Big data processing method based on cloud computing and cloud computing platform

Abstract

The invention discloses a big data processing method based on cloud computing and a cloud computing platform, and relates to the technical field of cloud computing, the method comprises the following steps: firstly, constructing a course knowledge graph comprising a course pre-repair dependency digraph and a knowledge unit density index, and constructing a course feature vector according to the course knowledge graph; then extracting the effective learning duration and the test score of the student to construct a student-course mastering matrix, and constructing a student feature vector in combination with a historical completion rate; calculating the similarity between the trainee and other trainees to generate a recommendation candidate set, predicting the completion probability and the test score of the trainee in the course in the recommendation candidate set by using a multi-target depth model, and finally recommending the course with the highest prediction score; and in combination with the course knowledge graph and the student feature vector, personalized course recommendation is realized, and the learning efficiency and effect of the student are improved.

Description

Technical Field

[0001] The present invention belongs to the field of cloud computing technology, and specifically relates to a big data processing method and a cloud computing platform based on cloud computing. Background Art

[0002] With the widespread popularization of Internet technology, the gradual maturity and integration of technologies such as computer networks, virtualization, distributed computing, parallel computing, and data center technology have laid the foundation for the rise and widespread application of cloud computing.

[0003] The rise of online education is closely related to cloud computing technology. The data center technology in cloud computing technology provides strong technical support for the development of online education. The recommendation and sharing of educational resources is one of the important goals of online education. Existing technologies often use traditional collaborative filtering algorithms, or simple keyword matching and rule matching in the process of processing recommendations based on data center technology. This approach has obvious shortcomings in dealing with complex user behaviors and diverse needs. The recommended content may be based more on group behavior rather than the specific interests of users. This lack of personalized recommendation may lead to low user satisfaction with the recommendation results. Due to the limitations of the recommendation algorithm, many high-quality educational resources may not be fully utilized, while some low-quality or irrelevant resources are over-recommended. This imbalance in resource allocation not only wastes computing resources, but also affects the learning efficiency of users. Based on this, this application proposes a big data processing method and a cloud computing platform based on cloud computing. Summary of the Invention

[0004] In response to the shortcomings of the existing technology, the present invention provides a big data processing method and cloud computing platform based on cloud computing, which solves the problems of poor recommendation accuracy and difficulty in personalized recommendation in online education systems in the existing technology.

[0005] The purpose of the present invention can be achieved through the following technical solutions:

[0006] A cloud computing-based big data processing method, which is applied to a cloud computing platform, includes the following:

[0007] Step 1: Construct a course knowledge graph, which contains the corresponding course C i The course prerequisite dependency directed graph G and the knowledge unit density index Constructing information about course C based on course knowledge graph i The course feature vector

[0008] Step 2: Extract any student S u As of now, study the current course C i Effective learning time tu , and test scores Based on test scores Effective learning time u Construct student S u Associated Learner-Course Mastery Matrix

[0009] Get student S u Historical completion rates, combined with the student-course mastery matrix And effective learning time t u Construct student S u The student feature vector

[0010] Step 3: Get other students who are studying the same knowledge field and calculate the difference between them and the student S u The similarity of students and generate recommendation candidate sets, while predicting student S u After learning the completion probability and test scores of the courses in the corresponding recommendation candidate set, we can extract the predicted student S u Courses with the highest completion probability and test scores are recommended.

[0011] As a further solution of the present invention, the specific method of constructing the course knowledge graph described in Step 1 is:

[0012] Determine the knowledge domain Z as the target domain, extract j different courses in the target domain, and the prerequisite dependencies between j different courses, and define the course prerequisite dependency directed graph G(C,E) based on the prerequisite dependencies, where C represents the course. According to the prerequisite dependencies, j courses are recorded as: course sequence C1, C2, ..., C j , C i For any course, i and j are counting indexes, starting from 1, and i does not exceed j;

[0013] In the course prerequisite dependency directed graph G(C,E), E represents the prerequisite dependency relationship between different courses. i→j represents that completing course i is a prerequisite for studying course j;

[0014] Extract any course C from the course sequence i , get C i The total number of knowledge units K in i , Course duration T i , according to the total number of knowledge units K i Divide by the course duration T i Get C i Knowledge unit density index combination and G(C,E) to get C i Course knowledge graph.

