A learning behavior data mining method, system, storage medium, and processing terminal

Abstract

The present invention belongs to the field of artificial intelligence technology, and discloses a learning behavior data mining method, system, storage medium, and processing terminal. The learning behavior data mining method includes: using an Elearn system to collect student learning behavior data in real time, performing a series of operations on PDF teaching materials, and uploading the operation information to a server for storage; serializing the collected student learning behavior data, using a long-short attention network to predict the characteristics of the student learning behavior data, and obtaining the predicted grade of the student's grades; dividing the students into three categories of excellent, good, and qualified according to their academic performance, and using a method based on vector quantization learning to supervise and cluster the learning behavior data of students with excellent and good grades, and mining the students' excellent learning behaviors. The Elearn system of the present invention is based on the design mode of Internet + artificial intelligence + education, which gets rid of the dependence of teaching on paper textbooks, and can access the system anytime and anywhere to carry out teaching and learning activities.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence technology, and in particular relates to a learning behavior data mining method, system, storage medium, and processing terminal. Background Art

[0002] At present, with the advent of the big data era and the development of artificial intelligence, combining educational technology with deep learning methods and exploring the development trends of educational technology changes will be able to effectively guide the continuous improvement of teaching methods and better serve students' growth needs.

[0003] At present, the development of deep learning technology is in the ascendant, and its application in education, medical and social sciences has become an inevitable trend. There is actually a lot of potential in applying artificial intelligence methods to the field of education. In addition to assisting teaching and learning activities, it can also analyze educational big data to condense learning behaviors and methods, and guide teachers and students to better complete teaching tasks. However, the application of artificial intelligence in the field of education is still in its initial stage. There are some robots or learning system products on the market, but their intelligence level is not high. There is no technical solution that can truly and objectively collect a large amount of student learning behavior data, analyze student learning behavior on this basis, and mine excellent learning behaviors. Therefore, there is an urgent need for such a technical solution that combines artificial intelligence methods with education.

[0004] Through the above analysis, the problems and defects of the existing technology are as follows: There is no report on the perfect technical solution for applying artificial intelligence methods to the field of education in the existing technology. On the one hand, when using questionnaires and videos to obtain student learning data, the subjective factors of students have a great influence, and it is impossible to collect real and objective learning behavior data, resulting in the data not being able to well reflect the relationship between students' learning behavior and academic performance; on the other hand, the limitations of the tools used lead to a small number of samples collected, or from specific samples (consisting of a small number of teachers and students), which cannot provide sufficient data support for learning behavior analysis. Therefore, there is currently a lack of tools that can effectively collect large-scale, real and objective student learning behavior data, which also makes it difficult to use deep learning technology to analyze student learning behavior data in the future.

[0005] The difficulty of solving the above problems and defects is as follows: First, it is necessary to develop a system that can effectively collect large-scale real and objective student learning behavior data. This requires that the newly designed system can provide students with electronic learning materials during the process of pre-class preparation, in-class learning and after-class review, while being able to accurately record students' learning behaviors in real time and store the data; in addition, in order to accurately analyze students' learning behavior data and make high-precision predictions of students' academic performance levels based on the collected data, it is necessary to design a model that is better than the currently popular deep learning model for processing time series data. At the same time, it is necessary to design an effective solution to tap into students' excellent learning behaviors in order to promote and guide other students' learning activities.

[0006] The significance of solving the above problems and defects is: developing a new system based on the design mode of Internet + artificial intelligence + education can make teaching free from the dependence on paper textbooks. Teachers and students can access the system anytime and anywhere in the Internet environment to carry out teaching and learning activities. In the process of students using the system to learn, the system can accurately record students' notes, page turning and other behaviors in real time and store them, providing a large amount of objective and real data for the subsequent analysis of students' learning behavior data using deep learning technology. In addition, new deep learning methods are developed to efficiently process time series data and make high-precision predictions. On this basis, based on students' learning behavior data, students' learning effects can be evaluated and students' learning situation can be closely monitored. In addition, the excellent learning behaviors of students obtained by data mining methods can be used to guide the learning process of more students and help them develop good learning habits and learning methods. It has very important promotion value among graduate students, undergraduates and even primary and secondary school students. Summary of the invention

[0007] In view of the problems existing in the prior art, the present invention provides a learning behavior data mining method, system, storage medium, and processing terminal, and more particularly, relates to a learning behavior data mining method, system, storage medium, and processing terminal based on an Elearn system and deep learning technology.

[0008] The present invention is implemented as follows: a learning behavior data mining method, the learning behavior data mining method comprising the following steps:

[0009] Step 1: Use the Elearn system to collect students’ learning behavior data in real time, perform column operations on PDF textbooks, and upload the operation information to the server for storage;

[0010] Step 2: Serialize the collected student learning behavior data, use the long-term and short-term attention network to predict the characteristics of the student learning behavior data, and obtain the predicted grade of the student's performance;

[0011] Step three, divide the students into three categories: excellent, good and qualified according to their academic performance, use the vector quantization learning-based method to supervise the learning behavior data of students with excellent and good grades and explore the students' excellent learning behaviors.

[0012] Furthermore, in step one, the operation information includes: student user name, student ID, highlight operation, underline operation, delete line operation, marked content, current reading page number and current reading time.

[0013] Furthermore, in step 2, the collected student learning behavior data is serialized, including:

[0014] In the process of preprocessing students’ learning behavior data, all data information is divided into different parts according to time stages, and in each part, the data of different students are counted separately;

[0015] Among them, the highlight, underline and strikethrough operations are counted according to the number of operations, the marked content is counted by adding up the number of characters in the marked content, the different numbers of read pages are counted, and the reading time of the day is counted by the difference between the latest operation time and the earliest operation time of the user in a day.

[0016] Furthermore, after the collected student learning behavior data are serialized, each user has a 6-dimensional data vector in each time segment, where each dimension corresponds to the statistics of 6 types of operation information, and each dimension is normalized respectively; the normalization processing includes: selecting the 6 operation values ​​with the largest value in each time segment, dividing all the data in this time segment by the maximum value of the current dimension, and obtaining the final normalized result.

[0017] Furthermore, in step 2, the use of the long-short time attention network to predict the characteristics of the student learning behavior data includes:

[0018] (1) When making predictions based on the long- and short-term attention network, the final course scores are divided into four levels: excellent, good, pass, and fail, and converted into one-hot vectors as the true labels;

[0019] (2) The data of each student in each time period is regarded as a sequence data. The number of time periods is the number of time steps input into the long and short time attention network. The dimension 6 of the data is the input dimension size of the long and short time attention network at each time step.

[0020] (3) The student data is divided into a training set and a test set. During the training process, after processing the time series data of the current student, the long and short time attention network generates a 4-dimensional vector and performs cross entropy loss with the label;

[0021] (4) The network is trained through back propagation, and the grade represented by the maximum dimension of the vector median is the grade of the current student predicted by the network; during the test, the predicted grade is compared with the actual label grade to obtain the error rate.

[0022] Furthermore, in step 2, the predicted grades of student performance include: excellent, good, pass and fail.

[0023] Further, in step 3, the supervised clustering of the learning behavior data of students with excellent and good grades using a method based on vector quantization learning includes:

[0024] (1) At the beginning of clustering, two samples are randomly selected from each class as the prototype points of the cluster;

[0025] (2) Randomly select a sample from the sample set, calculate its distance with each vector in the prototype vector group, and select the cluster where the prototype vector with the smallest distance belongs as its division result, and then compare it with the true class label;

[0026] (3) If the division result is correct, the corresponding prototype vector is closer to the sample; if the division result is incorrect, the corresponding prototype vector is farther away from the sample;

[0027] (4) Iterate until the algorithm converges or reaches the maximum step, and output a prototype vector group that represents the characteristics of good learning behavior data.

[0028] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor executes the following steps:

[0029] Use the Elearn system to collect students' learning behavior data in real time, perform column operations on PDF textbooks, and upload the operation information to the server for storage;

[0030] The collected student learning behavior data is serialized and the characteristics of the student learning behavior data are predicted using the long-short time attention network to obtain the predicted grade of the student's performance;

[0031] According to their academic performance, students are divided into three categories: excellent, good and qualified. A vector quantization learning-based method is used to perform supervised clustering on the learning behavior data of students with excellent and good grades to explore students' excellent learning behaviors.

[0032] Another object of the present invention is to provide an information data processing terminal, which is used to implement the learning behavior data mining method.

[0033] Another object of the present invention is to provide a learning behavior data mining system for implementing the learning behavior data mining method, the learning behavior data mining system comprising:

[0034] The Elearn system is used to collect students’ learning behavior data, perform a series of operations on PDF textbooks, and upload the operation records to the server for storage;

[0035] The grade prediction module is used to serialize the collected student learning behavior data and use the long-short time attention network to predict the student grade;

[0036] The learning behavior mining module is used to classify students into three categories: excellent, good, and qualified according to their academic performance. It uses a vector quantization learning-based method to perform supervised clustering on the learning behavior data of students with excellent and good grades to mine students' excellent learning behaviors.

[0037] Combined with all the above technical solutions, the advantages and positive effects of the present invention are as follows: the learning behavior data mining method provided by the present invention, the developed Elearn system is based on the design mode of Internet + artificial intelligence + education, which gets rid of the dependence of teaching on paper textbooks. Teachers and students can access the system anytime and anywhere in the Internet environment to carry out teaching and learning activities. In the process of students using the Elearn system to learn, the system can accurately record students' behaviors such as highlighting, turning pages, and emphasizing in real time. In addition, the deep learning method developed by the present invention can process time series data and make high-precision predictions with long and short time attention networks. Based on the students' learning behavior data, their learning effects can be evaluated and their learning situation can be closely monitored. In addition, the excellent learning behavior of students obtained by the data mining method can be used to guide the learning process of more students, help students develop better learning habits and learning methods, and have very important promotion value in graduate students, undergraduates and even primary and secondary school courses. The present invention also has the following technical effects:

[0038] (1) Using the Elearn system to collect student learning behavior data in real time. The Elearn system developed by the present invention is based on the design mode of Internet + artificial intelligence + education, which gets rid of the dependence of teaching on paper textbooks. Teachers and students can access the system anytime and anywhere in the Internet environment to carry out teaching and learning activities. In the process of students using the Elearn system to learn, the system can accurately record students' behaviors such as highlighting, emphasizing, turning pages, reading time, etc. in real time.

[0039] (2) Prediction based on long-short time attention network. The long-short time attention network studied in this invention adds an attention gate on the basis of the long short-term memory network (LSTM), which can better process time series data and make high-precision predictions. Based on the students' learning behavior data, their learning effects can be evaluated and their learning situation can be closely monitored.

[0040] (3) Based on the learning vector quantization method, the learning behavior data of students who have achieved excellent and good grades are studied by vector quantization method. The excellent learning behavior of students can be used to guide the learning process of more students and help them develop good learning habits and learning methods. It has very important promotion value among graduate students, undergraduates and even primary and secondary school students. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments of the present invention. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0042] Figure 1 It is a design block diagram of the learning behavior data mining method provided by an embodiment of the present invention.

[0043] Figure 2 It is a schematic diagram of a user login interface, a user registration interface and a user password modification interface provided by an embodiment of the present invention.

[0044] Figure 3 It is a schematic diagram of a user operation interface provided by an embodiment of the present invention.

[0045] Figure 4 It is a schematic diagram of a user management interface provided by an embodiment of the present invention.

[0046] Figure 5 It is a schematic diagram of a user operation record interface provided by an embodiment of the present invention.

[0047] Figure 6 It is a structural diagram of a long- and short-time attention network unit used for predicting student grades provided by an embodiment of the present invention.

[0048] Figure 7 It is a diagram showing the effect of supervised clustering of learning behavior data using a learning vector quantization method provided in an embodiment of the present invention.

[0049] Figure 8 It is a flow chart of the learning behavior data mining method provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0051] In view of the problems existing in the prior art, the present invention provides a learning behavior data mining method, system, storage medium, and processing terminal. The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0052] like Figure 8 As shown, the learning behavior data mining method provided by the embodiment of the present invention includes the following steps:

[0053] S101, using the Elearn system to collect students’ learning behavior data in real time, performing column operations on the PDF textbooks, and uploading the operation information to the server for storage;

[0054] S102, serializing the collected student learning behavior data, using the long-short time attention network to predict the characteristics of the student learning behavior data, and obtaining the predicted grade of the student's performance;

[0055] S103, divide the students into three categories: excellent, good and qualified according to their academic performance, use a vector quantization learning-based method to perform supervised clustering on the learning behavior data of students with excellent and good grades, and explore the students' excellent learning behaviors.

[0056] Ordinary technicians in the industry can also use other steps to implement the learning behavior data mining method provided by the present invention. Figure 1 The learning behavior data mining method provided by the present invention is only a specific embodiment.

[0057] The technical solution of the present invention is described in detail below in conjunction with the accompanying drawings.

[0058] Figure 1 The design block diagram of the present invention can be divided into three parts: using the Elearn system to collect student learning behavior data in real time, where the student learning behavior data includes: highlight, underline, delete line, marked content, page number, and reading time; after preprocessing the data, predicting the characteristics of student learning behavior data based on the long and short time attention network, and obtaining student prediction levels including: excellent, good, pass, and fail; based on the learning vector quantization method, conducting an investigation and clustering analysis on the learning behavior data to obtain the learning behavior in the optimal cluster.

[0059] Figure 2 It is the user login interface, user registration interface and user password change interface.

[0060] Figure 3It is the user interface. After logging in, students can perform a series of operations on the textbook during reading, such as "previous page", "next page", "zoom in", "zoom out", "search", "tag", etc. They can also "underline", "add mark", "highlight text" on the content in the textbook. The figure shows some of the operations that students perform on the PDF textbook.

[0061] Figure 4 This is the user management interface where administrators can view and manage user account information, including deleting, adding, and changing user information. When a user forgets their password or the information is incorrect, the administrator can help them modify it.

[0062] Figure 5 It is the user operation record interface. All operations performed by users will be recorded in the database and transmitted back to the back-end server via the Internet. In addition, the administrator can filter, view and manage the operation records.

[0063] Figure 6 This is a structural diagram of the long-short time attention network unit used to predict the grade of students. The long-short time attention network adds an attention gate on the basis of the long-short time memory network. The attention gate acts on the forget gate and the input gate. To some extent, it determines which information should be ignored and which information should be paid attention to. The network achieves a high accuracy in the process of predicting the grade of students.

[0064] Figure 7 This is a diagram showing the effect of supervised clustering of learning behavior data using the learning vector quantization method. Specifically, students are divided into three categories: excellent, good, and qualified according to their final grades in the Elearn system. The learning behavior data of excellent and good students are clustered using the LVQ algorithm and clustered into four small categories (two excellent and two good). Each student's learning behavior data and the four learned prototype vectors have 6-dimensional features and a sample label representing excellent or good. The four prototype vectors define a region associated with them. The distance between each sample in the region and this prototype vector is no greater than the distance to other prototype vectors. In order to demonstrate the clustering effect, two-dimensional features of the sample data and the obtained prototype vector are plotted. Figure 7 .

[0065] The technical solution of the present invention is further described below in conjunction with embodiments.

[0066] Example 1

[0067] With the advent of the big data era and the development of artificial intelligence, combining educational technology with deep learning methods and exploring the development trend of educational technology changes will be able to effectively guide the continuous improvement of teaching methods and better serve the growth needs of students. The Elearn system developed by the present invention is based on the design mode of Internet + artificial intelligence + education, which gets rid of the dependence of teaching on paper textbooks. Teachers and students can access the system anytime and anywhere in the Internet environment to carry out teaching and learning activities. In the process of students using the Elearn system to learn, the system can accurately record students' behaviors such as highlighting, turning pages, and emphasizing in real time. In addition, the deep learning method developed by the present invention pays attention to the network for a long and short time, can process time series data and make high-precision predictions. Based on the students' learning behavior data, their learning effects can be evaluated and the students' learning situation can be closely monitored. In addition, the excellent learning behavior of students obtained by the data mining method can be used to guide the learning process of more students, help students develop better learning habits and learning methods, and have very important promotion value in graduate students, undergraduates and even primary and secondary school students.

[0068] Python Web development framework Django, deep learning framework PyTorch and PyCharm 2020.1.2. The Elearn system used in the present invention is developed based on the Python Web framework Django and PyCharm 2020.1.2, and the operating environment is Ubuntu 18.04.4.

[0069] The long short-term memory network is built based on the deep learning framework PyTorch, and the operating environment is Ubuntu 18.04.4. The learning vector quantization method used here to cluster student learning behavior data is developed based on PyCharm 2020.1.2, and the operating environment is Windows10.

[0070] The present invention consists of three parts:

[0071] First, the Elearn system is used to collect students' learning behavior data in real time. The Elearn system developed by the present invention is based on the design mode of Internet + artificial intelligence + education, which gets rid of the dependence of teaching on paper textbooks. Teachers and students can access the system anytime and anywhere in the Internet environment to carry out teaching and learning activities. In the process of students using the Elearn system to learn, the system can record students' behaviors such as highlighting, emphasizing, turning pages, reading time, etc. in real time and accurately.

[0072] Second, prediction is made based on the long and short-term attention network. The long and short-term attention network studied in this invention adds an attention gate based on the long short-term memory network (LSTM), which can better process time series data and make high-precision predictions. Based on the students' learning behavior data, their learning effects can be evaluated and their learning situation can be closely monitored.

[0073] Third, based on the learning vector quantization method, the learning behavior data of students who obtained excellent and good grades were studied by vector quantization method. The students' excellent learning behavior was obtained, which can be used to guide the learning process of more students and help them develop good learning habits and learning methods. It has very important promotion value among graduate students, undergraduates and even primary and secondary school students.

[0074] The learning behavior data mining system based on the Elearn system and deep learning technology provided by the embodiment of the present invention includes the Elearn system, a grade prediction module, and a learning behavior mining module; wherein,

[0075] The Elearn system is used to collect student learning behavior data; specifically, it includes: using the Elearn system to perform a series of operations on the PDF teaching materials, and uploading the operation records to the server for storage.

[0076] The grade prediction module is used to predict the student grade based on the collected student learning behavior data; specifically includes: serializing the collected student learning behavior data and using the long-short time attention network to predict the student grade.

[0077] The learning behavior mining module is used to mine students' excellent learning behaviors; specifically, it includes: dividing students into three categories: excellent, good, and qualified according to their academic performance, using a vector quantization learning-based method to perform supervised clustering on the learning behavior data of students with excellent and good grades, and mining students' excellent learning behaviors.

[0078] Example 2

[0079] Before using the Elearn system, students need to register an account based on their information (the registration page is as follows: Figure 2 After successful registration, the student's information will be recorded in the database (the backend display interface is as shown in Figure 4 The administrator can add, modify, and delete the information. In addition, students can modify their passwords (the password modification interface is shown in the figure below). Figure 2 If the password is forgotten, the administrator can help you change it.

[0080] When using the Elearn system, students need to log in to their accounts (the login interface is as follows Figure 2When reading a PDF textbook, you can turn pages, zoom in and out, search, and highlight, delete, or underline the content of the article. Figure 3 shown.

[0081] The operation information performed by students will be recorded in the database, including student username, student ID, highlight operation, underline operation, delete operation, marked content, current reading page number, current reading time, and the background display interface is as follows: Figure 5 shown.

[0082] In the process of preprocessing students' learning behavior data, all data information is divided into different parts according to time stages. In each part, the data of different students are counted separately. The highlight, underline, and delete operations are counted according to the number of operations, the marked content is counted by adding up the number of characters in the marked content, the different numbers of read pages are counted, and the reading time of the day is counted by the difference between the latest operation time and the earliest operation time of the user in a day.

[0083] After the above processing, in each time segment, each user has a 6-dimensional data vector, where each dimension corresponds to the statistics of 6 types of operation information. In order to input into the deep network for better training and more accurate prediction, each dimension is normalized. Specifically, the 6 operation values ​​with the largest value in each time segment are selected, and all the data in this time segment are divided by the maximum value of the current dimension to obtain the final normalized result.

[0084] In the case of length-based attention networks (network structures such as Figure 6 When making predictions (as shown in Figure 2), the final course grades are divided into 4 levels, namely excellent, good, pass, and fail, and converted into one-hot vectors as true labels. Based on the above data processing process, the data of each student in each time period is regarded as a sequence data. The number of time periods is the number of time steps input to the long and short time attention network. The dimension 6 of the data is the input dimension size of the long and short time attention network at each time step.

[0085] The student data is divided into a training set and a test set. During the training process, after processing the time series data of the current student, the long and short attention network generates a 4-dimensional vector, performs cross entropy loss with the label, and performs network training through back propagation. The grade represented by the maximum dimension of the vector median is the grade of the current student predicted by the network. During the test process, the predicted grade is compared with the actual label grade to obtain the network prediction error rate.

[0086] When investigating based on the learning vector quantization method, in order to discover excellent learning behaviors, the learning behavior data with excellent and good grades are investigated in detail. As mentioned above, after data processing, we obtained a sample set of student learning behavior data containing excellent and good grades, in which each sample data has six-dimensional features and a sample label (i.e., supervision information, excellent is 1, and good is 0). The goal of learning is to generate a set of prototype vectors and corresponding excellent and good labels that can respectively represent the characteristics of excellent and good learning data. Each 6-dimensional prototype vector represents the characteristic information of an excellent or good cluster, that is, it defines a region related to it, and the distance between each sample in the region and this prototype vector is not greater than the distance with other prototype vectors. At the beginning of clustering, two samples are randomly selected from each class as the prototype points of the cluster. Then, a sample is randomly selected from the sample set, and its distance with each vector in the prototype vector group is calculated, and the cluster with the smallest distance to the prototype vector is selected as its partition result, and then compared with the true class label. If the partition result is correct, the corresponding prototype vector is closer to this sample; if the partition result is incorrect, the corresponding prototype vector is farther away from this sample. This process is iterated until the algorithm converges or executes to the maximum step, outputting a prototype vector group representing the characteristics of good and excellent learning behavior data (each dimension of the prototype vector corresponds to a behavior data feature). Figure 7 The figure below shows the clustering effect diagram drawn using two-dimensional features of sample data and prototype vector after clustering is completed. Based on the clustering results, we can discover the excellent behavioral characteristics of students.

