A course recommendation method and device based on big data

Abstract

The present invention provides a course recommendation method and device based on big data, which relates to the field of online education technology. The method includes: acquiring and analyzing the learning behavior data of a target user; constructing an interaction matrix between a user and a course according to the learning behavior data; calculating the similarity between courses or between users according to the interaction matrix to obtain a similarity calculation result; generating the target user's preference for courses according to the similarity calculation result; generating a course recommendation list according to the target user's learning needs, the target user's preference for courses, the course information in the course library, and a dynamic threshold; and sending the course recommendation list to the target user. The present invention can find other users with similar interests to the target user, and can find new courses similar to the courses liked by the target user.

Description

Technical Field

[0001] The present invention relates to the field of online education technology, and in particular to a course recommendation method and device based on big data. Background Art

[0002] With the development of the Internet, the number of courses in online education platforms continues to grow. However, faced with a huge number of courses, users often find it difficult to find courses that are truly suitable for them. In order to solve this problem, researchers have proposed various course recommendation methods based on big data.

[0003] Traditional recommendation systems mainly rely on content matching, that is, recommending similar new courses based on the user's historical learning records. This method has a cold start problem and cannot achieve effective recommendations for new users who have no learning records. In addition, it cannot distinguish the personalized learning preferences of different users. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a course recommendation method and device based on big data, which can find other users with similar interests to the target user and find new courses similar to the courses liked by the target user.

[0005] In order to solve the above technical problems, the technical solution of the present invention is as follows:

[0006] In a first aspect, a course recommendation method based on big data, the method comprising:

[0007] Obtain and analyze target users’ learning behavior data;

[0008] constructing an interaction matrix between the user and the course according to the learning behavior data;

[0009] Calculate the similarity between courses or between users according to the interaction matrix to obtain a similarity calculation result;

[0010] Generate the target user's preference for the course according to the similarity calculation result;

[0011] Generate a course recommendation list based on the target user's learning needs, the target user's preferences for courses, the course information in the course library, and dynamic thresholds;

[0012] The course recommendation list is sent to the target user.

[0013] Furthermore, the learning behavior data of target users is obtained and analyzed, including:

[0014] Obtain all learning behavior data of target users on the online learning platform;

[0015] Preprocessing the learning behavior data to obtain preprocessed data;

[0016] The preprocessed data is analyzed to obtain the user's learning patterns and preferences, as well as key factors that affect learning outcomes.

[0017] Furthermore, in the interaction matrix, rows represent users, columns represent courses, and the value of each cell is the interaction value of the corresponding user for a specific course.

[0018] Furthermore, based on the learning behavior data, an interaction matrix between the user and the course is constructed, including:

[0019] Traverse all learning behavior data in dynamic order and extract all corresponding relationships between users and courses;

[0020] According to the correspondence between all users and courses, obtain all user behavior records on the course;

[0021] Calculate the fusion interaction value based on all the user's course behavior records;

[0022] constructing an interaction matrix according to the fused interaction values;

[0023] The missing values ​​in the interaction matrix are processed, and the interaction matrix is ​​standardized to obtain a standardized interaction matrix.

[0024] Further, according to the interaction matrix, similarities between courses or between users are calculated to obtain similarity calculation results, including:

[0025] According to the interaction matrix, construct user vector and course vector respectively;

[0026] According to the user vector and the course vector, respectively calculating the average value of the interaction values ​​of all users and the average value of the interaction values ​​of all courses;

[0027] Based on the average interaction values ​​of all users and the average interaction values ​​of all courses, as well as the dynamic user weights, the similarity between every two users and the similarity between every two courses are calculated respectively.

[0028] Furthermore, according to the similarity calculation result, the target user's preference for the course is generated, including:

[0029] According to the similarity between every two users and every two courses, extract the N users with the highest similarity to each target user and the M courses most similar to the courses rated by the target user;

[0030] The target user's preference for courses is generated based on the N users with the highest similarity to each target user and the M courses most similar to the courses rated by the target user.

