CN116187852A - Online course recommendation method based on community association and behavior feature learning - Google Patents
Online course recommendation method based on community association and behavior feature learning Download PDFInfo
- Publication number
- CN116187852A CN116187852A CN202310164787.4A CN202310164787A CN116187852A CN 116187852 A CN116187852 A CN 116187852A CN 202310164787 A CN202310164787 A CN 202310164787A CN 116187852 A CN116187852 A CN 116187852A
- Authority
- CN
- China
- Prior art keywords
- user
- course
- node
- representing
- feature
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 230000006399 behavior Effects 0.000 title claims abstract description 81
- 238000000034 method Methods 0.000 title claims abstract description 39
- 239000011159 matrix material Substances 0.000 claims abstract description 114
- 230000006870 function Effects 0.000 claims abstract description 23
- 230000002776 aggregation Effects 0.000 claims abstract description 9
- 238000004220 aggregation Methods 0.000 claims abstract description 9
- 230000004931 aggregating effect Effects 0.000 claims abstract description 8
- 239000013598 vector Substances 0.000 claims description 44
- 238000004364 calculation method Methods 0.000 claims description 20
- 238000000605 extraction Methods 0.000 claims description 15
- 238000012549 training Methods 0.000 claims description 14
- 238000010586 diagram Methods 0.000 claims description 11
- 238000006243 chemical reaction Methods 0.000 claims description 10
- 230000004913 activation Effects 0.000 claims description 8
- 230000004927 fusion Effects 0.000 claims description 5
- 230000008569 process Effects 0.000 claims description 5
- 238000013507 mapping Methods 0.000 claims description 4
- 238000010606 normalization Methods 0.000 claims description 4
- 230000003542 behavioural effect Effects 0.000 claims description 3
- 230000003993 interaction Effects 0.000 description 6
- 238000004422 calculation algorithm Methods 0.000 description 2
- 238000007418 data mining Methods 0.000 description 2
- 230000007774 longterm Effects 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 239000002245 particle Substances 0.000 description 2
- 230000002787 reinforcement Effects 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000007787 long-term memory Effects 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000005295 random walk Methods 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 238000005728 strengthening Methods 0.000 description 1
- 239000000758 substrate Substances 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
- G06Q10/06393—Score-carding, benchmarking or key performance indicator [KPI] analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9536—Search customisation based on social or collaborative filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/10—Services
- G06Q50/20—Education
- G06Q50/205—Education administration or guidance
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Human Resources & Organizations (AREA)
- Databases & Information Systems (AREA)
- Theoretical Computer Science (AREA)
- Educational Administration (AREA)
- Strategic Management (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Tourism & Hospitality (AREA)
- Economics (AREA)
- Entrepreneurship & Innovation (AREA)
- Development Economics (AREA)
- General Engineering & Computer Science (AREA)
- General Business, Economics & Management (AREA)
- Marketing (AREA)
- Data Mining & Analysis (AREA)
- Educational Technology (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Primary Health Care (AREA)
- Quality & Reliability (AREA)
- Game Theory and Decision Science (AREA)
- Operations Research (AREA)
- Electrically Operated Instructional Devices (AREA)
Abstract
The invention discloses an online course recommendation method based on community association and behavior feature learning, which is characterized in that the community association learning in a module is used for calculating the node association degree based on communities, the node association degree after the semantic information of added nodes is enhanced, and the feature aggregation is carried out to obtain the entity features on each element path. Calculating importance of different element paths, aggregating entity features obtained by the different element paths according to the element paths to obtain entity association features, inputting two groups of association matrix sets and the entity features into a convolution layer to obtain entity feature matrices, inputting the entity feature matrices into a linear layer to obtain entity behavior features, splicing the entity association features and the behavior features, and inputting the entity association features and the behavior features into a scoring prediction function to obtain a course recommendation list. According to the method, the potential information existing in the original data of the online education is deeply mined, and more comprehensive characteristic representation is obtained, so that the accuracy of course recommendation is improved.
Description
Technical Field
The invention belongs to the field of online education course recommendation, and particularly relates to an online course recommendation method based on community association and behavior feature learning.
Background
In recent years, large-scale online open courses (Massive open online courses, MOOCs) have evolved very rapidly, and traditional offline education has been significantly impacted in the case of a world epidemic pandemic. On-line courses attract a great deal of teachers and students to learn by virtue of the convenience, the possession of high-quality teaching resources and the like. There are many well-known online education platforms in China, such as museums, classrooms, etc., and these high-quality platforms provide a great deal of learning resources for teachers and students.
Although online education is gradually developed into a mainstream educational mode, the efficiency of students' learning through an online education platform is widely questioned. Compared with traditional offline education, the online education course completion rate is low. The overall completion rate of the student's online course is statistically less than 5%, reflecting the lack of interest of the student in the course. Therefore, how to deeply mine the preferences of students so as to provide the students with courses of interest to them is one of the key points to promote the development of online education.
The Chinese patent application with publication number of CN113435685A discloses a course recommendation method of a hierarchical attention deep learning model, which calculates preprocessed user learning behavior data to obtain user vector representation, and extracts course knowledge points as course vector representation by using a convolution network. And capturing interactions of different historical session interests of the user by using a long-term and short-term memory algorithm to obtain interest vectors, sending the interest vectors to an attention network to obtain long-term interest vectors, sending the long-term interest vectors and the user vectors to the attention network to obtain mixed interest representations, and finally carrying out inner products with course representations to obtain recommended results. The method considers the time variation of the user interests, but the interest degree of the user corresponding to different behaviors between the user and the course is different, so the method does not fully utilize the data, and the final prediction performance is more general.
