training set as a feature pair, and calculate the Pearson correlation coefficient of each feature pair;

[0019] Based on the Pearson correlation coefficient of each feature pair, after deleting the features, the feature matrix is ​​input into the random forest model to learn the correlation coefficient ranking of all features, and all features are sorted from large to small according to the correlation coefficient.

[0020] Furthermore, after feature screening, the training set used for training the student learning effectiveness prediction model is expanded in data volume through the mean method to balance the ratio of positive and negative samples.

[0021] Furthermore, it also includes: using standard scores to normalize the classification features, and performing feature screening based on the normalized classification features.

[0022] Furthermore, the objective function of the trainer of the pre-training layer includes two parts: a loss function and a regularization term.

[0023] Furthermore, the constructed classification features include: average clicks, average extracurricular clicks, standard deviation of daily clicks, total registration time, average value of process-based assessment scores, and standard deviation of process-based assessment scores.

[0024] A second aspect of the present invention provides a learning plan recommendation system based on learning effectiveness prediction, which includes:

[0025] A data preprocessing module is configured to: obtain basic data of the user to be predicted and learning behavior data of a certain course, and perform preprocessing;

[0026] A feature enhancement module is configured to: construct and filter classification features based on the preprocessed data to obtain a feature vector of the user to be predicted;

[0027] The effectiveness evaluation module is configured to: input the feature vector of the user to be predicted into the trained online course student learning effectiveness prediction model to predict the learning effectiveness of the user to be predicted on the course;

[0028] The learning plan recommendation module is configured to: when the learning achievement fails the assessment, send a warning message to the user to be predicted, and use the user whose learning achievement of the course is predicted to pass the assessment as a reference user, obtain the learning efficiency and learning plan of the reference user, and send the learning plan of the reference user with the highest learning efficiency to the user to be predicted;

[0029] Among them, the student learning effectiveness prediction model includes a pre-training layer, a parallel processing layer and a meta-learning layer; the pre-training layer is used to generate the weight of the feature vector and add the weight to the feature vector; the parallel processing layer uses multiple parallel base learners to obtain output results based on the feature vector, and merges the output results of all base learners and inputs them into the meta-learning layer.

[0030] Furthermore, the constructed classification features include: average clicks, average extracurricular clicks, standard deviation of daily clicks, total registration time, average value of process-based assessment scores, and standard deviation of process-based assessment scores.

[0031] A third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the method for recommending a learning plan based on learning effectiveness prediction as described above.

[0032] The fourth aspect of the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps in the method for recommending a learning plan based on learning outcome prediction as described above are implemented.

[0033] Compared with the prior art, the present invention has the following beneficial effects:

[0034] The present invention provides a learning plan recommendation method based on learning achievement prediction, which adopts an integrated learning model to aggregate traditional machine learning models. It integrates multiple machine learning models, and the base classifier has the characteristics of high diversity, which improves the generalization ability of the model for different samples, can more accurately learn the behavior of students in online courses, and can effectively identify students with failure risks to remind students and recommend learning plans to students with failure risks. The plan is a learning plan for students at the same learning stage as the students, and will not be affected by environmental factors such as policies. It realizes the recommendation of the best and most suitable learning plan for the warned students, so as to improve the students' performance and enhance the students' learning ability.

[0035] The present invention provides a learning plan recommendation method based on learning effectiveness prediction, which uses the mean method to balance the number of positive and negative samples and improve the model's learning ability for unbalanced data sets.

[0036] The present invention provides a learning program recommendation method based on learning effectiveness prediction, which extracts and constructs features according to students' tracking data in a virtual learning environment, thereby improving the model's learning ability for key features of a data set.

[0037] The present invention provides a learning plan recommendation method based on learning effectiveness prediction, which adds a layer to the traditional stacking model, uses a base classifier to pre-train the data set, and inputs the prediction results into the model as new features, thereby improving the model's ability to reuse samples. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] The accompanying drawings in the specification, which constitute 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 descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0039] Figure 1 is a flow chart of a method for recommending a learning plan based on learning effectiveness prediction according to Embodiment 1 of the present invention;

[0040] Figure 2 is a flow chart of averaging positive and negative samples according to the first embodiment of the present invention;

[0041] Figure 3 is a structural diagram of a learning effectiveness prediction model according to the first embodiment of the present invention;

[0042] Figure 4 It is a system module diagram of the second embodiment of the present invention. DETAILED DESCRIPTION

[0043] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0044] It should be noted that the following detailed descriptions are all illustrative and 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.