[0015] As a further solution of the present invention, the construction of the course feature vector for the corresponding course in step 1 is as follows: The specific method is:

[0016] Income-Based Course C i Course knowledge graph, further obtain course C i Difficulty level D n , the D n ∈[D1,D m ], where the difficulty level range [D1,D m ] Prepared by the operator, D1 to D m Indicates from easy to difficult, n and m are both counting indexes, starting from 1, and n does not exceed m;

[0017] Extract C i Knowledge domain Z, knowledge unit density index and difficulty level D n Combine to get C i The course feature vector

[0018] As a further solution of the present invention, in the step Step 2, the student S is constructed u Associated student-course mastery matrix M u-C The specific method is:

[0019] Take the current time point as the deadline and get the student S u Learn C i Effective learning time t u , t u is calculated as follows:

[0020] Get student S u Learn C i The total number of times, determine any S u Learn C i duration, and then obtain the effective duration threshold preset by the operator. If S u This time I studied C i If the duration is lower than the effective duration threshold, then mark S u This time I studied C i is marked as invalid learning, otherwise it is marked as effective learning;

[0021] Extract S u Learn C i The sum of the duration of effective learning in all times is denoted as t u, that is, student S u Study Course C i Effective learning time;

[0022] Get S again u Study Course C i The average of all test scores during the course is recorded as the course average score The score of the test closest to the deadline and the standard deviation of the test scores and build a student-course mastery matrix Quantitative student S u For Course C i degree of mastery.

[0023] As a further solution of the present invention, in step 2, the student S is constructed. u The student feature vector The specific method is:

[0024] Extract S u The total number of courses studied in knowledge domain Z, and the courses that have been completed are extracted, and the proportion of completed courses in the total number of courses is calculated to obtain S u Historical completion rates

[0025] Based on the identified student S u For Course C i Degree of mastery Effective learning time t u and historical completion rates The combined student feature vector

[0026] As a further embodiment of the present invention, the quantified student S u For Course C i The degree of mastery, if S u C i Degree of mastery If the mastery level exceeds the operator's preset threshold, the next step is to u Make course recommendations;

[0027] Otherwise, no processing is performed.

[0028] As a further solution of the present invention, the step Step 3 is to obtain other students who are studying the same knowledge field and calculate the number of students S u The specific method of calculating the similarity of students and generating the recommendation candidate set is as follows:

[0029] Get with students S u Any student S who studies the same knowledge domain Zo , extract S o With S u The same courses among all the courses learned in knowledge domain Z are recorded as the same course set Including S u Learning C i ;

[0030] Based on the constructed student-course mastery matrix, calculate S u For the same course set The mastery level of all courses in the course is averaged and scored as S u right Average mastery level Similarly, get student S o right Average mastery level

[0031] Based on the determined S u and S o right The average mastery level of S u With S o The similarity SIM(S u ,S o );

[0032] If S u With S o The similarity of students SIM(S u ,S o )≥80%, then S o Marked as S u similar students and extract S o About Course C in all courses studied i The next course in the course prerequisite dependency directed graph G(C,E) is denoted as

[0033] If the student similarity SIM(S u ,S o )<80%, skip S o , continue to search for the next one;

[0034] Similarly, select y students S u Similar students of course C i The next course in the course prerequisite dependency directed graph G(C,E) is denoted as As the recommended candidate set, where y is the operator's preset value, Any course in the recommended candidate set.

[0035] As a further solution of the present invention, the predicted student S in step 3 u The specific methods of completing the courses in the recommended candidate set and the test scores are as follows:

[0036] Extract student S u The student feature vector and any one course in the recommended candidate set The course feature vector use: The student feature vector With the course feature vector Spliced ​​into a joint input vector, where d is the student feature vector The dimension of k is the course feature vector The dimensions are preset by the operator;

[0037] Based on the multi-objective deep model, a shared bottom layer and task-specific layer as well as a loss function are constructed to predict each course in the recommended candidate set and obtain the student S u For each course, the probability of completion and the test score.