[0031] Furthermore, a course recommendation list is generated based on the target user's learning needs, the target user's preferences for courses, the course information in the course library, and the dynamic threshold, including:

[0032] Separate users u and courses i The original feature vector of X u and Y i Through deep neural network NN u and NN i Convert to feature vector U u and I i ,in, U u =NN u (X u )=σ(W u X u +b u ) I i =NN i (Y i )=σ(W i Y i +b i ) ,in, W u and W i is the weight matrix, b u and b i is the bias vector, s is the activation function;

[0033] according to Calculating prediction scores ,in, is a dot product operation, representing the converted user feature vector U u and the course feature vector I iThe similarity between b u is the user bias term, representing the user u Overall rating tendency; is the global average rating, which represents the average rating of all users on all courses; is the course bias term, representing the course i Overall popularity of T Used to represent the transpose of a matrix;

[0034] According to user Adjust dynamic thresholds based on historical behavior and feedback i u ,in, ,in, Is a user u The historical average rating of α is a constant;

[0035] Count all courses i Prediction score of , and obtain the prediction score ≥ i u of N A list of courses is provided as a recommendation.

[0036] In the second aspect, a course recommendation device based on big data includes:

[0037] An acquisition module is used to acquire and analyze the learning behavior data of the target user; based on the learning behavior data, an interaction matrix between the user and the course is constructed; based on the interaction matrix, similarities between courses or between users are calculated to obtain similarity calculation results;

[0038] A processing module is used to generate the target user's preference for courses based on the similarity calculation result; generate a course recommendation list based on the target user's learning needs, the target user's preference for courses, the course information in the course library and the dynamic threshold; and send the course recommendation list to the target user.

[0039] According to a third aspect, a computing device includes:

[0040] one or more processors;

[0041] The storage device is used to store one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement the above method.

[0042] In a fourth aspect, a computer-readable storage medium stores a program, and the program implements the above method when executed by a processor.

[0043] The above solution of the present invention includes at least the following beneficial effects:

[0044] The above scheme of the present invention can collect the user's historical learning data, understand the user's learning interests and preferences, and analyze the learning data to discover the user's learning rules by acquiring and analyzing the target user's learning behavior data. An interaction matrix between users and courses is constructed, and the matrix intuitively represents the interactive relationship between users and courses. By calculating the similarity between courses or users, other users with similar interests to the target user can be found, and new courses similar to the courses liked by the target user can be found. The target user's preference for courses is generated, and the user's interests and historical learning behaviors are integrated to accurately express the user's course preferences, and recommendations are made based on various information to improve the quality of recommendations. Dynamic thresholds are introduced to make recommendations more intelligent and dynamic. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is a flowchart of a course recommendation method based on big data provided by an embodiment of the present invention.

[0046] Figure 2 It is a schematic diagram of a course recommendation device based on big data provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0047] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0048] like Figure 1 As shown, an embodiment of the present invention proposes a course recommendation method based on big data, the method comprising:

[0049] Step 11, obtaining and analyzing the learning behavior data of the target users;

[0050] Step 12, constructing an interaction matrix between the user and the course based on the learning behavior data;

[0051] Step 13, calculating the similarity between courses or between users according to the interaction matrix to obtain a similarity calculation result;

[0052] Step 14, generating the target user's preference for courses according to the similarity calculation result;

[0053] Step 15, generating a course recommendation list according to the target user's learning needs, the target user's preferences for courses, the course information in the course library, and the dynamic threshold;

[0054] Step 16: Send the course recommendation list to the target user.

[0055] In an embodiment of the present invention, by acquiring and analyzing the learning behavior data of the target user, the user's historical learning data can be collected, the user's learning interests and preferences can be understood, and the user's learning rules can be discovered by analyzing the learning data. An interaction matrix between users and courses is constructed, and the matrix intuitively represents the interactive relationship between users and courses. By calculating the similarity between courses or users, other users with similar interests to the target user can be found, and new courses similar to the courses liked by the target user can be found. The target user's preference for courses is generated, and the user's interests and historical learning behaviors are integrated to accurately express the user's course preferences, and recommendations are made based on various information to improve the quality of recommendations. Dynamic thresholds are introduced to make recommendations more intelligent and dynamic.

[0056] In a preferred embodiment of the present invention, the above step 11 may include:

[0057] Step 111, obtaining all learning behavior data of the target user on the online learning platform;

[0058] Step 112, preprocessing the learning behavior data to obtain preprocessed data;

[0059] Step 113: Analyze the preprocessed data to obtain the user's learning pattern and preference, as well as key factors affecting the learning effect.