[0045] 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.

[0046] Embodiment 1

[0047] This embodiment provides a learning plan recommendation method based on learning effectiveness prediction. Figure 1 As shown, the specific steps include:

[0048] Step 1: Obtain a training set, which includes the basic characteristics of several students, as well as the learning behavior characteristics and learning achievement labels of each student for a certain course.

[0049] Step 101: For the courses that students have completed and obtained accurate learning outcomes in the virtual learning environment, collect the classification data (including basic data and learning behavior data for a certain course) and learning outcome labels of each student.

[0050] Step 102: preprocess the classified data. In order to enhance the model feature learning effect, the student feature vector f i Data preprocessing is performed before input into the model. Data preprocessing includes missing value processing and format conversion. Specifically, it includes:

[0051] (1) Fill in missing values ​​in student classification data; if the attributes of click volume and grade are null values, fill them with 0; if the date attribute is null value, fill it with the date of learning effectiveness evaluation; due to the fact that students did not submit or participate in learning, there are problems such as missing data in the training set; such samples will affect the model's learning of features; fill in the missing values ​​by analyzing the feature definition;

[0052] (2) Convert the text data of students’ learning behaviors in the virtual learning environment into digital data, that is, convert varchar data into digital data, including replacing binary features with 0 and 1; convert enumeration data into natural numbers; for example, health status (poor: 0, good: 1), final evaluation results (failed: 0, passed: 1).

[0053] Among them, students' basic data include gender and health status; learning behavior data refers to students' learning behavior data in the virtual learning environment, including the number of times the course has been studied, course duration, grade weight, late homework rate, failure rate, pass rate, process-based assessment scores, daily clicks, registration time, and logout time, etc.

[0054] The learning outcomes are labeled as passed and failed.

[0055] Step 103: Based on the basic data of students, construct classification features: average clicks, average extracurricular clicks, standard deviation of daily clicks, total registration time, average of process-based assessment scores, and standard deviation of process-based assessment scores.

[0056] The construction method of the feature avgClick (average click volume) is as follows: Students learn a certain course in a virtual learning environment, and the frequency of learning the course is represented by the click volume. The daily average click volume is used to track the students' daily learning results. The daily average click volume is calculated as where sumclick i It represents the total number of daily clicks of student i during the learning period, and modulePresentationLength is the duration of the learning course.

[0057] The construction method of feature ac_avgclicks (average extracurricular clicks) is: The average extracurricular clicks reflect the students' efforts and activity in extracurricular learning. The higher the average extracurricular clicks, the higher the probability of passing the assessment of the course. The calculation method of the average extracurricular clicks is: date>modulePresentationLength OR date<0, where ac_avgclicks irepresents the average extracurricular clicks of student i, clicki represents the daily clicks of students in the virtual learning environment, ac_daysi represents the number of extracurricular learning days, and date represents the date when the clicks were generated.

[0058] The construction method of the feature sdevClicks (the standard deviation of daily clicks) is: The standard deviation of daily clicks reflects the degree of dispersion between students' daily clicks. The larger the standard deviation of daily clicks, the greater the difference in students' daily activity. The calculation method of the standard deviation of daily clicks in class is: where click id represents the number of clicks by student i on day d, avgClick i represents the average number of clicks by student i, and D is the total number of course days.

[0059] The construction method of the feature totalreg (total registration time) is: the total registration time reflects the students' enthusiasm for learning in the virtual learning environment to a certain extent. The calculation method of the total registration time is as follows:

[0060]

[0061] Among them, totalreg i Indicates the total registration time of student i, date_registration i represents the date when student i registered for the course, date_unregistration i It represents the date when student i drops out of the course, and modulePresentationLength is the duration of studying the course.

[0062] The construction method of the feature score_avg (average of process assessment scores) is: the average score is the average score of students in the process assessment of the course, reflecting the average performance of students during the course. The higher the average score, the higher the probability of passing the assessment of the course. The average score calculation method is Where score_avg i Represents the average score of student i, score ij It represents the score of the jth process-based assessment of student i in the virtual learning environment for this course, and n represents the number of assessments.

[0063] The construction method of the feature score_sdev (standard deviation of process assessment scores) is: the standard deviation of process assessment scores reflects the fluctuation of students' scores during the course. The larger the score_sdev, the greater the fluctuation of process assessment scores and the more unstable the students' learning status. The calculation method of the standard deviation of process assessment scores is where score_sdev i represents the standard deviation of student i's process assessment score, score ij represents the score of the jth process assessment of student i in the virtual learning environment for this course, n represents the number of assessments, score_avg i Represents the average score of student i.