[0038] As a further solution of the present invention, the student S u After calculating the completion probability and test score of each course, normalize the completion probability and test score of all courses, take the sum of the normalized completion probability and test score of any course, and select the course with the largest sum to give to student S. u Make a recommendation.

[0039] As a further solution of the present invention, a cloud computing platform includes the following: a memory, a processor, and a platform interface;

[0040] The memory is used to store the analysis and calculation results of any of the steps described above, and save the big data processing method based on cloud computing;

[0041] The processor is used to read the cloud computing-based big data processing method stored in the memory and execute any one of the steps described above;

[0042] The platform interface interacts with online education software or websites to obtain relevant data from the online education software or websites;

[0043] The relevant data are the non-analyzed data and non-calculated data described in any one of the above contents.

[0044] Beneficial effects of the present invention:

[0045] (1) The present invention provides an intelligent and personalized course recommendation solution that combines course knowledge graphs, student feature vectors, and similarity calculations. While optimizing resource allocation, it improves the accuracy and efficiency of course recommendations, and can enhance students' learning motivation and satisfaction to a certain extent. The application of this technology can significantly improve the user experience and operational effectiveness of online education platforms, software, and websites.

[0046] (2) The present invention constructs a course knowledge graph and a course feature vector to represent the prerequisite dependencies and knowledge unit density indicators between courses in a structured manner, thereby improving the accuracy and efficiency of recommendations. By analyzing the difficulty level and knowledge unit density indicators of courses, it can provide students with personalized learning path suggestions, optimize resource allocation, and enhance learning motivation. In addition, by analyzing course relationships, it can alleviate the cold start problem to a certain extent and provide data-driven decision support for educational institutions.

[0047] (3) The present invention provides a method for calculating student similarity based on student feature vectors and course feature vectors. This method can screen out other students with similar behavior patterns to the target student and generate a recommendation candidate set using group intelligence, thereby ensuring the diversity and accuracy of the recommendation results. At the same time, this method avoids invalid recommendations to students through a preset mastery level threshold, thereby improving the pertinence and efficiency of the recommendation.

[0048] (4) The multi-objective deep model of the present invention has been extended to a method that combines student feature vectors and course feature vectors, achieving accurate prediction of student learning behavior, including course completion probability and test scores. By sharing the design of the bottom layer and task-specific layer, it can extract the common characteristics and nonlinear characteristics of the interaction between students and courses, and optimize the prediction results of the two tasks at the same time. In addition, the added loss function balances the contribution of the two tasks by weight, further improving the robustness and adaptability of the model. In the final recommendation link, it ensures that the recommendation results take into account both the completion probability and the test score, thereby providing students with courses that best meet their learning needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] The present invention will be further described below with reference to the accompanying drawings.

[0050] Figure 1 It is a structural diagram of the platform of the present invention;

[0051] Figure 2 Schematic diagram of the process of the method described in Example 2 of the present invention;

[0052] Figure 3 Schematic diagram of the process of the method described in Example 3 of the present invention. DETAILED DESCRIPTION

[0053] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0054] Example 1

[0055] A cloud computing platform, such as Figure 1 As shown, the platform includes the following:

[0056] The memory is one of the core components of a cloud computing platform described in the present invention, and is used to store and manage all data and methods related to course recommendations (a big data processing method based on cloud computing), where the data includes student-related data obtained from online education platforms, software, and websites, as well as course-related data. In addition, the memory is also used to save the parameters of the multi-objective deep model, student feature vectors, and course feature vectors. The memory should meet the requirements of using distributed storage technology, be able to efficiently process large-scale data sets, and ensure fast reading and writing of data and persistent storage.

[0057] The processor, as a core part of a cloud computing platform, is responsible for executing cloud computing-based big data processing methods stored in the memory. By reading the methods and data in the memory, the processor runs a multi-objective deep model to complete complex computing tasks such as interactive analysis of student characteristics and course characteristics, and prediction of completion probabilities and test scores. The processor should support multi-threading and parallel computing and be able to efficiently process large-scale data sets. In addition, the processor is also responsible for updating model parameters in real time to ensure the accuracy and adaptability of the prediction results.