[0060] In the embodiment of the present invention, as comprehensive user learning behavior data as possible is collected, the learning behavior data is preprocessed, invalid and abnormal data is cleaned, and the data is processed into a structured format to improve the efficiency and quality of subsequent analysis; the user's learning time, learning content, etc. can be analyzed to discover learning rules, analyze the user's preferences for different courses and teachers, establish a user interest model, analyze the key factors affecting learning effects, and provide improvement suggestions for the platform. Through the above steps, valuable information can be extracted from the original user behavior data, and an in-depth understanding of user learning can be established. Preprocessing improves analysis efficiency, and the analysis results can describe the user's learning characteristics from multiple dimensions, making the recommendation more accurate and suitable for user needs.

[0061] In another preferred embodiment of the present invention, the above step 111 may include:

[0062] Step 1111, according to Get the user's behavior log, where: is the behavior log of the ith user at timestamp j, is a function to obtain the user's behavior log. represents the i-th user, represents timestamp j;

[0063] Step 1112, according to Convert the behavior log into a matrix form, where represents the data of type k behavior performed by the i-th user at timestamp j, is a function that converts the behavior log into a matrix form, represents k types of learning behaviors.

[0064] In the embodiment of the present invention, by obtaining the user's behavior log, all the user's behavior data on the online learning platform can be obtained, including when they are online, their learning activities, etc. The behavior log is converted into a matrix form. The data in the matrix form is easier to be processed and analyzed by computer programs. By converting complex behavior logs into matrix data, data analysis and model training can be performed more effectively.

[0065] In another preferred embodiment of the present invention, the above step 112 may include:

[0066] Step 1121, according to The learning behavior data is processed with missing values, wherein: Represents the data of the i-th user at timestamp j after processing missing values; is a function that handles missing values. is the original data of the sth attribute, and the parameter for handling missing values;

[0067] Step 1122, according to Handling outliers, where represents the data of the i-th user at timestamp j after processing outliers, is a function for handling outliers, is the data after processing missing values. Represents the processed data, that is, the result after outlier processing;

[0068] Step 1123, according to Handling repeated values, where, where, represents the data of the i-th user at timestamp j after processing duplicate values, is a function that handles repeated values, The processed data, i, j and k They represent user ID, timestamp and type of learning behavior respectively; ，oand r Represent the processes of handling missing values, outliers, and duplicate values, respectively.

[0069] In another preferred embodiment of the present invention, the pre-processed data is analyzed to obtain the user's learning pattern and preference, as well as key factors affecting the learning effect, which may include:

[0070] Step 1131, according to Calculating the user's learning time includes calculating the user's learning time in each time period, where: represents the total learning time of the i-th user, is the learning time of the ith user at timestamp j;

[0071] Step 1132, according to Analyze the user's learning preferences, including analyzing the user's preferred learning method, for example, whether they prefer to listen to lectures or do exercises. represents the preference of the i-th user for the k-th learning method, is the number of times the i-th user uses the k-th learning method. Represents the total number of times the i-th user uses all learning methods.

[0072] Step 1133, according to E i = f ( T i ，P i1 ，P i2 ，…，P in ) Analyze the key factors that affect learning outcomes, including the impact of user learning time, learning methods, and other factors on learning outcomes. E i represents the learning effect of the i-th user, f ( ) is a function that affects the learning effect, T i is the learning time of the ith user, P i1 ，P i2 ，…，P in is the preference of the i-th user for n learning methods.

[0073] In a preferred embodiment of the present invention, in the interaction matrix, rows represent users, columns represent courses, and the value of each cell is the interaction value of the corresponding user for a specific course; the above step 12 may include:

[0074] Step 121, traverse all learning behavior data in dynamic sorting in sequence, and extract all corresponding relationships between users and courses;

[0075] Step 122, based on the correspondence between all users and courses, obtain all user behavior records for the course;

[0076] Step 123, calculating the fusion interaction value based on all the user's behavior records on the course;

[0077] Step 124, constructing an interaction matrix according to the fused interaction value;

[0078] Step 125 , processing missing values ​​in the interaction matrix and standardizing the interaction matrix to obtain a standardized interaction matrix.

[0079] In an embodiment of the present invention, in step 121, by traversing all learning behavior data and extracting all user-course correspondences, a complete user-course interaction data set can be obtained. In step 122, it helps to gain an in-depth understanding of the user's specific behavior for each course, such as when the user started learning the course, how long he studied, whether he completed the course, etc. In step 123, it helps to quantify the user's degree of interaction with each course. For example, a comprehensive interaction value can be calculated by combining multiple behavioral data such as the number of user clicks, dwell time, completion status, etc. In step 124, it helps to organize user-course interaction data in a structured manner. In this matrix, rows represent users, columns represent courses, and the value of each cell is the interaction value of the corresponding user for a specific course. In step 125, it helps to deal with the incompleteness and heterogeneity of the data, allowing us to compare different interaction values ​​on the same scale.