Chinese patent application publication No. CN114154839 discloses a course recommendation method based on online education platform data. The method obtains interaction data between students and courses, between students and attribute data of the students from an online education platform. And obtaining a characteristic representation and scoring matrix of the course through interaction data of the students and the course, establishing an adjacency matrix through interaction data of the students, and inputting attribute data of the adjacency matrix and the students into the reinforcement learning model to obtain the characteristic representation of the students. And obtaining a course recommendation list for each student by the scoring matrix, the course characteristic representation and the user characteristic representation. Although the method considers the interaction between students, the degree of interaction between students is not considered to be different, so that the obtained characteristic representation is not accurate enough, and the final recommended performance is general.
The prior art has the common problems that data mining is not deep enough, community information of users and courses is not utilized effectively, behavior characteristics are not fully mined, final recommendation is not accurate enough, and courses of interest to the users cannot be provided accurately.
Disclosure of Invention
The online course recommendation method based on community association and behavior feature learning is used for solving the problems that data mining is not deep enough, community information is not utilized effectively, behavior features are not fully mined, recommendation accuracy is low and the like in the prior art.
In order to achieve the above purpose, the technical scheme of the application is as follows:
an online course recommendation method based on community association and behavior feature learning, comprising:
obtaining course context information, user personal information and user behavior logs from an online education platform to generate a user element path adjacency matrix set A U Course element path adjacency matrix set A C User initial feature H U Course context feature H C User community set C U Course community set C C ;
Adjoining the element path to the matrix set A U 、A C User initial feature H U Course context feature H C With community collection C U 、C C Input to the association feature extraction module which is built and trained to obtain association features F 'of the user respectively' U Associated features F 'with course' C ;
Building a bipartite graph set B according to the user behavior log, and respectively integrating the initial characteristics H of the user U Course context feature H C Inputting the training result and the bipartite graph set B into a behavior feature extraction module which is constructed and trained to obtain behavior features F', of a user and a course respectively U and F″C ;
Associating the user with the course feature F' U 、F' C Behavior characteristics F U 、F″ C Respectively carrying out feature fusion to respectively obtain final feature representations E of users and courses U and EC ;
Representing E according to final characteristics U and EC And calculating a scoring matrix of the user and the courses, and recommending the courses with the scores being the front to the user.
Further, the course context information, the user personal information and the user behavior log are obtained from the online education platform to generate a user element path adjacency matrix set A U Course element path adjacency matrix set A C User initial feature H U Course context feature H C User community set C U Course community set C C Comprising:
step 1.1, acquiring course context information comprising course names, course introduction and belonging categories from an online education platform, creating a corpus according to the acquired course introduction, training a Word2vec model on the corpus, calculating Word vectors corresponding to the course names through the Word2vec model, and combining the obtained Word vectors to generate course context characteristics H C ;
Step 1.2, slaveThe online education platform acquires personal information of the user including gender, age and academic, and inputs the personal information into the single-heat encoder to obtain initial characteristics H of the user U ;
Step 1.3, constructing a meta-path adjacency matrix set A of the users according to the potential relation among the users U Constructing a meta-path adjacency matrix set A of courses according to potential relations among courses C :
A U ={MP 1 MP 2 ......MP z };
A C ={MP′ 1 MP′ 2 ......MP′ t };
Wherein, the number of user element paths is z, the number of course element paths is t, and MP z Representing the z-th user element path, MP' t Representing a t-th course element path;
step 1.4, according to the user element path adjacency matrix set A U Creating a user association diagram for any user node u i Constructing communities of the nodeRepresenting user node u within the community i Is a node of all k-th order neighbors,representing u in the community i For u for all k-th order neighbor nodes of (2) i Weights of (2); node c for any course i Constructing a community of the node>Representing course node c within the community i Is +.>All k-th order neighbor nodes representing ci within the community are for c i Weights of (2); /> and />The calculation formula of (2) is as follows:
wherein I represents an identity matrix, c is a trainable parameter for dynamically controlling the size of the community,is a start vector of all zeros except the position corresponding to the center node, R ui Is an adjacency matrix of the community where the user node is located, +.>Is an adjacency matrix of communities where course nodes are located;
by the above, the community set C of all users can be obtained U Community set C with all courses C 。
Further, the element path is adjacent to the matrix set A U 、A C User initial feature H U Course context feature H C With community collection C U 、C C Input to the association feature extraction module which is built and trained to obtain association features F 'of the user' U Associated features F 'with course' C Comprising:
step 2.1, by calculating different user nodes u i And u is equal to j Similarity among communities to obtain relevance based on communitiesThe calculation formula is as follows:
wherein ,Vui And Vu j Respectively represent user nodes u i And u is equal to j Neighbor node set of u l Representing Vu i And Vu j Nodes in the intersection, u m Representing Vu i And Vu j The nodes in the union are connected in parallel,representing user node u i For user node u l Is used for the weight of the (c),representing user node u j For user node u l Weight of->Representing user node u i For user node u m Weight of->Representing user node u j For user node u m Weights of (2);
step 2.2, according to the initial characteristics H U Calculating different user nodes u i And u is equal to j Semantic similarity betweenThe calculation formula is as follows:
wherein ,and->Respectively are provided withRepresenting user node u i With user node u j Semantic information of->The method comprises the steps of converting a high-dimensional vector into a real number, wherein W is a conversion matrix for converting node characteristics into the same space;
step 2.3, adding the semantic similarity of the nodes on the basis of the node association degree based on the community to obtain the enhanced node association degree
Where softmax (·) represents the normalization function, mapping the input values to [0,1 ]]The numerical value of the two components is equal to the numerical value of the two components,and (3) withRespectively indicate +.> and />Normalized values were performed. Sigma represents an activation function, exp (·) represents an exponential function with the natural constant e as a base, and α (·) and β (·) represent a conversion function;