[0058] The platform interface is a bridge for the cloud computing platform described in the present invention to interact with external systems (online education platforms, software, websites, etc.), and is responsible for exchanging data with online course software or websites; the platform interface communicates with external systems to obtain students' learning behavior data, course content data and other non-analytical (calculated) data, and transmits them to the memory through the bus for use by the processor. The platform interface should contain encrypted transmission and identity authentication mechanisms to ensure the security and privacy of data transmission.

[0059] Example 2

[0060] This embodiment discloses a method for constructing a student feature vector and a course feature vector, such as Figure 2 As shown, the specific steps include:

[0061] Before implementing this method, it is necessary to first determine any knowledge domain as the target domain. In this method, knowledge domain Z is used as the target domain. The knowledge domain Z can be computer science, mathematics, physics, or linguistics, etc., depending on the actual situation.

[0062] Based on the determined knowledge domain Z, several different courses are obtained from the knowledge domain Z, and the prerequisite dependencies between different courses are locked. The prerequisite dependency refers to the connection between two courses. For example, the existing course C j with C i , they are two courses in this knowledge area Z, of which course C j The knowledge points in C are more i Difficult, and as a course C i The next level of courses to be studied is defined as Course C j The knowledge level in the course should be higher than that of course C i , must complete Course C i Only then can you continue to study Course C j .

[0063] Based on the determined prerequisite dependencies, a course prerequisite dependency directed graph G(C,E) can be defined, where C represents the course and E represents the prerequisite dependency between two different courses. For example, if we extract j different courses, we can record them as course sequences C1, C2, ..., C according to the prerequisite dependency between the two-year courses. j , where C i C1 to C j For any course in E, i and j are count indexes, starting from 1, and i does not exceed j; i→j Represents completion of course C i The next step is to continue studying course C j prerequisites.

[0064] Based on the determined course sequence C1, C2, ..., C j , extract any one course from it and record it as C i , get the course C i The total number of knowledge units covered, denoted as K i The total number of knowledge units refers to the smallest learnable unit of course content, which can also be understood as a separate knowledge point;

[0065] Then obtain the course C i The course duration is recorded as T i , according to the following formula:

[0066]

[0067] Calculate the current course C i Knowledge unit density index The resulting course C i Knowledge unit density index and Course C i Combine the course prerequisite dependency directed graph G(C,E) to get course C i Course knowledge graph;

[0068] From the course sequence C1, C2, ..., C j Extract any course C i , get course C i The total number of knowledge units in: K i , and Course C i Course duration: T i , according to the following formula:

[0069] Calculate the current course C i Knowledge unit density index Based on the knowledge unit density index and Course C i Course C is obtained by combining the directed graph of course prerequisite dependencies i The course knowledge graph reflects the prerequisite relationship between each course and the next, clarifies the order of learning different courses, and quantifies the amount of knowledge in a corresponding course through the knowledge unit density index. This can help students better understand the content compactness of the current corresponding course and lay a data support foundation for subsequent recommendation algorithms.

[0070] Determine the difficulty range [D1, D m ] and lock the course C i The degree of difficulty is recorded as D n , the D n ∈[D1,D m ], indicating course C i Difficulty level D n Included in the difficulty range [D1,D m ], the difficulty level range [D1,D m ] is formulated by the operator in combination with the courses and experience in this knowledge field, from D1 to D m Indicates the difficulty level of the course from easy to difficult, n and m are both counting indexes, starting from 1, and n does not exceed m;

[0071] Based on the above steps, the course C i Knowledge area Z, course C iKnowledge unit density index and Course C i Difficulty level D n , combine and jointly build Course C i The course feature vector of is denoted as:

[0072] Get the current time point as the deadline, and get any student S before the deadline u Any course that is currently being studied but not completed C i Effective learning time t u , the effective learning time t u The calculation method is as follows:

[0073] Extract the student S u Study Course C before the deadline i The total number of times, and each time you study the course C i The duration of the test is determined by the operator, and the effective duration threshold is obtained, and then the trainee S is judged. u Each study course C i The duration of the training exceeds the effective duration threshold preset by the operator. If the trainee S u Any one-time study course C i If the duration of the course is lower than (does not exceed) the effective duration threshold, the student will be i Mark as invalid learning, filter students S based on this method u Ineffective learning behaviors;

[0074] If student S u Any one-time study course C i If the duration of the course is greater than or equal to the effective duration threshold, the student will be i Tagged as effective learning;

[0075] Repeat the above steps and extract the student S u Study Course C i The total number of times belongs to the number of effective learning, obtain the duration of all effective learning, and sum them up to get the total duration of all effective learning, and record it as t u , that is, student S u Study Course C i The effective learning time is t u ;

[0076] Then get student S u Studying Course C i All test scores in the process are averaged and recorded as and the score of the most recent test Calculate the standard deviation of test scores Build a student-course mastery matrix through the above three types of data The details are as follows:

[0077]

[0078] The student-course mastery matrix can be used to quantify and obtain the student S u For Course C i Degree of mastery Where e is the natural logarithm;

[0079] Then extract student S in knowledge domain Z u The total number of all courses studied in the past, and the total number of completed courses, calculate the proportion of the total number of completed courses in the total number of all courses, and use the obtained proportion as the student S u The historical completion rate of

[0080] Based on the historical completion rates determined Combined with student S u For Course C i Degree of mastery And student S u For Course C i Effective learning time t u , get student S u The current student feature vector is denoted as The student feature vector Shows the student S in a multi-dimensional form u The current learning situation provides data support about the students for subsequent recommendation algorithms, making it easier to provide different recommended courses based on the actual learning situation of different students, thereby optimizing the recommendation strategy.

[0081] Example 3

[0082] This embodiment discloses a method for generating a set of recommended candidates for a course for any student, such as Figure 3 As shown, the specific steps include:

[0083] Extract student S u For Course C i Degree of mastery And obtain the operator's preset mastery threshold, and convert the obtained S u For Course C i Degree of mastery Compare the operation with the mastery threshold preset by the operator. If the trainee S uFor Course C i Degree of mastery If the mastery level exceeds the operator's preset threshold, it means that the recommended candidate set can be generated for the student and subsequent recommendation operations can be performed. u For Course C i Degree of mastery If the mastery level threshold preset by the operator is not exceeded, the student S will be skipped. u Yes, no processing.

[0084] If student S is determined u For Course C i Degree of mastery If the mastery level exceeds the operator's preset threshold, the trainee S u Anyone who is different from student S in the same knowledge domain Z u The students are denoted as S o ;

[0085] Extract the student S o All courses studied in knowledge area Z, and student S u All courses learned in knowledge domain Z are obtained by taking the intersection of the student S o All courses studied with students S u The same courses are recorded as the same course set The same set of courses Need to include S u Currently studying course C i .

[0086] If the same course set Excluding Course C i , then skip the student S o , continue to search for the next student;

[0087] Based on the constructed student-course mastery matrix, calculate the student S u For the same course set The mastery level of any course in the course is averaged and the student S is obtained. u For the same course set The average mastery level of all courses in And so on, get student S o For the same course set The average mastery level of

[0088] According to the formula for calculating student similarity:

[0089]

[0090] The student S is calculated by the above formula u With student S o The similarity SIM(S u ,S o ), as described in the formula Indicates that the value range of course C is student S o With student S u The same set of courses, i.e.

[0091] Based on the identified student S u With student S o The similarity SIM(S u ,S o ), perform similarity judgment processing, if the student similarity SIM(S u ,S o )≥80%, then the similarity is judged and the student S is selected. o Marked as student S u Similar students, and then extract student S o About Course C in all courses studied i The next course in its corresponding course prerequisite dependency directed graph G(C,E) is recorded as

[0092] If student S u With student S o Failed to pass the similarity judgment, that is, the student similarity SIM (S u ,S o )<80%, then skip this student S o , continue to search for the next student;

[0093] Repeat the above to obtain course C i The next course in its corresponding course prerequisite dependency directed graph G(C,E) Steps to reselect y students S u For similar students, a total of y about course C i The next stage of the course is recorded as And as the recommended candidate set, where y is the operator's preset value, To recommend any course in the candidate set, the value of x starts from 1 and the maximum value is y.