[0064] After combining the basic data with the constructed classification features (average clicks, average extracurricular clicks, standard deviation of daily clicks, total registration time, average of process-based assessment scores, and standard deviation of process-based assessment scores), we can obtain the classification features (including basic features and learning behavior features for a certain course) and learning outcome labels of each student, that is, we get the training set {(t i ,s i ),y i |i=1,…,n}, where t i are all demographic characteristics (basic characteristics) of student i, s i are all the learning behavior characteristics of student i in the virtual learning environment, y i is the course assessment of student i (learning achievement label), and n is the number of students in the training set. Define the feature matrix F = [f1, f2, ..., f n ] T , where f i is the eigenvector of the i-th student, and f i =[t i1 ,t i2 ,…,t is ,s is+1 ,s is+2 ,……,s ik ], where s is the number of demographic features, ks is the number of learning behavior features, and t ia represents the ath demographic feature in the feature vector of the i-th student, s ia The ath learning behavior feature in the feature vector representing the i-th student. The time outside the class period in the virtual learning environment is called extracurricular.

[0065] Step 104: Normalization. Use the standard score (z-score) to normalize the student classification features, and perform feature screening based on the normalized classification features. Normalize the student classification features to convert data of different magnitudes into a uniform magnitude to ensure comparability between data. The calculation formula used for normalization is: or Among them, μ is the data mean, δ is the standard deviation of the data. The normalized feature matrix F * =[f1 * ,f2* ,……,f n * ] T , where f i * =[x i1 * ,x i2 * ,……,x ik * ], x ia * Represents the normalized value of the a-th feature in the i-th feature vector.

[0066] Step 105, feature screening. Any two different feature combinations in the feature matrix of the training set are regarded as a feature pair. Calculate the Pearson correlation coefficient of a feature pair as the correlation between the two features. A Pearson correlation coefficient greater than 0 indicates that the features are positively correlated; a Pearson correlation coefficient less than 0 indicates that the features are negatively correlated; a Pearson correlation coefficient equal to 0 indicates that there is no linear relationship between the features. According to the correlation coefficient between the features, delete one of the feature pairs with high correlation. By inputting the feature matrix into the random forest model, the correlation coefficient ranking is learned. Input the features into the model from large to small according to the correlation coefficient until the accuracy no longer improves.

[0067] Based on the Pearson correlation coefficient of each feature pair, after deleting the features, the feature matrix is ​​input into the random forest model to learn the correlation coefficient ranking of all features, and all features are sorted from large to small according to the correlation coefficient. Specifically: randomly delete a feature in the feature pair with correlation of 1 and -1; use the random forest model to train the learning behavior characteristics of the students' virtual learning environment, generate the correlation coefficient of each feature with the evaluation results of the students' learning effectiveness, sort them from large to small according to the correlation coefficient, and input the features into the random forest model for learning until the accuracy rate no longer improves, and obtain the filtered feature vector f i * =[x i1 * ,x i2 * ,……,x im * ], where m represents the number of features after screening.

[0068] Step 106: The party with the smaller number makes the number of positive samples and negative samples in the training set equal. Figure 2 As shown in the figure, the training set after feature screening is used as input, 5 negative samples are randomly selected, and the average of each column feature is calculated to obtain a new sample. Finally, the new negative sample set is merged into the original sample set.

[0069] Extract part of the data set to form a sample population, where each piece of data is a sample. The label refers to the sample classification corresponding to the data, usually 0 or 1. The expected positive result is called a positive sample (represented by "1"), and the opposite result is called a negative sample (represented by "0"). Select the set with less data in the positive sample or negative sample, expand its data volume through the mean method, and balance the ratio of positive and negative samples.

[0070] Define the feature matrix of the sample with label 0 as F0=[f01,f02,…,f0 g ] T , where g is the number of samples with label 0, f0 i is the feature vector of the i-th sample with label 0, f0 i =[x i1 ,x i2 ,…,x im ], x ia Represents the ath feature in the i-th sample with label 0. Randomly extract 5 samples with replacement from the sample population with label 0 or 1, and calculate the mean of the feature values ​​of each column of these five samples. New Sample And add it to the training set; determine whether the number of samples with labels 0 and 1 in the training set is the same. If not, continue to amplify the samples until the number of samples with labels 0 and 1 is the same. The number of samples before amplification is n, and the number of samples after amplification is u, that is, F * =[f1 * ,f2 * ,……,f u * ] T , f i * =[x i1 * ,x i2 * ,……,x im * ], x ia * Represents the ath feature in the i-th feature vector.