step 2.4, obtaining a user node u according to the enhanced node association degree and semantic information of the neighbor nodes i Feature vector under a certain element path:
Step 2.5, calculating that all user nodes are in a primary path MP z The feature vectors are spliced to obtain the user features on the meta-pathWherein n represents the number of user nodes;
step 2.6, calculating that all course nodes are in a meta-path MP' t The feature vectors are spliced to obtain course features on the meta-pathWherein m represents the number of course nodes;
step 2.7, calculating the user characteristics on each element path to obtain a user characteristic set
Step 2.8, calculating course characteristics on each element path to obtain a course characteristic set
Wherein U represents a set of user nodes, L u Representing trainable vectors, W u Representing a weight matrix, b u Representing a bias vector;
the aggregation is carried out according to the element paths, and user association characteristics under Z element paths are obtained:
Wherein C represents course node set, L c Representing trainable vectors, W c Representing a weight matrix, b c Representing a bias vector;
aggregating according to the meta-paths to obtain course association features under T meta-paths:
further, constructing a bipartite graph set B according to the user behavior log, and respectively integrating the initial characteristics H of the user U Course context feature H C Inputting the training result and the bipartite graph set B into a behavior feature extraction module which is constructed and trained to obtain behavior features F', of a user and a course respectively U and F″C Comprising:
step 3.1, constructing a user-course bipartite graph set B= { according to the user behavior logB 1 B 2 ......B r}, wherein ,Br Representing a user-course bipartite graph corresponding to the behavior type r;
step 3.2, for any bipartite graph, if a plurality of user entities are connected with one course entity, one superside exists to connect the plurality of user nodes; if a plurality of course entities are connected with a user entity, an overtravel exists to connect the course nodes; thereby, the user-course bipartite graph is converted into two isomorphic hypergraph sets G U and GC :
G U ={G U,global G U,1 ......G U,r };
G C ={G C,global G C,1 ......G C,r };
wherein GU,r =(U,E U,r ) Representing user hypergraph corresponding to the r-th behavior type, U representing user node set, E U,r Representing a set of r-type superedges. G C,r =(C,E C,r ) Representing a curriculum hypergraph corresponding to the r-th behavior type, E C,r Representing a set of r-type superedges. G U,global =(U,E U,1 ∪E U,2 ...∪E U,r ) Representing a global user hypergraph whose hyperedge set is defined by G U,1 ......G U,r Is formed by the union of the supersides of the above-mentioned two parts. G C,global =(U,E C,1 ∪E C,2 ...∪E C,r ) Representing global curriculum hypergraphs whose hyperedge sets are defined by G C,1 ......G C,r Is formed by the union of the supersides of the two groups;
step 3.4, constructing an association matrix corresponding to each hypergraph, wherein for each association matrix, if one entity node is associated with one hyperedge, the corresponding position in the association matrix is 1, and if the entity node is not associated with one hyperedge, the corresponding position is 0;
step 3.5, obtaining two groups of incidence matrix sets X through the step 3.4 U and XC :
X U ={X U,global X U,1 ......X U,r };
X C ={X C,gobal X C,1 ......X C,r };
Aggregating two sets of incidence matrices X U 、X C User initial feature H U Course context feature H C Input to the convolutional layer, the operation of layer 1+1 is as follows:
wherein sigma represents a nonlinear activation function,is a convolution calculation unit by which a convolution operation is performed. D (D) U,i Is a user node degree matrix representing a diagonal array of degrees of each node for a diagonal element, the degree of a node representing the number of edges associated with that node. B (B) U,i The super-edge degree matrix is a diagonal matrix which represents the degree that diagonal elements are each super-edge, and the super-edge degree represents the number of nodes contained in the super-edge; w (W) U Representing an identity matrix>Representing a matrix of learnable filters. Due to G U,global Comprising G U,1 ......G U,r So that the convolution calculation unit is additionally added with the aggregation information of the (E)To H U,global Incorporating global information contained into each H U,i The method comprises the steps of carrying out a first treatment on the surface of the G, according to the process of converting the bipartite graph into the hypergraph in the step 3.2 U,i One of the user nodes corresponds to G C,i Is added with ++>To build the corresponding relation between the above nodes and the superside, strengthen H U,i And H is c,i Similar attributes between;
step 3.6, according to step 3.5, obtaining the user characteristic matrix through l iterationsAnd course feature matrix->
Step 3.7, user characteristic matrixAnd course feature matrix->Inputting to the linear layer to obtain user behavior feature F U With course behavior feature F C The linear operation is as follows:
wherein OU With O C Is a conversion matrix, k U And k is equal to C Is a training parameter.
Further, the representation E according to the final characteristics U and EC Calculating scoring matrix for user and courseComprising:
representing E according to final characteristics U and EC The scoring matrix of the user and the course is calculated, and the calculation formula is as follows:
wherein vU ,v C Is a trainable parameter to ensure E U And E is connected with C In the same feature space; beta U ,β C Is an adjustment parameter, p U Is a potential feature matrix of the user, q C Is a potential feature matrix for the course.
The online course recommendation method based on community association and behavior feature learning is provided. Unlike traditional graphic neural network, the correlation between nodes can be captured only on node level, the method designs different element paths to obtain global information in different patterns, then constructs node communities for each element path, calculates the node correlation degree based on communities through community correlation learning, and adds semantic information of the nodes to strengthen the node correlation degree based on communities, and captures the correlation between the nodes together on community and node level. And then, aggregating the entity characteristics obtained by different meta-paths by using a meta-path attention mechanism to respectively obtain the associated characteristics of the user entity and the course entity. In addition, considering the influence of different user behaviors on a recommendation result, the method constructs the user hypergraph and the course hypergraph according to different behaviors of the entity, and obtains the behavior characteristics of the user and the course respectively through graph convolution operation. And finally, splicing the associated features and the behavior features, inputting the spliced features and the behavior features into a scoring prediction function to obtain a scoring matrix, and recommending courses with the scores in front to a user.
Drawings
FIG. 1 is a flow chart of an online course recommendation method based on community association and behavioral feature learning in the present application;
FIG. 2 is a schematic diagram of the extraction of relevant features of the present application;
FIG. 3 is a community association reinforcement schematic diagram of the present application;
fig. 4 is a schematic diagram of behavior feature extraction in the present application.
Detailed Description
In order to make the objects, technical solutions and advantages of the present application more apparent, the present application will be further described in detail with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the present application.