[0094] Example 4

[0095] This embodiment further discloses a method for recommending courses to students based on a recommendation candidate set based on Embodiments 2 and 3, specifically comprising the following steps:

[0096] Based on the course C determined in Example 2 i The course feature vector Method to obtain any one of the recommended candidate courses determined in Example 3 The course feature vector is denoted as

[0097] Extract the student S determined in Example 2 u The student feature vector Combined courses The course feature vector use The student feature vector With the course feature vector Spliced ​​into a joint input vector (as input value, output value is student S u Complete the course The completion probability and test score of the course will be explained in the subsequent content). The joint input vector is used to combine the characteristics of the students and the course to form a joint input vector describing the student-course interaction, where d is the student feature vector The dimension of k is the course feature vector Dimensions;

[0098] A multi-objective deep model related to student-course interaction (hereinafter referred to as the model, which is an extension of the existing technology for student-course interaction) is constructed, including a shared bottom layer, a task-specific layer, and a loss function. The shared bottom layer of the model mainly consists of two layers of fully connected neural networks (FCNs) to extract common features of student-course interaction. The first layer h1 is as follows:

[0099]

[0100] in, is the weight matrix, which reduces the input dimension from d+k to the dimension γ preset by the operator, b1 is the bias term, and ReLU is the activation function, which introduces nonlinearity to help the multi-target deep model learn complex feature relationships;

[0101] The second layer h1 is as follows:

[0102] h2=ReLU(W2·h1+b2)

[0103] Where W1∈R γ×δ , is the weight matrix, which reduces the input dimension γ to the dimension δ preset by the operator, b2 is the bias term, and ReLU is the activation function to further extract nonlinear features;

[0104] The shared bottom layer extracts common features of the interaction between students and courses, providing basic feature representation for the subsequent task-specific layer;

[0105] The task-specific layer includes two prediction branches, namely, a completion probability prediction branch and a test score prediction branch. The completion probability prediction branch h3 is shown as follows:

[0106] h3=Sigmoid(W3·h2+b1)

[0107] Among them, Sigmoid is the Sigmoid function, the output value range is [0,1], W3∈R δ×1 , is the weight matrix, reducing the input dimension from δ to dimension 1;

[0108] The test score prediction branch h4 is as follows:

[0109] h4=W3·h2+b2

[0110] Among them, the output range of h4 is constrained to [0,100] by the loss function;

[0111] The loss function is as follows:

[0112] τ=α·MSE(h3^,h3)+(1-α)·BCE(h4^,h4)

[0113] Where α is the weight preset by the operator, α∈[0,1], which is used to balance the two tasks in the loss function: the weights of MSE(h3^,h3) and BCE(h4^,h4), MSE represents mean square error, BCE represents binary cross entropy, h3^ is the completion probability predicted by the multi-target deep model, h4^ is the test score predicted by the multi-target deep model, h3 is the actual completion probability, and h4 is the actual test score. The loss function is constructed based on the multi-task learning method, so the predicted value and the true value need to appear in the loss function at the same time. The purpose is to correct the relevant parameters in the above prediction branch based on the true value. Each prediction will be corrected until the constructed model achieves the best prediction effect. Therefore, this model requires a certain adaptation period in the actual application process, or it can be fed through a large amount of historical data of student-course interaction to improve the accuracy of the relevant parameters in the model, thereby shortening the adaptation period of this model in the actual application process;

[0114] Based on the completion probability prediction branch h3 and the test score prediction branch h4 in the model, each recommended course in the recommendation candidate set is predicted. Finally, the model obtains the completion probability and test score of each course.

[0115] Then normalize the completion probabilities of all courses in the recommended candidate set, normalize the test scores of all courses in the recommended candidate set, obtain the completion probability and test scores of any course after normalization, and sum them up. Then select the course with the largest sum and present it to student S. u Make recommendations;

[0116] This embodiment constructs a multi-objective deep model, which combines the student feature vector and the course feature vector to achieve accurate prediction of student learning behavior and personalized course recommendation. Specifically, by splicing the student feature vector and the course feature vector into a joint input vector and using this as the input of the model, it can comprehensively describe the interaction characteristics between the student and the course, and simultaneously optimize the prediction of completion probability and test score through a multi-task learning method; the shared bottom layer of the model extracts common features, and the task-specific layer processes the prediction of completion probability and test score respectively. The loss function improves the prediction accuracy by balancing the weights of the two tasks; through normalization processing and summation operations, the course that best suits the student is finally recommended.

[0117] Some of the data in the formulas described above are dimensionless and numerically calculated. Meanwhile, the contents not described in detail in this specification belong to the prior art known to those skilled in the art.

[0118] The above contents are merely examples and explanations of the present invention. Those skilled in the art may make various modifications or additions to the described specific embodiments or replace them in similar ways. As long as they do not deviate from the invention or exceed the scope defined by the claims, they should all fall within the scope of protection of the present invention.

[0119] It is important to note that all user data collected in this application is collected with the user's consent and authorization. Furthermore, the use of user data is legal and compliant, and the use and processing of user data complies with the relevant laws, regulations, and standards of the relevant regions.

Claims

1. A big data processing method based on cloud computing, characterized in that: This method is applied to the cloud computing platform and includes the following: Step 1: Construct a course knowledge graph, which contains the corresponding course C i The course prerequisite dependency directed graph G and the knowledge unit density index Constructing information about course C based on course knowledge graph i The course feature vector Step 2: Extract any student S u As of now, study the current course C i Effective learning time t u , and test scores Based on test scores Effective learning time u Construct student S u Associated Learner-Course Mastery Matrix Get student S u Historical completion rates, combined with the student-course mastery matrix And effective learning time t u Construct student S u The student feature vector Step 3: Get other students who are studying the same knowledge field and calculate the difference between them and the student S u The similarity of students and generate recommendation candidate sets, while predicting student S u After learning the completion probability and test scores of the courses in the corresponding recommendation candidate set, we can extract the predicted student S u Courses with the highest completion probability and test scores are recommended.

2. The cloud computing-based big data processing method according to claim 1, characterized in that: The specific method of constructing the course knowledge graph described in Step 1 is: Determine the knowledge domain Z as the target domain, extract j different courses in the target domain, and the prerequisite dependencies between j different courses, and define the course prerequisite dependency directed graph G(C,E) based on the prerequisite dependencies, where C represents the course. According to the prerequisite dependencies, j courses are recorded as: course sequence C1, C2, ..., C j , C i For any course, i and j are counting indexes, starting from 1, and i does not exceed j; In the course prerequisite dependency directed graph G(C,E), E represents the prerequisite dependency relationship between different courses. i→j represents that completing course i is a prerequisite for studying course j; Extract any course C from the course sequence i , get C i The total number of knowledge units K in i , Course duration T i , according to the total number of knowledge units K i Divide by the course duration T i Get C i Knowledge unit density index combination and G(C,E) to get C i Course knowledge graph.