[0080] In another preferred embodiment of the present invention, all learning behavior data in dynamic sorting are sequentially traversed to extract all corresponding relationships between users and courses, which may include:

[0081] Step 1211, dynamically sort all learning behavior data, and the dynamic sorting is performed based on the timestamp, user activity, and course popularity dimensions. For example, if the timestamp is used for sorting, the learning behavior data set can be sorted in the order of occurrence time;

[0082] Step 1212, traverse the sorted learning behavior data, and create a loop to traverse the learning behavior data set from beginning to end S Each record in;

[0083] Step 1213: for each learning behavior data, extract the user ID and courses ID, in the process of traversal, for the data set S Each learning behavior record in S i .

[0084] In the embodiment of the present invention, in step 1211, it is helpful to discover the development trend and pattern of learning behavior, for example, whether the user's learning activities have a specific time pattern, whether certain courses have attracted the attention of the majority of users in a specific time period, etc.; using multi-dimensional sorting (such as timestamp, user activity, course popularity, etc.) can understand learning behavior from multiple perspectives, provide more comprehensive learning dynamic information, and help to better understand the user's learning behavior and the popularity of the course. In step 1212, by traversing the sorted data, each learning behavior record can be processed one by one, which can ensure that we will not miss any important learning behavior information. The traversal process can be combined with other processing steps (such as data cleaning, feature extraction, etc.) to improve the efficiency and effect of data processing. In step 1213, extracting the user ID and course ID can establish a corresponding relationship between the user and the course. By extracting the user ID and course ID, in-depth data analysis and mining can be performed for specific users or specific courses, for example, analyzing the user's learning habits, the popularity of the course, etc.

[0085] In another preferred embodiment of the present invention, constructing an interaction matrix according to the fused interaction value may include:

[0086] Step 1241, construct a user and course fusion interaction value table I1, I1 is a set of triples, set;

[0087] Step 1242, encode all user IDs and course IDs, and map each user ID and course ID to a continuous integer, so that it is convenient to construct the interaction matrix. Assuming there are m users and n courses, then two mapping functions can be obtained, which can map the user ID and course ID to the range of [1, m] and [1, n], where each element is a correspondence between a user ID, a course ID and a fused interaction value.

[0088] Step 1243, construct an m1×n1 interaction matrix M2, where each element M2[i][j] in the matrix represents the fused interaction value of user i for course j. If user i does not participate in course j, the value of M2[i][j] can be set to 0 or a specific missing value.

[0089] Step 1244, use the interaction matrix M2 for subsequent analysis and modeling. Each element in the interaction matrix M2 represents the integrated interaction value of a user for a course, which can quantify the user's participation and enthusiasm for the course.

[0090] In an embodiment of the present invention, multi-dimensional interactive behavior data can be gathered together. Constructing an interaction matrix is ​​to store the information of the user-course-fusion interaction value table in a more intuitive and easier to process manner. Through the interaction matrix M2, the fusion interaction value of any user for any course can be easily obtained, thereby performing more in-depth user behavior analysis and course recommendation tasks.

[0091] In a preferred embodiment of the present invention, the above step 13 may include:

[0092] Step 131, constructing a user vector and a course vector respectively according to the interaction matrix;

[0093] Step 132, calculating the average values ​​of interaction values ​​of all users and the average values ​​of interaction values ​​of all courses respectively according to the user vector and the course vector;

[0094] Step 133, based on the average interaction values ​​of all users and the average interaction values ​​of all courses, as well as the dynamic user weights, respectively calculate the similarity between every two users and the similarity between every two courses.

[0095] In an embodiment of the present invention, based on the interaction matrix, user vectors and course vectors are constructed respectively, and the interaction information between users and courses can be quantified and converted into vector form. Based on the user vector and course vector, the average value of the interaction value of all users and the average value of the interaction value of all courses are calculated respectively, and a global perspective can be obtained to understand the general interaction behavior of users and the general popularity of courses. This information can be used as a reference point to evaluate whether a specific user or course is active or popular. Based on the average value of the interaction value of all users and the average value of the interaction value of all courses, as well as the dynamic user weight, the similarity between every two users and the similarity between every two courses are calculated respectively, which can help establish a similarity matrix between users and between courses. By introducing dynamic user weights, the current interests and behavior patterns of users can be more accurately reflected, thereby improving the accuracy and personalization of recommendations.

[0096] In a preferred embodiment of the present invention, the similarity between every two users and the similarity between every two courses are calculated respectively according to the average value of the interaction values ​​of all users and the average value of the interaction values ​​of all courses, as well as the dynamic user weights, which may include:

[0097] Step 1331, according to Calculate the average interaction value of all users m u ,in, is the total number of users; Is a user uWeight w u The calculation formula of weight w u Reflects the user's activity; Is a user u With Courses i The sum of the interaction values, which are defined according to specific application scenarios, such as user ratings of courses, number of clicks, number of completed courses, etc.

[0098] Step 1332, according to Calculate the average of the interaction values ​​for all courses m c ;in, is the total number of courses; is the weight of the course, It is the sum of the course values, which can be defined according to specific application scenarios, such as the number of times a course is clicked by users, the score of a course, etc.

[0099] Step 1333, according to Calculate unnormalized weights ; Based on the unstandardized weights , calculate the sum of all users' unstandardized weights T1, where, ; Calculate user t based on the sum of all users’ unstandardized weights T1 i The standardized weight Z(i) ,in, According to the user i The standardized weight Z(i) , and the sum of all users’ unnormalized weights T1, calculate user i The final weight F (i) ,in, ,in, Indicates user i activity, is a parameter that controls the weight distribution, is the unstandardized weight of user i, T1 is the sum of the unstandardized weights of all users, Z(i) is the standardized weight of user i, F(i) is the final weight of user i, Indicates the total number or range of users; j Indicates the index or number of the user, from 1 to , used to distinguish different users; and Respectively represent the minimum and maximum values ​​of the unnormalized weights of all users;

[0100] Step 1334, according to Calculate the similarity between every two users ,in, and Respectively represent users i and users l exist j The value of the item interaction.

[0101] In the embodiment of the present invention, by calculating the similarity between users, it is possible to more accurately predict what content the user may be interested in, thereby improving the accuracy of the recommendation system; by calculating the final weight F(i) of the user, it is possible to better handle the situations of active users and inactive users; by calculating the average value of the interaction value of all users, m u The average of the interaction values ​​of all courses m c , the interaction values ​​of users and courses are balanced, so that the effect of the recommendation system is not affected by the extremely high or low interaction values ​​of some users or courses. By calculating the standardized weight Z(i) of the user, the weight of the user is standardized, so that the effect of the recommendation system is not affected by the extremely high or low weight of some users.

[0102] In a preferred embodiment of the present invention, the above step 14 may include:

[0103] Step 141, based on the similarity between every two users and the similarity between every two courses, extract the N users with the highest similarity to each target user and the M courses most similar to the courses rated by the target user;

[0104] Step 142 , based on the N users with the highest similarity to each target user and the M courses that are most similar to the courses rated by the target user, generate the target user's preference for courses.

[0105] In an embodiment of the present invention, taking both the similarity between users and the similarity between courses into consideration can better reflect the specific interests of each user, thereby achieving personalized recommendations; by extracting the N users most similar to the target user and the M courses most similar to the courses rated by the target user, the content that the target user may be interested in can be predicted more accurately, thereby improving the accuracy of the recommendation system; considering only the N users most similar to the target user and the M courses most similar to the courses rated by the target user can greatly reduce the amount of calculation and improve the efficiency of recommendations, and not only the similarity of users but also the similarity of courses is considered, so that more diverse recommendation results can be provided to meet the different needs of users.

[0106] In a preferred embodiment of the present invention, the above step 15 may include:

[0107] Step 151:u and courses i The original feature vector of X u and Y i Through deep neural network NN u and NN i Convert to feature vector U u and I i ,in, U u =NN u (X u )=σ(W u X u +b u ) I i =NN i (Y i )=σ(W i Y i +b i ) ,in, W u and W i is the weight matrix, b u and b i is the bias vector, s is the activation function;

[0108] Step 152, according to Calculating prediction scores ,in, is a dot product operation, representing the converted user feature vector U u and the course feature vector I i The similarity between b u is the user bias term, representing the user u Overall rating tendency; is the global average rating, which represents the average rating of all users on all courses; is the course bias term, representing the course i Overall popularity ofT Used to represent the transpose of a matrix;

[0109] Step 153, according to the user Adjust dynamic thresholds based on historical behavior and feedback i u ,in, ,in, Is a user u The historical average rating of α is a constant;

[0110] Step 154, calculate all courses i Prediction score of , and obtain the prediction score ≥ i u of N A list of courses is provided as a recommendation.

[0111] In an embodiment of the present invention, the deep neural network is used to transform the feature vector to capture the deep characteristics of users and courses. Combining the bias items and global average scores of users and courses can more accurately predict the user's score for the course, thereby improving the recommendation accuracy. By adjusting the dynamic threshold, the recommendation results can be adjusted in real time according to the user's historical behavior and feedback, making the recommendation system more dynamic. Only recommending courses with predicted scores exceeding the dynamic threshold can reduce invalid recommendations and improve recommendation efficiency.

[0112] like Figure 2 As shown, an embodiment of the present invention further provides a course recommendation device 20 based on big data, comprising:

[0113] The acquisition module 21 is used to acquire and analyze the learning behavior data of the target user; construct an interaction matrix between the user and the course according to the learning behavior data; calculate the similarity between courses or between users according to the interaction matrix to obtain a similarity calculation result;

[0114] The processing module 22 is used to generate the target user's preference for courses based on the similarity calculation result; generate a course recommendation list based on the target user's learning needs, the target user's preference for courses, the course information in the course library and the dynamic threshold; and send the course recommendation list to the target user.

[0115] Optionally, obtain and analyze the target user's learning behavior data, including:

[0116] Obtain all learning behavior data of target users on the online learning platform;

[0117] Preprocessing the learning behavior data to obtain preprocessed data;

[0118] The preprocessed data is analyzed to obtain the user's learning patterns and preferences, as well as key factors that affect learning outcomes.

[0119] Optionally, in the interaction matrix, rows represent users, columns represent courses, and the value of each cell is the interaction value of the corresponding user for a specific course.

[0120] Optionally, an interaction matrix between the user and the course is constructed based on the learning behavior data, including:

[0121] Traverse all learning behavior data in dynamic order and extract all corresponding relationships between users and courses;

[0122] According to the correspondence between all users and courses, obtain all user behavior records on the course;

[0123] Calculate the fusion interaction value based on all the user's course behavior records;

[0124] constructing an interaction matrix according to the fused interaction values;

[0125] The missing values ​​in the interaction matrix are processed, and the interaction matrix is ​​standardized to obtain a standardized interaction matrix.

[0126] Optionally, similarities between courses or between users are calculated according to the interaction matrix to obtain similarity calculation results, including:

[0127] According to the interaction matrix, construct user vector and course vector respectively;

[0128] According to the user vector and the course vector, respectively calculating the average value of the interaction values ​​of all users and the average value of the interaction values ​​of all courses;

[0129] Based on the average interaction values ​​of all users and the average interaction values ​​of all courses, as well as the dynamic user weights, the similarity between every two users and the similarity between every two courses are calculated respectively.

[0130] Optionally, generating the target user's preference for courses according to the similarity calculation result includes:

[0131] According to the similarity between every two users and every two courses, extract the N users with the highest similarity to each target user and the M courses most similar to the courses rated by the target user;

[0132] Based on the N users with the highest similarity to each target user and the M courses most similar to the courses rated by the target user, the target user's preference for courses is generated.

[0133] Optionally, a course recommendation list is generated based on the target user's learning needs, the target user's preferences for courses, the course information in the course library, and the dynamic threshold, including:

[0134] Separate users u and courses i The original feature vector of X u and Y i Through deep neural network NN u and NN i Convert to feature vector U u and I i ,in, U u =NN u (X u )=σ(W u X u +b u ) I i =NN i (Y i )=σ(W i Y i +b i ) ,in, W u and W i is the weight matrix, b u and b i is the bias vector, s is the activation function;

[0135] according to Calculating prediction scores ,in, is a dot product operation, representing the converted user feature vector U u and the course feature vector I i The similarity between b u is the user bias term, representing the user u Overall rating tendency; is the global average rating, which represents the average rating of all users on all courses; is the course bias term, representing the course i Overall popularity of T Used to represent the transpose of a matrix;

[0136] According to user Adjust dynamic thresholds based on historical behavior and feedback i u ,in, ,in, Is a user u The historical average rating of α is a constant;

[0137] Count all courses i Prediction score of , and obtain the prediction score ≥ i u of N A list of courses is provided as a recommendation.

[0138] It should be noted that the device is a device corresponding to the above method, and all implementation methods in the above method embodiment are applicable to this embodiment and can achieve the same technical effect.

[0139] The embodiment of the present invention further provides a computing device, comprising: a processor, a memory storing a computer program, wherein when the computer program is executed by the processor, the method described above is executed. All implementations in the above method embodiment are applicable to this embodiment and can achieve the same technical effect.

[0140] The embodiment of the present invention also provides a computer-readable storage medium storing instructions, which, when executed on a computer, enable the computer to execute the method described above. All implementations in the above method embodiment are applicable to this embodiment and can achieve the same technical effect.

[0141] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0142] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0143] In the embodiments provided by the present invention, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0144] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0145] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0146] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, ROM, RAM, magnetic disks, or optical disks.

[0147] In addition, it should be pointed out that in the apparatus and method of the present invention, it is obvious that each component or each step can be decomposed and / or recombined. These decompositions and / or recombinations should be regarded as equivalent schemes of the present invention. Moreover, the steps of performing the above series of processing can naturally be performed in chronological order according to the order of description, but it is not necessary to perform them in chronological order, and some steps can be performed in parallel or independently of each other. For those of ordinary skill in the art, it is understandable that all or any steps or components of the method and apparatus of the present invention can be implemented in hardware, firmware, software or a combination thereof in any computing device (including processors, storage media, etc.) or a network of computing devices, which can be achieved by those of ordinary skill in the art using their basic programming skills after reading the description of the present invention.

[0148] Therefore, the purpose of the present invention can also be achieved by running a program or a group of programs on any computing device. The computing device can be a well-known general device. Therefore, the purpose of the present invention can also be achieved by simply providing a program product containing a program code that implements the method or device. That is to say, such a program product also constitutes the present invention, and the storage medium storing such a program product also constitutes the present invention. Obviously, the storage medium can be any well-known storage medium or any storage medium developed in the future. It should also be pointed out that in the device and method of the present invention, it is obvious that each component or each step can be decomposed and / or recombined. These decompositions and / or recombinations should be regarded as equivalent schemes of the present invention. In addition, the steps of performing the above-mentioned series of processing can naturally be performed in chronological order according to the order of description, but it is not necessary to perform them in chronological order. Some steps can be performed in parallel or independently of each other.

[0149] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A course recommendation method based on big data, characterized in that: The method comprises: Obtain and analyze target users’ learning behavior data; constructing an interaction matrix between the user and the course based on the learning behavior data; Calculate the similarity between courses or between users according to the interaction matrix to obtain a similarity calculation result; Generate the target user's preference for the course according to the similarity calculation result; Generate a course recommendation list based on the target user's learning needs, the target user's preferences for courses, the course information in the course library, and dynamic thresholds; Sending the course recommendation list to a target user; Obtain and analyze the learning behavior data of target users, including: Obtain all learning behavior data of target users on the online learning platform; Preprocessing the learning behavior data to obtain preprocessed data; Analyzing the preprocessed data to obtain the user's learning patterns and preferences, as well as key factors affecting learning effects; In the interaction matrix, rows represent users, columns represent courses, and the value of each cell is the interaction value of the corresponding user for a specific course; According to the learning behavior data, an interaction matrix between the user and the course is constructed, including: Traverse all learning behavior data in dynamic order and extract all corresponding relationships between users and courses; According to the correspondence between all users and courses, obtain all user behavior records on the course; Calculate the fusion interaction value based on all the user's course behavior records; constructing an interaction matrix according to the fused interaction values; Processing missing values ​​in the interaction matrix and standardizing the interaction matrix to obtain a standardized interaction matrix; According to the interaction matrix, similarities between courses or between users are calculated to obtain similarity calculation results, including: According to the interaction matrix, construct user vector and course vector respectively; According to the user vector and the course vector, respectively calculating the average value of the interaction values ​​of all users and the average value of the interaction values ​​of all courses; Based on the average interaction values ​​of all users and the average interaction values ​​of all courses, as well as the dynamic user weights, the similarity between every two users and the similarity between every two courses are calculated respectively; According to the similarity calculation result, the target user's preference for the course is generated, including: According to the similarity between every two users and every two courses, extract the N users with the highest similarity to each target user and the M courses most similar to the courses rated by the target user; Generate the target user's preference for courses based on the N users with the highest similarity to each target user and the M courses most similar to the courses rated by the target user; Generate a course recommendation list based on the target user's learning needs, the target user's preferences for courses, the course information in the course library, and dynamic thresholds, including: Separate users u and courses i The original feature vector of X u and Y i Through deep neural network NN u and NN i Convert to feature vector U u and I i ,in, U u =NN u (X u )=σ(W u X u +b u ) I i =NN i (Y i )=σ(W i Y i +b i ) ,in, W u and W i is the weight matrix, b u and b i is the bias vector, σ is the activation function; according to Calculating prediction scores ,in, is a dot product operation, representing the converted user feature vector U u and the course feature vector I i The similarity between b u is the user bias term, representing the user u Overall rating tendency; is the global average rating, which represents the average rating of all users on all courses; is the course bias term, representing the course i Overall popularity of T Used to represent the transpose of a matrix; According to user Adjust dynamic thresholds based on historical behavior and feedback θ u ,in, ,in, Is a user u The historical average rating of α is a constant; Count all courses i Prediction score of , and obtain the prediction score ≥ θ u of N courses as a recommended list; The pre-processed data is analyzed to obtain the user's learning patterns and preferences, as well as key factors affecting the learning effect, including: according to Calculating the user's learning time includes calculating the user's learning time in each time period, where: represents the total learning time of the i-th user, is the learning time of the ith user at timestamp j; according to Analyze the user's learning preferences, including analyzing the user's preferred learning method, for example, whether they prefer to listen to lectures or do exercises. represents the preference of the i-th user for the k-th learning method, is the number of times the i-th user uses the k-th learning method. represents the total number of times the i-th user uses all learning methods; According to the fused interaction value, an interaction matrix is ​​constructed, including: Construct a fusion interaction value table I1 between users and courses. I1 is a set of triples. Encode all user IDs and course IDs, and map each user ID and course ID to a continuous integer, so that it is convenient to construct the interaction matrix. Assuming there are m users and n courses, then two mapping functions can be obtained, which can map user IDs and course IDs to the range of [1, m] and [1, n], where each element is a corresponding relationship between a user ID, a course ID and a fused interaction value; Construct an m1×n1 interaction matrix M2. Each element M2[i][j] in the matrix represents the fusion interaction value of user i for course j. If user i does not participate in course j, the value of M2[i][j] can be set to 0 or a specific missing value. The interaction matrix M2 is used for subsequent analysis and modeling. Each element in the interaction matrix M2 represents the integrated interaction value of a user for a course, which can quantify the user's participation and enthusiasm for the course; Based on the average interaction values ​​of all users and the average interaction values ​​of all courses, as well as the dynamic user weights, the similarity between every two users and the similarity between every two courses are calculated, including: according to Calculate the average interaction value of all users μ u ,in, is the total number of users; Is a user u Weight w u The calculation formula of weight w u Reflects the user's activity; Is a user u With Courses i The sum of the interaction values ​​of ; according to Calculate the average of the interaction values ​​for all courses μ c ;in, is the total number of courses; is the weight of the course, is the sum of the course values; according to Calculate unnormalized weights ; Based on the unstandardized weights , calculate the sum of all users' unstandardized weights T1, where, ; Calculate user t based on the sum of all users’ unstandardized weights T1 i The standardized weight Z(i) ,in, According to the user i The standardized weight Z(i) , and the sum of all users’ unnormalized weights T1, calculate user i The final weight F(i) ,in, ,in, Indicates user i activity, is a parameter that controls the weight distribution, is the unstandardized weight of user i, T1 is the sum of the unstandardized weights of all users, Z(i) is the standardized weight of user i, F(i) is the final weight of user i, Indicates the total number or range of users; j Indicates the index or number of the user, from 1 to , used to distinguish different users; and Respectively represent the minimum and maximum values ​​of the unnormalized weights of all users; according to Calculate the similarity between every two users ,in, and Respectively represent users i and users l exist j The value of the item interaction.

2. A course recommendation device based on big data, characterized in that: Applied to the method as claimed in claim 1, comprising: An acquisition module is used to acquire and analyze the learning behavior data of the target user; based on the learning behavior data, an interaction matrix between the user and the course is constructed; based on the interaction matrix, similarities between courses or between users are calculated to obtain similarity calculation results; A processing module is used to generate the target user's preference for courses based on the similarity calculation result; generate a course recommendation list based on the target user's learning needs, the target user's preference for courses, the course information in the course library and the dynamic threshold; and send the course recommendation list to the target user.

3. A computing device, characterized in that: include: one or more processors; A storage device, used for storing one or more programs, when the one or more programs are executed by the one or more processors, the one or more processors implement the method as claimed in claim 1.

4. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a program, which implements the method according to claim 1 when executed by a processor.

Images