The online course recommendation method based on community association and behavior feature learning provided by the application, as shown in fig. 1, comprises the following steps:
step S1, acquiring course context information, user personal information and user behavior logs from an online education platform to generate a user element path adjacency matrix set A U Course element path adjacency matrix set A C User initial feature H U Course context feature H C User community set C U Course community set C C 。
When a user accesses the online educational platform, the background will record and retain a lot of raw data, including different behaviors between the user and the course (viewing course information, viewing video, answering questions), course context information (course name, course profile, category to which the user belongs), user personal information (gender, age, school). The application processes the original data and deep mines the data to obtain a user element path adjacency matrix set A U Course element path adjacency matrix set A C User initial feature H U Course context feature H C User community set C U Course community set C C And the like.
The course context information, the user personal information and the user behavior log are obtained from the online education platform to generate a user element path adjacency matrix set A U Course element path adjacency matrix set A C User initial feature H U Course context feature H C User community set C U Course community set C C The method specifically comprises the following steps:
step 1Firstly, acquiring course context information comprising course names, course introduction and belonging categories from an online education platform, creating a corpus according to the acquired course introduction, training a Word2vec model on the corpus, calculating Word vectors corresponding to the course names through the Word2vec model, and combining the obtained Word vectors to generate course context characteristics H C ;
Step 1.2, acquiring personal information including gender, age and academic from an online education platform, and inputting the personal information into a single-heat encoder to obtain initial characteristics H of a user U ;
Step 1.3, constructing a meta-path adjacency matrix set A of the users according to the potential relation among the users U Constructing a meta-path adjacency matrix set A of courses according to potential relations among courses C :
A U ={MP 1 MP 2 ......MP z };
A C ={MP′ 1 MP′ 2 ......MP′ t };
Wherein, the number of user element paths is z, the number of course element paths is t, and MP z Representing the z-th user element path, MP' t Representing a t-th course element path;
step 1.4, for any user node u i Constructing communities of the nodeRepresenting user node u within the community i Is +.>All k-th order neighbor nodes representing ui within the community are for u i Weights of (2); node c for any course i Constructing a community of the node>Representing course node c within the community i Is +.>Representing c in the community i For c i Weights of (2); /> and />The calculation formula of (2) is as follows:
wherein, I represents an identity matrix, c is a trainable parameter used for dynamically controlling the size of the community, and the value of c is adjusted by training a model.Is a start vector of all zeros except the location corresponding to the center node, < >>Is an adjacency matrix of the community where the user node is located, +.>Is an adjacency matrix of the community where the course nodes are located.
By the above, the community set C of all users can be obtained U Community set C with all courses C 。
In this embodiment, the data set used contains 1944 users, 694 courses, and the initial nesting size=100 is set, so that the user initial feature H U Is [1944, 100.)]Course context H C Is of dimension [694, 100]。
In this embodiment, multiple meta paths are constructed for the user entity to represent the relationship of different user entity objects under different relationships.
For example, the semantic information of the user element path UCU is: two different users learn the same course.
If the user element path UU, the semantic information of the element path is: two different users have watched the same teaching video.
In this embodiment, different meta-path types are designed for users and courses, and the types are shown in table 1:
TABLE 1
In this embodiment, communities are firstly partitioned according to 2-order neighbors, then the weight of other nodes in the communities to the center node is calculated by adopting a restarting random walk method on the communities, particles start from the center node and randomly walk to the neighbor nodes with transition probability, a certain probability returns to the center node in the walk process, and the community information C of users and courses is obtained by calculating the probability that the particles stay in each node as the weight of each node U 、C C 。
Step S2, collecting the element path adjacent matrix set A U 、A C User initial feature H U Course context feature H C With community collection C U 、C C Input to the association feature extraction module which is built and trained to obtain association features F 'of the user respectively' U Associated features F 'with course' C 。
Specifically, the element path is adjacent to the matrix set A U 、A C User initial feature H U Course context feature H C With community collection C U 、C C Input to the association feature extraction module which is built and trained to obtain association features F 'of the user respectively' U Associated features F 'with course' C Comprising:
step 2.1, by calculating different user nodes u i And u is equal to j Similarity among communities to obtain relevance based on communitiesThe calculation formula is as follows:
wherein ,Vui And Vu j Respectively represent user nodes u i And u is equal to j Neighbor node set of u l Representing Vu i And Vu j Nodes in the intersection, u m Representing Vu i And Vu j The nodes in the union are connected in parallel,representing user node u i For user node u l Is used for the weight of the (c),representing user node u j For user node u l Weight of->Representing user node u i For user node u m Weight of->Representing user node u j For user node u m Weights of (2);
step 2.2, according to the initial characteristics H U Calculating different user nodes u i And u is equal to j Semantic similarity betweenThe calculation formula is as follows:
wherein ,and->Respectively represent user nodes u i With user node u j Semantic information of->The method comprises the steps of converting a high-dimensional vector into a real number, wherein W is a conversion matrix for converting node characteristics into the same space;
step 2.3, adding the semantic similarity of the nodes on the basis of the node association degree based on the community to obtain the enhanced node association degree
Where softmax (·) represents the normalization function, mapping the input values to [0,1 ]]The numerical value of the two components is equal to the numerical value of the two components,and (3) withRespectively indicate +.> and />Normalized values were performed. Sigma represents the activation function, exp (·) represents an exponential function with the natural constant e as a base, and α (·) and β (·) represent a transfer function;
step 2.4, obtaining the feature vector of the user node ui under a certain element path according to the enhanced node association degree and the semantic information of the neighbor node:
wherein ,Nui Representing a set of all neighbor nodes of node ui on the element path;
I.e. by calculating different course nodes c i And c j Similarity among communities to obtain relevance based on communitiesThe calculation formula is as follows:
wherein ,Vci With Vc j Respectively representing course nodes c i And c j Is set of neighbor nodes, c l Representation Vc i With Vc j Nodes in intersection, c m Representation Vc i With Vc j The nodes in the union are connected in parallel,representing course node c i For course node c l Is used for the weight of the (c),representing course node c j For course node c l Weight of->Representing course node c i For course node c m Weight of->Representing course node c j For course node c m Weights of (2);
according to the initial characteristics H C Calculating nodes c of different courses i And c j Semantic similarity betweenThe calculation formula is as follows:
wherein ,and->Respectively representing course nodes c i And course node c j Semantic information of->The method comprises the steps of converting a high-dimensional vector into a real number, wherein W is a conversion matrix for converting node characteristics into the same space;
adding semantic similarity of nodes on the basis of the node relevance based on the community to obtain the enhanced node relevance
Where softmax (·) represents the normalization function, mapping the input values to [0,1 ]]The numerical value of the two components is equal to the numerical value of the two components,and (3) withRespectively indicate +.> and />Normalized values were performed. Sigma represents an activation function, exp (·) represents an exponential function with the natural constant e as a base, and α (·) and β (·) represent a conversion function;
obtaining a course node c according to the enhanced node association degree and semantic information of the neighbor nodes i Feature vector under a certain element path:
wherein ,Nci Representing node c i A set of all neighbor nodes on the element path;
step 2.5, calculating that all user nodes are in a primary path MP z The feature vectors are spliced to obtain the user features on the meta-pathWherein n represents the number of user nodes;
step 2.6, calculating that all course nodes are in a meta-path MP' t The feature vectors are spliced to obtain course features on the meta-pathWherein m represents the number of course nodes;
step 2.7, calculating the user characteristics on each element path to obtain a user characteristic set
Step 2.8, calculating course characteristics on each element path to obtain a course characteristic set
Wherein U represents a set of user nodes, L u Representing trainable vectors, W u Representing a weight matrix, b u Representing a bias vector;
the aggregation is carried out according to the element paths, and user association characteristics under Z element paths are obtained:
Wherein C represents course node set, L c Representing trainable vectors,W c Representing a weight matrix, b c Representing a bias vector;
aggregating according to the meta-paths to obtain course association features under T meta-paths:
in this embodiment, the association feature generation schematic diagram is shown in fig. 2, the community association strengthening module schematic diagram is shown in fig. 3, and the aggregation network contains 8 hidden units and 8 attention points. The number of user nodes is 1944, i.e., n=1944, and the number of course nodes is 694, i.e., m=694. The ReLu function is adopted as the activation function to better mine relevant characteristics, and the problems of gradient saturation and gradient disappearance are avoided when training data are fitted. Jaccord similarity is used to measure community relevance between communities where different nodes are located.
It should be noted that the number of the substrates,a matrix that respectively illustrates what dimensions the feature is.
Step S3, constructing a bipartite graph set B according to the user behavior log, and respectively integrating the initial characteristics H of the user U Course context feature H C Inputting the training result and the bipartite graph set B into a behavior feature extraction module which is constructed and trained to obtain behavior features F', of a user and a course respectively U and F″C 。
Specifically, a bipartite graph set is built according to the user behavior log, and the user initial characteristics H are respectively obtained U Course context feature H C Inputting the training result and the bipartite graph set B into a behavior feature extraction module which is constructed and trained to obtain behavior features F', of a user and a course respectively U and F″C Comprising:
step 3.1, constructing a user-course bipartite graph set B= { B according to the user behavior log 1 B 2 ......B r}, wherein ,Br Representing a user-course bipartite graph corresponding to the behavior type r;
step 3.2, for any bipartite graph, if a plurality of user entities are connected with one course entity, one superside exists to connect the plurality of user nodes; if a plurality of course entities are connected with a user entity, an overtravel exists to connect the course nodes; thereby, the user-course bipartite graph is converted into two isomorphic hypergraph sets G U and GC :
G U ={G U,global G U,1 ......G U,r };
G C ={G C,global G C,1 ......G C,r };
wherein GU,r =(U,E U,r ) Representing user hypergraph corresponding to the r-th behavior type, U representing user node set, E U,r Representing a set of r-type superedges. G C,r =(C,E C,r ) Representing a curriculum hypergraph corresponding to the r-th behavior type, E C,r Representing a set of r-type superedges. G U,global =(U,E U,1 ∪E U,2 ...∪E U,r ) Representing a global user hypergraph whose hyperedge set is defined by G U,1 ......G U,r Is formed by the union of the supersides of the above-mentioned two parts. G C,global =(U,E C,1 ∪E C,2 ...∪E C,r ) Representing global curriculum hypergraphs whose hyperedge sets are defined by G C,1 ......G C,r Is formed by the union of the supersides of the two groups;
step 3.4, constructing an association matrix corresponding to each hypergraph, wherein for each association matrix, if one entity node is associated with one hyperedge, the corresponding position in the association matrix is 1, and if the entity node is not associated with one hyperedge, the corresponding position is 0;
step 3.5, obtaining two groups of incidence matrix sets X through the step 3.4 U and XC :
X U ={X U,global X U,1 ......X U,r };
X C ={X C,gobal X C,1 ......X C,r };
Aggregating two sets of incidence matrices X U 、X C User initial feature H U Course context feature H C Input to the convolutional layer, the operation of layer 1+1 is as follows:
wherein sigma represents a nonlinear activation function,is a convolution calculation unit by which a convolution operation is performed. D (D) U,i Is a user node degree matrix representing a diagonal array of degrees of each node for a diagonal element, the degree of a node representing the number of edges associated with that node. B (B) U,i The super-edge degree matrix is a diagonal matrix which represents the degree that diagonal elements are each super-edge, and the super-edge degree represents the number of nodes contained in the super-edge; w (W) U Representing an identity matrix>Representing a matrix of learnable filters. Due to G U,global Comprising G U,1 ......G U,r So that the convolution calculation unit is additionally added with the aggregation information of the (E)To H U,global Incorporating global information contained into each H U,i The method comprises the steps of carrying out a first treatment on the surface of the G, according to the process of converting the bipartite graph into the hypergraph in the step 3.2 U,i One of the user nodes corresponds to G C,i Is added with ++>To build the corresponding relation between the above nodes and the superside, strengthen H U,i And H is c,i Similar attributes between;
step 3.6, according to step 3.5, obtaining the user characteristic matrix through l iterationsAnd course feature matrix->
Step 3.7, user characteristic matrixAnd course feature matrix->Inputting to the linear layer to obtain user behavior feature F U With course behavior feature F C The linear operation is as follows:
wherein OU With O C Is a conversion matrix, k U And k is equal to C Is a training parameter.