3. The cloud computing-based big data processing method according to claim 2, characterized in that: Construct the course feature vector for the corresponding course as described in Step 1 The specific method is: Income-Based Course C i Course knowledge graph, further obtain course C i Difficulty level D n , the D n ∈[D1,D m ], where the difficulty level range [D1,D m ] Prepared by the operator, D1 to D m Indicates from easy to difficult, n and m are both counting indexes, starting from 1, and n does not exceed m; Extract C i Knowledge domain Z, knowledge unit density index and difficulty level D n Combine to get C i The course feature vector 4. The cloud computing-based big data processing method according to claim 1, characterized in that: In the step Step 2, construct the student S u Associated student-course mastery matrix M u-C The specific method is: Take the current time point as the deadline and get the student S u Learn C i Effective learning time t u , t u is calculated as follows: Get student S u Learn C i The total number of times, determine any S u Learn C i duration, and then obtain the effective duration threshold preset by the operator. If S u This time I studied C i If the duration is lower than the effective duration threshold, then mark S u This time I studied C i is marked as invalid learning, otherwise it is marked as effective learning; Extract S u Learn C i The sum of the duration of effective learning in all times is denoted as t u , that is, student S u Study Course C i Effective learning time; Get S again u Study Course C i The average of all test scores during the course is recorded as the course average score The score of the test closest to the deadline and the standard deviation of the test scores and build a student-course mastery matrix Quantitative student S u For Course C i degree of mastery.

5. The cloud computing-based big data processing method according to claim 4, characterized in that: In step 2, build student S u The student feature vector The specific method is: Extract S u The total number of courses studied in knowledge domain Z, and the courses that have been completed are extracted, and the proportion of completed courses in the total number of courses is calculated to obtain S u Historical completion rates Based on the identified student S u For Course C i Degree of mastery Effective learning time t u and historical completion rates The combined student feature vector 6. The cloud computing-based big data processing method according to claim 5, characterized in that: Quantitative student S u For Course C i The degree of mastery, if S u C i Degree of mastery If the mastery level exceeds the operator's preset threshold, the next step is to u Make course recommendations; Otherwise, no processing is performed.

7. The cloud computing-based big data processing method according to claim 5, characterized in that: As described in Step 3, other students who study the same knowledge domain are obtained and the number of students S is calculated. u The specific method of calculating the similarity of students and generating the recommendation candidate set is as follows: Get with students S u Any student S who studies the same knowledge domain Z o , extract S o With S u The same courses among all the courses learned in knowledge domain Z are recorded as the same course set Including S u Learning C i ; Based on the constructed student-course mastery matrix, calculate S u For the same course set The mastery level of all courses in the course is averaged and scored as S u right Average mastery level Similarly, get student S o right Average mastery level Based on the determined S u and S o right The average mastery level of S u With S o The similarity SIM between students (S u ,S o ); If S u With S o The similarity of students SIM(S u ,S o )≥80%, then S o Marked as S u similar students and extract S o About Course C in all courses studied i The next course in the course prerequisite dependency directed graph G(C,E) is denoted as If the student similarity SIM(S u ,S o )<80%, skip S o , continue to search for the next one; Similarly, select y students S u Similar students of course C i The next course in the course prerequisite dependency directed graph G(C,E) is denoted as As the recommended candidate set, where y is the operator's preset value, Any course in the recommended candidate set.

8. The cloud computing-based big data processing method according to claim 7, characterized in that: The predicted student S described in Step 3 u The specific methods of completing the courses in the recommended candidate set and the test scores are as follows: Extract student S u The student feature vector and any one course in the recommended candidate set The course feature vector use: The student feature vector With the course feature vector Spliced ​​into a joint input vector, where d is the student feature vector The dimension of k is the course feature vector The dimensions are preset by the operator; Based on the multi-objective deep model, a shared bottom layer and task-specific layer as well as a loss function are constructed to predict each course in the recommended candidate set and obtain the student S u For each course, the probability of completion and the test score.

9. The cloud computing-based big data processing method according to claim 8, characterized in that: Get student S u After calculating the completion probability and test score of each course, normalize the completion probability and test score of all courses, take the sum of the normalized completion probability and test score of any course, and select the course with the largest sum to give to student S. u Make a recommendation.

10. Cloud computing platform, characterized in that The platform includes the following: memory, processor, platform interface; The memory is used to store the analysis and calculation results of the steps described in any one of claims 1 to 9, and to save the cloud computing-based big data processing method; The processor is configured to read the cloud computing-based big data processing method stored in the memory and execute the method to implement the steps of any one of claims 1 to 9; The platform interface interacts with online education software or websites to obtain relevant data from the online education software or websites; The relevant data are the non-analytical data and non-calculated data described in any one of claims 1 to 9.

Images