[0071] Step 2: Use the training set to train the online course student learning effectiveness prediction model. Obtain the classification data of the user to be predicted (including basic data and learning behavior data for a certain course), perform preprocessing and normalization; based on the normalized data, construct and filter out the classification features (directly select the features filtered out based on the training set in step 106), and obtain the feature vector f of the user to be predicted. i * =[x i1* ,x i2 * ,……,x im * ], and input the trained online course student learning effectiveness prediction model to predict the learning effectiveness of the user to be predicted for the course. When the learning effectiveness fails the assessment, a warning message is sent to the user to be predicted (student terminal), and the learning effectiveness of each student is sent to the teacher terminal for display.

[0072] like Figure 3 As shown in the figure, the first layer of the student learning effectiveness prediction model is the pre-training layer, the middle layer is the parallel processing layer, and the third layer is the meta-learning layer. The pre-training layer is used to generate the weight of the feature vector and add the weight to the feature vector; the parallel processing layer uses multiple parallel base learners to obtain the output results based on the feature vector, and merges the output results of all base learners and inputs them into the meta-learning layer.

[0073] Step 201: Pre-training. Pre-training includes calling the XGBoost trainer and setting the cost matrix.

[0074] The objective function of the pre-trained layer trainer consists of two parts: the loss function and the regularization term. The objective function of the XGBoost trainer is in is the regularization term, f k is the function of the kth tree, K is the number of trees, and for a function f of a tree, its regularization term is Measures the gap between the true label and the predicted label, l is the loss function, y i is the true label, is the predicted label. In the regularization term, T represents the number of leaf nodes, represents the score of leaf nodes, γ controls the number of leaf nodes, and λ controls the score of leaf nodes to prevent overfitting.

[0075] The setting of the cost matrix includes constructing the cost matrix and calculating the error probability.

[0076] Let the error cost of misclassifying a sample with a classification result of passing the assessment (or failing the assessment) as failing the assessment (or passing the assessment) be recorded as e 10 (or e 01 ), there are two types of classification, forming a 2*2 cost matrix: Classification error 10 (or e 01 ) are n 10 (or n 01 ). The cost-sensitive error rate is calculated as: Where II(*) is the indicator function, which is 1 if the training is correct and 0 otherwise. *is the data set, F *+ is the positive sample data set, F *- is a negative sample dataset. Refers to the predicted value of the characteristics of the i-th student in the model.

[0077] Confusion matrix is ​​established to visualize the performance of the algorithm. The probability that a sample with the classification result of passing the assessment (or failing the assessment) is misclassified as failing the assessment (or passing the assessment) is calculated as P 10 (or P 01 ). Specifically, establish the confusion matrix: The probability (error probability) of misclassifying a sample with a classification result of passing the assessment (or failing the assessment) as failing the assessment (or passing the assessment) can be calculated based on the confusion matrix:

[0078] Initialize the weight of each feature vector D = (w1, w2, ..., w u ), weighted equally, both After XGBoost training, samples with incorrect classification are screened out and their weights are amplified. The cost-sensitive error rate is calculated for each training. When the cost-sensitive error rate is the lowest, the weight at this time is the final weight. The classification result and the weight are combined as a new feature and added to the original feature vector f i * =[x i1 * ,x i2 * ,……,x im * ] to form f i ′=w i ×f i * =[w i × i1 * ,w i × i2 * ,……,w i × im * ], the characteristic matrix is ​​F'=[f1',f2',……,f u '] T .

[0079] Step 202: Parallel processing. Four base learners are selected in the parallel processing layer, and the base learners must meet the following characteristics: strong performance, small correlation, and not too large a performance gap. The four base learners selected in this layer are: Extreme Gradient Boosting (XGBoost), CatBoost (Category Boosting), Binary Random Forest (BRF) and Extremely randomized trees (ET). The four base learners are distributed side by side and receive the feature matrix F' at the same time. Among them, ET uses random features and random thresholds to divide nodes based on the decision tree, resulting in greater and more random differences in the decision tree results. BRF extracts feature information through random forests, that is, t trees are taken, each tree has f leaves, and each leaf corresponds to a one-bit binary number. The training error of the kth learner is recorded as where f i ' is an eigenvector in F', u is the number of eigenvectors in F', C(f i ') is the learner training f i 'After the output, y is f i ' label. When C(f i ′)≠y, [|C(f i ′)≠y|]=1. The error probability P of the kth base learner is calculated as in step 201. k The training result of the kth base learner is recorded as R k =[C k (f1′),C k (f2′),…,C k (f u ′)] T Assign weights to the training results of the four base learners. The weight assignment formula is: And construct a new feature matrix F1 = [weight1×R1, weight2×R2, weight3×R3, weight4×R4].

[0080] Step 203: Meta-learning layer. The meta-learning layer uses CatBoost as a meta-learner and obtains an output result of 0 or 1 for the validation set by training the feature matrix F1.

[0081] Step 3: For this course, at a certain point in time, after the learning outcomes of all users are predicted through step 2, when the learning outcomes of the user to be predicted are not passed, the users whose learning outcomes of this course are predicted to be passed are used as reference users, and the learning efficiency and learning plan of the reference users are obtained, and the learning plan of the reference user with the highest learning efficiency is sent to the user to be predicted.

[0082] The learning efficiency is the ratio of the average value of the process assessment scores of the reference users for the course to the daily click volume. The method for constructing a learning plan can adopt the method disclosed in the patent application number 201911034859.3, a learning plan recommendation method and device.

[0083] This embodiment uses an integrated learning model to aggregate traditional machine learning models. It integrates multiple machine learning models. The base classifier has the characteristics of high diversity, which improves the generalization ability of the model for different samples. It can more accurately learn the behavior of students in online courses and can effectively identify students at risk of failure to remind students and recommend learning plans for students at risk of failure. The plan is a learning plan for students at the same learning stage as the students and will not be affected by environmental factors such as policies. It realizes the recommendation of the best and most suitable learning plan for the warned students, so as to improve students' performance and enhance their learning ability.

[0084] Embodiment 2

[0085] This embodiment provides a learning plan recommendation system based on learning effectiveness prediction. Figure 4 As shown, it specifically includes the following modules:

[0086] A data preprocessing module is configured to: obtain basic data of the user to be predicted and learning behavior data of a certain course, and perform preprocessing;

[0087] A feature enhancement module is configured to: construct and filter classification features based on the preprocessed data to obtain a feature vector of the user to be predicted;

[0088] The effectiveness evaluation module is configured to: input the feature vector of the user to be predicted into the trained online course student learning effectiveness prediction model to predict the learning effectiveness of the user to be predicted on the course;

[0089] The learning plan recommendation module is configured to: when the learning achievement is not passed, send a warning message to the terminal of the user to be predicted, and use the user whose learning achievement of the course is passed as a reference user, obtain the learning efficiency and learning plan of the reference user, and send the learning plan of the reference user with the highest learning efficiency to the terminal of the user to be predicted;

[0090] The prediction display module is configured to send each student's learning achievement to the teacher's terminal for display.

[0091] Among them, the student learning effectiveness prediction model includes a pre-training layer, a parallel processing layer and a meta-learning layer; the pre-training layer is used to generate the weight of the feature vector and add the weight to the feature vector; the parallel processing layer uses multiple parallel base learners to obtain output results based on the feature vector, and merges the output results of all base learners and inputs them into the meta-learning layer.

[0092] It should be noted here that each module in this embodiment corresponds to each step in Example 1 one by one, and the specific implementation process is the same, which will not be repeated here.

[0093] Embodiment 3

[0094] This embodiment provides a computer-readable storage medium having a computer program stored thereon. When the program is executed by a processor, the steps in the method for recommending a learning plan based on learning effectiveness prediction as described in the first embodiment above are implemented.

[0095] Embodiment 4

[0096] This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps in a method for recommending a learning plan based on learning effectiveness prediction as described in the first embodiment above are implemented.

[0097] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage and optical storage, etc.) containing computer-usable program code.

[0098] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0099] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0100] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0101] A person skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the embodiments of the above-mentioned methods. The storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc.

[0102] 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 learning effectiveness prediction, characterized in that: include: Obtain training sets, collect classification data and learning performance labels for each student, and preprocess the classification data for each student; the classification data includes basic data and learning behavior data for a certain course; the learning performance labels are passed and failed; Based on the students' learning behavior data, classification features are constructed. Specifically, the constructed classification features include: average click volume, average extracurricular click volume, standard deviation of daily click volume, total registration time, average value of process-based assessment scores, and standard deviation of process-based assessment scores; after combining the basic data with the constructed classification features, the combined classification features and learning achievement labels of each student are obtained; that is, the training set is obtained. ,in t i For Students i All demographic characteristics of s i For Students i All learning behavior characteristics in the virtual learning environment, y i For Students i Learning outcomes label, n is the number of students in the training set; demographic characteristics refer to the basic characteristics extracted based on basic data; Define the feature matrix ,in f i is the eigenvector of the i-th student, and ,in, s is the number of demographic characteristics, -s is the number of learning behavior features, t ia Representative i The feature vector of the student a demographic characteristics, s ia Representative i The feature vector of the student a Learning behavior characteristics; The combined classification features are normalized using the standard score, and feature screening is performed based on the normalized combined classification features; the normalization calculation formula is: or ,in, μ is the data mean, δ is the standard deviation of the data; the normalized feature matrix ,in , x ia * Representative i The first feature vector a The normalized value of a feature; The method of classification feature screening is as follows: any two different feature combinations in the feature matrix of the training set are taken as a feature pair, and the Pearson correlation coefficient of each feature pair is calculated; based on the Pearson correlation coefficient of each feature pair, the feature is deleted, and the feature matrix is ​​input into the random forest model to learn the correlation coefficient ranking of all features, and all features are sorted from large to small according to the correlation coefficient; after feature screening, the training set used for the student learning achievement prediction model is expanded by the mean method to balance the ratio of positive and negative samples; The training set is used to train the online course student learning effectiveness prediction model; the classification data of the user to be predicted is obtained for preprocessing and normalization; based on the normalized data, the classification features are constructed and screened to obtain the feature vector of the user to be predicted. , and input the trained online course student learning effectiveness prediction model to predict the learning effectiveness of the user to be predicted for the course. When the learning effectiveness is not passed, send a warning message to the user to be predicted, and use the user whose learning effectiveness of the course is passed as a reference user to obtain the learning efficiency and learning plan of the reference user, and send the learning plan of the reference user with the highest learning efficiency to the user to be predicted; Among them, the student learning effectiveness prediction model includes a pre-training layer, a parallel processing layer and a meta-learning layer; the pre-training layer is used to generate the weight of the feature vector and add the weight to the feature vector; the parallel processing layer uses multiple parallel base learners to obtain output results based on the feature vector, and merges the output results of all base learners and inputs them into the meta-learning layer; among them, the base learners are: extreme gradient decision tree, CatBoost, binary random forest and extreme random tree.

2. A method for recommending a learning plan based on learning effectiveness prediction as claimed in claim 1, characterized in that: The objective function of the trainer of the pre-training layer includes two parts: a loss function and a regularization term.

3. A learning plan recommendation system based on learning effectiveness prediction, using a learning plan recommendation method based on learning effectiveness prediction as described in any one of claims 1-2, characterized in that: include: A data preprocessing module is configured to: obtain basic data of the user to be predicted and learning behavior data of a certain course, and perform preprocessing; A feature enhancement module is configured to: construct and filter classification features based on the preprocessed data to obtain a feature vector of the user to be predicted; The effectiveness evaluation module is configured to: input the feature vector of the user to be predicted into the trained online course student learning effectiveness prediction model to predict the learning effectiveness of the user to be predicted on the course; The learning plan recommendation module is configured to: when the learning achievement fails the assessment, send a warning message to the user to be predicted, and use the user whose learning achievement of the course is predicted to pass the assessment as a reference user, obtain the learning efficiency and learning plan of the reference user, and send the learning plan of the reference user with the highest learning efficiency to the user to be predicted; Among them, the student learning effectiveness prediction model includes a pre-training layer, a parallel processing layer and a meta-learning layer; the pre-training layer is used to generate the weight of the feature vector and add the weight to the feature vector; the parallel processing layer uses multiple parallel base learners to obtain output results based on the feature vector, and merges the output results of all base learners and inputs them into the meta-learning layer.

4. A learning plan recommendation system based on learning effectiveness prediction as claimed in claim 3, characterized in that: The constructed classification features include: average clicks, average extracurricular clicks, standard deviation of daily clicks, total registration time, average value of process-based assessment scores, and standard deviation of process-based assessment scores.

5. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps in a method for recommending a learning plan based on learning effectiveness prediction as described in any one of claims 1-2 are implemented.

6. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps in the method for recommending a learning plan based on learning effectiveness prediction as described in any one of claims 1-2 are implemented.

Images