In this embodiment, the behaviors of the user are shown in table 2, and are divided into two isomorphic supera-graph sets according to different behaviors of the user:
G U =[G U,global G u, view information G U, watch video G U,Answering questions ]
G C =[G C,global G C, checking information G C, watching video G C, answering questions ]
Sequence number | User behavior |
1 | Viewing curriculum information |
2 | Watching video of courses |
3 | Answering post-class questions |
TABLE 2
The behavior feature generation diagram in this embodiment is shown in fig. 4, and includes two convolution layers, one linear layer, where the number of convolution layers may be changed.
S4, associating the user with the course to obtain the associated feature F 'of the course' U 、F′ C Behavior characteristics F U 、F″ C Respectively carrying out feature fusion to respectively obtain final feature representations E of users and courses U and EC 。
Specifically, the association characteristic F 'of the user and the course' U 、F′ C Behavior characteristics F U 、F″ C Respectively carrying out feature fusion to respectively obtain final feature representations E of users and courses U and EC Comprising:
associating the user with the course feature F' U 、F′ C With behavioral characteristics F U 、F″ C Respectively splicing to obtain final characteristic representation E of the user and the course U and EC :
E U =concat(F′ U ,F″ U )
E C =concat(F′ C ,F″ C )
In this embodiment, F 'U, F "U, F' C, F C Is 32, and E is obtained after splicing U And E is connected with C Is [1944,64 ] respectively],[694,64]。
Step S5, representing E according to the final characteristics U and EC And calculating a scoring matrix of the user and the courses, and recommending the courses with the scores being the front to the user.
Specifically, E is expressed in terms of final characteristics U and EC The scoring matrix of the user and the course is calculated, and the calculation formula is as follows:
wherein vU ,v C Is a trainable parameter to ensure E U And E is connected with C In the same feature space. Beta U ,β C Is an adjustment parameter, p U Is a potential feature matrix of the user, q C Is a potential feature matrix for the course.
If the corresponding value of the user and the course in the matrix is larger, the interest of the user in the course is larger, otherwise, if the corresponding value of the matrix is smaller, the interest of the user in the course is smaller, and therefore the course with high score is recommended to the user.
In this embodiment, a matrix decomposition-based method is adopted to decompose the scoring matrix into two potential feature matrices, and the learned final feature representation E of the user and course is then obtained U 、E C Inputting to obtain a predictive scoring matrix, taking the deviation between the actual scoring matrix and the predictive scoring matrix as Loss, updating the trainable parameters in the model by using a back propagation algorithm, and obtaining the user through training-course scoring matrix recommending courses of top 10 scoring to the user.
According to the method, the relevance of the nodes is strengthened through community relevance, the relevance is used as a weight aggregation feature, relevance between user entities and courses is captured together at a community level and a node level, and then entity features obtained by different meta-paths are aggregated through a meta-path attention mechanism, so that the entity relevance features are obtained. And modeling the correlation between different behaviors through the behavior characteristic extraction module, learning to obtain the behavior characteristics of the user and the course, and further mining potential information in different behaviors between the user and the course. The final characteristic representation obtained by carrying out characteristic fusion on the associated characteristic and the behavior characteristic more comprehensively covers potential information existing in the original data, so that the accuracy of the recommended course of the application is remarkably improved.
After the scoring matrix of the user-courses is obtained, the 10 courses with the highest scores can be recommended to the user, the user can evaluate and feed back the recommendation result, and the model can be further optimally trained according to the evaluation feedback of the user.
The above examples only represent a few embodiments of the present application, which are described in more detail and are not to be construed as limiting the scope of the invention. It should be noted that it would be apparent to those skilled in the art that various modifications and improvements could be made without departing from the spirit of the present application, which would be within the scope of the present application. Accordingly, the scope of protection of the present application is to be determined by the claims appended hereto.
Claims (5)
1. An online course recommendation method based on community association and behavior feature learning is characterized by comprising the following steps:
obtaining course context information, user personal information and user behavior logs from an online education platform to generate a user element path adjacency matrix set A U Course element path adjacency matrix set A C User initial feature H U Course up and downCharacter feature H C User community set C U Course community set C C ;
Adjoining the element path to the matrix set A U 、A C User initial feature H U Course context feature H C With community collection C U 、C C Input to the association feature extraction module which is built and trained to obtain association features F 'of the user respectively' U Associated features F 'with course' C ;
Building a bipartite graph set B according to the user behavior log, and respectively integrating the initial characteristics H of the user U Course context feature H C Inputting the training result and the bipartite graph set B into a behavior feature extraction module which is built and trained to obtain behavior features F' of users and courses respectively " U and F”C ;
Associating the user with the course feature F' U 、F' C Behavior feature F' U 、F” C Respectively carrying out feature fusion to respectively obtain final feature representations E of users and courses U and EC ;
Representing E according to final characteristics U and EC And calculating a scoring matrix of the user and the courses, and recommending the courses with the scores being the front to the user.