[0087] Example 3

[0088] With the advent of the 5G big data era, educational technology will inevitably have corresponding technological changes. 5G not only provides greater bandwidth and faster network speed, but more importantly, it allows more devices to connect to the network at the same time, and can also achieve more reliable and faster responses. Therefore, it is also a general trend to put Internet technology on the cloud. The Elearn system developed by the present invention is based on this concept. Unlike other applications, it does not require specific application development for different devices. The device can be connected to the Internet and logged in through a browser. This not only saves the manpower and material resources consumed by development, but also facilitates user use. There is no need to download the corresponding application on different devices, and the system can be accessed anytime, anywhere.

[0089] After students log in to the Elearn system, they can select the textbooks they need to read and study in the cloud. During the process of students using the Elearn system to study, the system can record the various operations performed by students on the textbooks and process data in real time.

[0090] During the one-semester course study, the system has also been optimized for the periodic tests in the education process. The system can divide the collected student learning behavior information more finely, perform statistical and normalized processing on the weekly data, make the results of each stage into a serialized data, and perform back propagation training on the long and short attention networks according to the results of each periodic test. Among them, different learning rates are used when training the network for the periodic score data and the final score data, so that the network can be optimized more accurately according to the importance of different exams.

[0091] For each stage test, the learning behavior data of students with excellent and good stage results are clustered through vector quantization analysis to obtain excellent learning behaviors, which are compared with the excellent learning behaviors analyzed from the final results to better guide students' learning process and help them develop better learning habits and methods.

[0092] Example 4

[0093] With the continuous development of artificial intelligence, the combination of deep learning methods and educational technology can effectively guide the continuous improvement of teaching methods. The Elearn system and long-short attention network developed by the present invention are not only for a single course in a single period, but can continuously optimize and improve the method by jointly and continuously collecting and training student learning behavior data of multiple courses in different periods.

[0094] Specifically, the Elearn system can be used in multiple courses at different times, and student data can be collected into the database according to the corresponding time periods and courses, and classified and managed. The long-short attention network can be continuously trained and improved based on the student learning behavior data of multiple courses at different times, gradually improving the prediction accuracy.

[0095] In the course learning process, the well-trained long-short attention network can be used to analyze the students' learning behavior data to obtain the students' predicted grades. The vector quantization method is used to analyze the learning behavior of students with high predicted grades, and the excellent learning behavior is obtained to guide the students with poor predicted grades, dynamically helping them to continuously adjust and improve their learning methods and habits.

[0096] The present invention was implemented in the Deep Learning Theory and Practice course of Ocean University of China in the spring semester of 2020 for one semester. Students who took the course registered according to their user name, student number, and contact information before the course started (such as Figure 2As shown in Figure 1.1.1. After the course started, students read, studied and marked the textbooks in the Elearn system, and the students' learning behavior data were recorded in the database. After the end of the semester, the students' learning behavior data were preprocessed and divided into training and test sets. The training set data was used to perform back propagation training on the long and short time attention network, and the test set student grades were predicted. The prediction accuracy reached 91%, and the test time for each data did not exceed 0.001 seconds. In order to further explore the students' excellent learning behaviors, the learning vector quantization method was used to supervise the clustering of the students' learning behavior data. Specifically, according to the final grades of the Elearn system, the students were divided into three categories: excellent, good, and qualified. The learning behavior data of excellent and good students were clustered using LVQ, and clustered into four subcategories (two excellent and two good). By examining each subcategory, it was found that the learning time and the degree of detail in marking knowledge points during reading were two outstanding excellent learning behaviors. In addition, students who read more pages will also have higher final grades.

[0097] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When the use is implemented in whole or in part in the form of a computer program product, the computer program product includes one or more computer instructions. When the computer program instructions are loaded or executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions may be transmitted from one website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL) or wireless (e.g., infrared, wireless, microwave, etc.) mode) to another website site, computer, server or data center. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integrated. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state hard disk SolidState Disk (SSD)), etc.

[0098] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with the technical field within the technical scope disclosed by the present invention and within the spirit and principle of the present invention should be covered by the protection scope of the present invention.

Claims

1. A learning behavior data mining method, characterized in that: The learning behavior data mining method comprises: Use the Elearn system to collect students' learning behavior data in real time, perform a series of operations on PDF textbooks, and upload the operation information to the server for storage; The collected student learning behavior data is serialized and the characteristics of the student learning behavior data are predicted using the long-short time attention network to obtain the predicted grade of the student's performance; According to their academic performance, students are divided into three categories: excellent, good and qualified. A vector quantization learning-based method is used to supervise the learning behavior data of students with excellent and good grades and cluster them to discover students' excellent learning behaviors. The operation information includes: highlight operation, underline operation, strikethrough operation, marked content, number of different pages read and reading time on the day; The serialization processing of the collected student learning behavior data includes: in the process of preprocessing the student learning behavior data, all the data information is divided into different parts according to the time stage, and in each part, the data of different students are counted separately; Among them, highlight, underline and strikethrough operations are counted according to the number of operations, the marked content is counted by adding up the number of characters in the marked content, the number of different reading pages is counted, and the reading time of the day is counted by the difference between the latest operation time and the earliest operation time of the user in a day; After serializing the collected student learning behavior data, each user has a 6-dimensional data vector in each time segment, where each dimension corresponds to the statistics of 6 types of operation information, and each dimension is normalized; the normalization process includes: selecting the 6 operation values ​​with the largest value in each time segment, dividing all the data in this time segment by the maximum value of the current dimension, and obtaining the final normalized result; The method of using the long-short time attention network to predict the characteristics of student learning behavior data includes: (1) When making predictions based on the long- and short-term attention network, the final course scores are divided into four levels: excellent, good, pass, and fail, and converted into one-hot vectors as the true labels; (2) The data of each student in each time period is regarded as a sequence data. The number of time periods is the number of time steps input into the long and short time attention network. The dimension of the data is the input data dimension size of the long and short time attention network at each time step. (3) The student data is divided into a training set and a test set. During the training process, after processing the time series data of the current student, the long and short time attention network generates a 4-dimensional vector and performs cross entropy loss with the label; (4) The network is trained through back propagation. The grade represented by the maximum dimension of the vector median is the grade of the current student predicted by the network. During the test, the predicted grade is compared with the actual label grade to obtain the network prediction error rate.

2. The learning behavior data mining method according to claim 1, characterized in that: The predicted grades of student performance include: excellent, good, pass and fail.

3. The learning behavior data mining method according to claim 1, characterized in that: The method based on vector quantization learning is used to supervise clustering the learning behavior data of students with excellent and good grades, including: (1) At the beginning of clustering, two samples are randomly selected from each class as the prototype points of the cluster; (2) Randomly select a sample from the sample set, calculate its distance with each vector in the prototype vector group, and select the cluster where the prototype vector with the smallest distance belongs as its division result, and then compare it with the true class label; (3) If the division result is correct, the corresponding prototype vector is closer to the sample; if the division result is incorrect, the corresponding prototype vector is farther away from the sample; (4) Iterate until the algorithm converges or reaches the maximum step, and output a prototype vector group that represents the characteristics of good learning behavior data.

4. A computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor executes the steps of the learning behavior data mining method according to any one of claims 1 to 3.

5. An information data processing terminal, characterized in that: The information data processing terminal is used to implement the learning behavior data mining method described in any one of claims 1 to 3.

6. A learning behavior data mining system for implementing the learning behavior data mining method according to any one of claims 1 to 3, characterized in that: The learning behavior data mining system comprises: The Elearn system is used to collect students’ learning behavior data, perform a series of operations on PDF textbooks, and upload the operation records to the server for storage; The grade prediction module is used to serialize the collected student learning behavior data and use the long-short time attention network to predict the student grade; The learning behavior mining module is used to classify students into three categories: excellent, good, and qualified according to their academic performance. It uses a vector quantization learning-based method to perform supervised clustering on the learning behavior data of students with excellent and good grades to mine students' excellent learning behaviors.

Images