2. The online course recommendation method based on community association and behavior feature learning of claim 1, wherein the obtaining course context information, user personal information, and user behavior log from the online education platform generates a set of user element path adjacency matrices a U Course element path adjacency matrix set A C User initial feature H U Course context feature H C User community set C U Course community set C C Comprising:
step 1.1, acquiring course context information comprising course names, course introduction and belonging categories from an online education platform, creating a corpus according to the acquired course introduction, training a Word2vec model on the corpus, calculating Word vectors corresponding to the course names through the Word2vec model, and obtaining wordsVector combination to obtain course context feature H C ;
Step 1.2, acquiring personal information of the user including gender, age and academic from an online education platform, and inputting the personal information into a single-heat encoder to obtain initial characteristics H of the user U ;
Step 1.3, constructing a meta-path adjacency matrix set A of the users according to the potential relation among the users U Constructing a meta-path adjacency matrix set A of courses according to potential relations among courses C :
A U ={MP 1 MP 2 ...... MP z };
A C ={MP' 1 MP' 2 ...... MP' t };
Wherein, the number of user element paths is z, the number of course element paths is t, and MP z Representing the z-th user element path, MP' t Representing a t-th course element path;
step 1.4, according to the user element path adjacency matrix set A U Creating a user association diagram, for any user node u in the diagram i Constructing communities of the node Representing user node u within the community i Is a node of all k-th order neighbors,representing u in the community i For u for all k-th order neighbor nodes of (2) i Weights of (2); adjacency matrix set A based on course element path C Creating a course association diagram, and for any course node c in the diagram i Constructing a community of the node> Representing course node c within the community i Is +.>Representing c in the community i For c i Weights of (2); /> and />The calculation formula of (2) is as follows:
wherein I represents an identity matrix, c is a trainable parameter for dynamically controlling the size of the community,is a start vector of all zeros except the location corresponding to the center node, < >>Is an adjacency matrix of the community where the user node is located, +.>Is an adjacency matrix of communities where course nodes are located;
by the above, the community set C of all users can be obtained U Community set C with all courses C 。
3. The online course recommendation method based on community association and behavior feature learning of claim 1, wherein the meta-path adjacency matrix set a U 、A C User initial feature H U Course context feature H C With community collection C U 、C C Input to the association feature extraction module which is built and trained to obtain association features F 'of the user' U Associated features F 'with course' C Comprising:
step 2.1, by calculating different user nodes u i And u is equal to j Similarity among communities to obtain relevance based on communitiesThe calculation formula is as follows:
wherein ,Vui And Vu j Respectively represent user nodes u i And u is equal to j Neighbor node set of u l Representing Vu i And Vu j Nodes in the intersection, u m Representing Vu i And Vu j The nodes in the union are connected in parallel,representing user node u i For user node u l Weight of->Representing user node u j For user node u l Weight of->Representing user node u i For user node u m Weight of->Representing user node u j For user node u m Weights of (2);
step 2.2, according to the initial characteristics H U Calculating different user nodes u i And u is equal to j Semantic similarity betweenThe calculation formula is as follows:
wherein ,and->Respectively represent user nodes u i With user node u j Semantic information of->The method comprises the steps of converting a high-dimensional vector into a real number, wherein W is a conversion matrix for converting node characteristics into the same space;
step 2.3, adding the semantic similarity of the nodes on the basis of the node association degree based on the community to obtain the enhanced node association degree
Where softmax (·) represents the normalization function, mapping the input values to [0,1 ]]The numerical value of the two components is equal to the numerical value of the two components,and->Respectively represent pair-> and />Normalized values are performed; sigma represents an activation function, exp (·) represents an exponential function with the natural constant e as a base, and α (·) and β (·) represent a conversion function;
step 2.4, obtaining a user node u according to the enhanced node association degree and semantic information of the neighbor nodes i Feature vector under one element path:
wherein ,Nui Representing node u i A set of all neighbor nodes on the element path;
Step 2.5, sequentially calculating that all user nodes are in one element path MP z The feature vectors are spliced to obtain the user features on the element pathWherein n is as followsShowing the number of user nodes;
step 2.6, sequentially calculating all course nodes in a meta-path MP' t The feature vectors are spliced to obtain course features on the meta-pathWherein m represents the number of course nodes;
step 2.7, calculating the user characteristics on each element path to obtain a user characteristic set
Step 2.8, calculating course characteristics on each element path to obtain a course characteristic set
Wherein U represents a set of user nodes, L u Representing trainable vectors, W u Representing a weight matrix, b u Representing a bias vector;
the aggregation is carried out according to the element paths, and user association characteristics under Z element paths are obtained:
Wherein C represents course node set, L c Representing trainable vectors, W c Representing a weight matrix, b c Representing a bias vector;
aggregating according to the meta-paths to obtain course association features under T meta-paths:
4. the online course recommendation method based on community association and behavior feature learning according to claim 1, wherein the building of the bipartite graph set B from the user behavior log respectively uses the initial features H of the user U Course context feature H C Inputting the training result and the bipartite graph set B into a behavior feature extraction module which is built and trained to obtain behavior features F' of users and courses respectively " U and F”C Comprising:
step 3.1, constructing a user-course bipartite graph set B= { B according to the user behavior log 1 B 2 ......B r}, wherein ,Br Representing a user-course bipartite graph corresponding to the behavior type r;
step 3.2, for any bipartite graph, if a plurality of user entities are connected with one course entity, one superside exists to connect the plurality of user nodes; if a plurality of course entities are connected with a user entity, an overtravel exists to connect the course nodes; thereby, the user-course bipartite graph is converted into two isomorphic hypergraph sets G U and GC :
G U ={G U,global G U,1 ......G U,r };
G C ={G C,global G C,1 ......G C,r };
wherein GU,r =(U,E U,r ) Representing user hypergraph corresponding to the r-th behavior type, U representing user node set, E U,r Representing a set of r-type hyperedges; g C,r =(C,E C,r ) Representing a curriculum hypergraph corresponding to the r-th behavior type, E C,r Representing a set of r-type hyperedges; g U,global =(U,E U,1 ∪E U,2 ...∪E U,r ) Representing a global user hypergraph whose hyperedge set is defined by G U,1 ......G U,r Is formed by the union of the supersides of the two groups; g C,global =(U,E C,1 ∪E C,2 ...∪E C,r ) Representing global curriculum hypergraphs whose hyperedge sets are defined by G C,1 ......G C,r Is formed by the union of the supersides of the two groups;
step 3.4, constructing an association matrix corresponding to each hypergraph, wherein for each association matrix, if one entity node is associated with one hyperedge, the corresponding position in the association matrix is 1, and if the entity node is not associated with one hyperedge, the corresponding position is 0;
step 3.5, obtaining two groups of incidence matrix sets X through the step 3.4 U and XC :
X U ={X U,global X U,1 ......X U,r };
X C ={X C,gobal X C,1 ......X C,r };
Aggregating two sets of incidence matrices X U 、X C User initial feature H U Course context feature H C Input to the convolutional layer, the operation of layer 1+1 is as follows:
wherein sigma represents a nonlinear activation function,is a convolution calculation unit, and carries out convolution operation by the formula; d (D) U,i A user node degree matrix, which represents a diagonal matrix in which diagonal elements are degrees of each node, wherein the degrees of the nodes represent the number of edges associated with the nodes; b (B) U,i The super-edge degree matrix is a diagonal matrix which represents the degree that diagonal elements are each super-edge, and the super-edge degree represents the number of nodes contained in the super-edge; w (W) U Representing an identity matrix>Representing a matrix of learnable filters; due to G U,global Comprising G U,1 ......G U,r So that the convolution calculation unit is additionally added with the aggregation information of the (E)To H U,global Incorporating global information contained into each H U,i The method comprises the steps of carrying out a first treatment on the surface of the G, according to the process of converting the bipartite graph into the hypergraph in the step 3.2 U,i One of the user nodes corresponds to G C,i Is added with ++>To build the corresponding relation between the above nodes and the superside, strengthen H U,i And H is c,i Similar attributes between;
step 3.6, according to step 3.5, obtaining the user characteristic matrix through l iterationsAnd course feature matrix->
Step 3.7, user characteristic matrixAnd course feature matrix->Input to the linear layer to obtain user behavior feature F' U Course behavior feature F' C The linear operation is as follows:
wherein OU With O C Is a conversion matrix, k U And k is equal to C Is a training parameter.
5. The online course recommendation method based on community correlation and behavioral characteristic learning of claim 1, wherein the final characteristic representation E U and EC Calculating a scoring matrix for the user and the course, comprising:
representing E according to final characteristics U and EC The scoring matrix of the user and the course is calculated, and the calculation formula is as follows:
wherein vU ,v C Is a trainable parameter to ensure E U And E is connected with C In the same feature space; beta U ,β C Is an adjustment parameter, p U Is a potential feature matrix of the user, q C Is a potential feature matrix for the course.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310164787.4A CN116187852A (en) | 2023-02-13 | 2023-02-13 | Online course recommendation method based on community association and behavior feature learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310164787.4A CN116187852A (en) | 2023-02-13 | 2023-02-13 | Online course recommendation method based on community association and behavior feature learning |
Publications (1)
Publication Number | Publication Date |
---|---|
CN116187852A true CN116187852A (en) | 2023-05-30 |
Family
ID=86444141
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202310164787.4A Pending CN116187852A (en) | 2023-02-13 | 2023-02-13 | Online course recommendation method based on community association and behavior feature learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116187852A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117811851A (en) * | 2024-03-01 | 2024-04-02 | 深圳市聚亚科技有限公司 | Data transmission method for 4G communication module |
-
2023
- 2023-02-13 CN CN202310164787.4A patent/CN116187852A/en active Pending
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117811851A (en) * | 2024-03-01 | 2024-04-02 | 深圳市聚亚科技有限公司 | Data transmission method for 4G communication module |
CN117811851B (en) * | 2024-03-01 | 2024-05-17 | 深圳市聚亚科技有限公司 | Data transmission method for 4G communication module |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110046304B (en) | User recommendation method and device | |
CN110955834A (en) | Knowledge graph driven personalized accurate recommendation method | |
CN111881363B (en) | Recommendation method based on graph interaction network | |
CN111881342A (en) | Recommendation method based on graph twin network | |
CN114117220A (en) | Deep reinforcement learning interactive recommendation system and method based on knowledge enhancement | |
CN108563755A (en) | A kind of personalized recommendation system and method based on bidirectional circulating neural network | |
CN109871504B (en) | Course recommendation system based on heterogeneous information network and deep learning | |
CN109063147A (en) | Online course forum content recommendation method and system based on text similarity | |
CN112016002A (en) | Mixed recommendation method integrating comment text level attention and time factors | |
CN106780071A (en) | A kind of online community network Information Communication modeling method based on multi-mode mixed model | |
Seo et al. | FaNDeR: fake news detection model using media reliability | |
Lai et al. | Transconv: Relationship embedding in social networks | |
CN116187852A (en) | Online course recommendation method based on community association and behavior feature learning | |
Wang | Research on online learner modeling and course recommendation based on emotional factors | |
CN113283488B (en) | Learning behavior-based cognitive diagnosis method and system | |
CN113065342B (en) | Course recommendation method based on association relation analysis | |
CN112101358B (en) | Method for aligning phrase and picture region based on decoupling and intervention graph network | |
Imran et al. | A framework to provide personalization in learning management systems through a recommender system approach | |
CN111311997B (en) | Interaction method based on network education resources | |
Yu | SPOC-MFLP: A multi-feature learning prediction model for SPOC students using machine learning | |
CN111401525A (en) | Adaptive learning system and method based on deep learning | |
CN115310004A (en) | Graph nerve collaborative filtering recommendation method fusing project time sequence relation | |
Zeng | Application of conditional random field model based on machine learning in online and offline integrated educational resource recommendation | |
Liu et al. | Inductive Cognitive Diagnosis for Fast Student Learning in Web-Based Intelligent Education Systems | |
Zhou | Research on teaching resource recommendation algorithm based on deep learning and cognitive diagnosis |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination |