CN114297498A - Opinion leader identification method and device based on key propagation structure perception - Google Patents
Opinion leader identification method and device based on key propagation structure perception Download PDFInfo
- Publication number
- CN114297498A CN114297498A CN202111638795.5A CN202111638795A CN114297498A CN 114297498 A CN114297498 A CN 114297498A CN 202111638795 A CN202111638795 A CN 202111638795A CN 114297498 A CN114297498 A CN 114297498A
- Authority
- CN
- China
- Prior art keywords
- user
- node
- nodes
- key
- matrix
- 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
- 238000000034 method Methods 0.000 title claims abstract description 29
- 230000008447 perception Effects 0.000 title claims abstract description 18
- 238000000605 extraction Methods 0.000 claims abstract description 22
- 238000005065 mining Methods 0.000 claims abstract description 12
- 238000003062 neural network model Methods 0.000 claims abstract description 11
- 230000005540 biological transmission Effects 0.000 claims abstract description 10
- 230000007246 mechanism Effects 0.000 claims abstract description 7
- 238000005516 engineering process Methods 0.000 claims abstract description 5
- 238000013528 artificial neural network Methods 0.000 claims abstract description 4
- 239000011159 matrix material Substances 0.000 claims description 63
- 239000010410 layer Substances 0.000 claims description 31
- 239000013598 vector Substances 0.000 claims description 22
- 230000006870 function Effects 0.000 claims description 20
- 238000004458 analytical method Methods 0.000 claims description 16
- 230000004931 aggregating effect Effects 0.000 claims description 9
- 238000012216 screening Methods 0.000 claims description 8
- 230000004913 activation Effects 0.000 claims description 6
- 230000002776 aggregation Effects 0.000 claims description 5
- 238000004220 aggregation Methods 0.000 claims description 5
- 238000012512 characterization method Methods 0.000 claims description 5
- 238000012549 training Methods 0.000 claims description 4
- 238000010606 normalization Methods 0.000 claims description 3
- 239000002356 single layer Substances 0.000 claims description 3
- OAICVXFJPJFONN-UHFFFAOYSA-N Phosphorus Chemical compound [P] OAICVXFJPJFONN-UHFFFAOYSA-N 0.000 claims description 2
- 238000004422 calculation algorithm Methods 0.000 abstract description 4
- 238000013461 design Methods 0.000 abstract description 3
- 230000009977 dual effect Effects 0.000 abstract 1
- 238000007418 data mining Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000007621 cluster analysis Methods 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000004806 packaging method and process Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 230000000644 propagated effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012163 sequencing technique Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
Images
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses an opinion leader identification method and device based on key propagation structure perception by a method in the field of network technology processing. The method is based on a neural network algorithm, a user dual feature extraction module and a key propagation structure mining module are designed to be composed of two logic modules, and opinion leaders and incidence relation outputs with representative features in input data are obtained by inputting data of different topics in a microblog. The invention designs a message transmission mechanism based on node centrality, fully combines the importance of nodes in a topological structure, and constructs a new graph neural network model to extract the characteristics of users in a social network. The method is characterized in that key structure information in different events is mined by using a graph classification task for the first time, top-k opinion leaders in a social network are mined, potential connections among the opinion leaders can be learned from three angles of node connectivity, node similarity and node centrality, and a key propagation structure in the network is constructed.
Description
Technical Field
The invention relates to the technical field of data mining, in particular to an opinion leader identification method and device based on key propagation structure perception.
Background
With the development and popularization of internet technology in recent years, more and more netizens use social networks to communicate, network public opinion ecology is gradually changed, and transfer of public speaking rights is promoted. In the social network, information is propagated along the social network formed by the association relationship among the users through the propagation behavior among the users. The functions of each individual in the network structure and the functions are greatly different, and the opinion leaders are special individuals which can influence the network structure and the functions formed in the information propagation process to a greater extent. For example, microboda V may accelerate the spread of facts or rumors in social networks. Therefore, in order to better guide network public opinions, opinion leaders in social networks need to be accurately discovered from massive users, and development trends of network public opinions need to be captured and predicted.
Opinion leader identification is an important research direction in the field of data mining, and plays an important role as an intermediary or filter in the formation of mass-propagation effects. A plurality of social network opinion leader identification algorithms are proposed at present, and classical methods based on network topology structures comprise degree centrality, betweenness centrality, approach centrality, feature vector centrality and the like. The most intuitive and basic method is centrality, the calculation is simple and effective, only local information of a complex network is reflected, and the self characteristics of the nodes and the mutual influence among the nodes are not considered. And each evaluation method is based on statistics and has specific limitations, so that the evaluation method cannot be well applied to all types of networks. And also identifies opinion leader nodes in the network based on cluster analysis, a PageRank algorithm, etc., and these conventional methods have the advantages of simple model and capability of converging in a relatively short time when there are many samples. But the clustering scale cannot be controlled, the nodes possibly in the same category do not all have similar characteristics, and the PageRank algorithm assumes that the nodes have the same hop probability and has the defect of non-unique sequencing result. Some neural network models mostly adopt a single relation modeling mode, and have the problem that key structures in a network formed by important nodes and neighbors of the important nodes cannot be sufficiently excavated.
Disclosure of Invention
An object of the present invention is to solve at least the above problems or disadvantages and to provide at least the advantages described hereinafter.
The invention provides an opinion leader identification method based on key propagation structure perception, which comprises the steps of acquiring microblog contents of different topics as input, searching a social network library according to the input topics, directly ending if the topics do not exist, taking out an information propagation network formed by forwarding relations of users and postings in microblogs of the topics if the topics do not exist, and further inputting the information propagation network into a feature extraction module;
the user feature extraction module is internally composed of two sequential operations of user centrality learning and aggregation updating of user features, the user centrality learning operation is executed to randomly initialize a user feature matrix, and an adjacent matrix of a network is constructed according to a topic propagation condition; then, aggregating and updating user characteristics, aggregating user information based on a message transmission mechanism of node centrality, and constructing a new graph neural network model to extract the characteristics of users in the network; the user characteristic extraction module outputs a user characterization matrix to the key propagation structure mining module;
the key propagation structure mining module comprises a key propagation structure learning submodule and five units, wherein the key propagation structure learning submodule comprises a key user screening unit, a user similarity learning unit, a user connectivity analysis unit and a user centrality analysis unit, the five units are used for updating a key structure adjacency matrix, the key user screening unit inputs a user representation matrix output by the user characteristic extraction module, and the user similarity learning unit, the user connectivity analysis unit and the user centrality analysis unit input user representation vectors and index information which are sequenced by users and select top-ki nodes as opinion leader nodes in the current convolutional layer; updating the output values of the user similarity learning unit, the user connectivity analysis unit and the user centrality analysis unit input by the key structure adjacency matrix, and calculating to obtain a key propagation structure in the network;
the key user screening unit is mainly used for sorting the importance of users, selecting top-ki nodes as opinion leader nodes in the current convolutional layer, calculating the node scores through a distance function, analyzing the information quantity of each node containing neighbors and sorting the importance of the users according to the information quantity;
the updating key structure adjacent matrix unit is mainly used for updating the topological structure information of the selected opinion leader nodes, selecting node pairs with potential connection possibility greater than the threshold value among the nodes in the key propagation structure learning submodule to form new connecting edges by setting the threshold value, updating the adjacent matrix of the leader nodes, and forming a key propagation structure in the network together with the excavated opinion leader;
and then, judging whether the user feature extraction module is more than or equal to 2, if so, outputting an identification result of the opinion leader after training and tuning, and if not, inputting a generated result into the user centrality learning operation for re-execution.
The step of obtaining the microblog contents of different topics comprises the following steps: the method comprises the steps of obtaining microblogs in a preset time period based on topic keywords, obtaining related microblog contents by a crawler technology through the preset time period and the keywords, if the microblog contents are in a forwarding relation, enabling a connecting edge to exist between two corresponding nodes, respectively constructing an information propagation network based on user post forwarding relations under different topic keywords, and customizing and dividing different topics into different categories: politics, culture, society, and economy.
The operation of the user feature extraction module for aggregating and updating the user features adopts a new graph convolution neural network model based on node centrality, the feature matrix X of the user nodes in the social network library and the relation matrix A among the user nodes are input, the user feature matrix is initialized at random, constructing an adjacency matrix of the network according to the topic propagation condition, then aggregating user information based on a message transmission mechanism of node centrality, constructing a new graph neural network model to extract the characteristics of users in the network, wherein the importance of other users to a certain user is measured by the proximity center of each node in the potential space, meanwhile, the topological structure information in the network can be considered, the proximity centrality of one point is higher, which shows that the distance from the point to other points in the network is closer overall, otherwise, the distance is farther, and the message aggregation mode is as follows:
where i is a node, | V | represents the number of all nodes in the graph, W and WcIs a parameter matrix, hiThe feature vector representing node i, exp represents an exponential function with a natural constant e as the base, cos (Wh)i,Whj) Measuring the distance of two nodes in a potential space by using the similarity of the nodes, wherein the closer the distance in the potential space is, the more similar the two nodes are, sigma represents a ReLU nonlinear activation function, NiAnd representing all neighbor nodes around the node i, wherein the updated feature vector of the node is related to all the adjacent nodes, neighbor information is aggregated by different weights based on the centrality of each neighbor node, and the user feature extraction module forms a new graph convolution layer based on the centrality of the node and is used for extracting user feature vectors of different levels and inputting the user feature vectors into the key propagation structure mining module.
The specific step of calculating the node score is that the score of each node in the neighborhood is defined as the Euclidean distance between the node representation and the neighbor representation thereof: sL=||(IL-(DL)-1AL)HL||2In which ILOne identity matrix, D, representing the L-th convolutional layerLRepresents the adjacency matrix ALDiagonal angle matrix of (H)LA feature matrix representing nodes, | · | > non-woven phosphor2Denotes the L2 norm, sLThe Euclidean distance between the L convolutional layer and the neighbor node of each node is calculated, and the closer the distance represents that the node is similar to the neighbor distance, the more information contained in the neighbor can be contained, and the node is more important relative to other nodes.
The opinion leaderThe user-characterized vectors of nodes are used to update the feature matrix H in convolutional layersL+1And adjacency matrix AL+1。
Selected kiThe nodes can cause the disconnection of the highly-related nodes in the subgraph, thereby losing the integrity of the graph structure information and further hindering the message transmission process, and potential relations exist among the nodes, and the selected k is learned through the updated key structure adjacency matrix (propagation structure learning submodule)iThe potential connection between each node reconstructs an adjacent matrix in the convolutional layer, and is realized by considering the similarity between two nodes, wherein the adjacent matrix of the node and the centrality of the node are specifically defined as follows:
where σ denotes the ReLU nonlinear activation function, hi,hjRepresenting the feature vectors of the nodes i and j, f is a single-layer neural network, and the similarity between the two nodes is learned, for the importance of the user learned in the user feature extraction module, the larger the average value of the distances from the node to other nodes in the L convolutional layer is, the lower the probability of connecting edges between the node and the node j is, and the lower the probability of potential connection exists, wherein alpha and beta are two hyper-parameters, AL(i, j) indicates whether there is a connecting edge between two nodes in the L convolutional layer, if so, the value of the item is 1; otherwise it is 0. ELThe probability of potential continuous edges of two nodes is measured integrally, in order to enable scores to be compared on different nodes, a softmax function is used for normalization, meanwhile, a threshold value is set, continuous edges of nodes above the threshold value are reserved, and an adjacency matrix in a convolutional layer is updated.
The technical effects to be realized by the invention are as follows:
the method aims at identifying the opinion leader in the social network and the application scene of related groups. The invention provides an end-to-end opinion leader identification method based on a graph convolution neural network model. The method provided by the invention has the following characteristics:
1. the invention designs a message transmission mechanism based on node centrality, fully combines the importance of nodes in a topological structure, and constructs a new graph neural network model to extract the characteristics of users in a social network.
2. The method provided by the invention is used for mining key structure information in different events by using the graph classification task for the first time, so that the potential relation among opinion leaders can be learned from three angles of node connectivity, node similarity and node centrality while top-k opinion leaders in the social network are mined, and a key propagation structure in the network is constructed.
Drawings
FIG. 1 is a flow chart of an opinion leader identification method based on key propagation structure perception
Detailed Description
The following is a preferred embodiment of the present invention and is further described with reference to the accompanying drawings, but the present invention is not limited to this embodiment.
The invention provides an opinion leader identification method based on key propagation structure perception, which can realize the mining of opinion leaders and incidence relations of specific topics and consists of two logic modules, namely a user characteristic extraction module and a key propagation structure mining module.
Firstly, microblogs in a preset time period are obtained based on topic keywords, and related microblog contents are obtained by a crawler technology through the preset time period and the keywords. If the microblog contents are in a forwarding relation, a connecting edge exists between the two corresponding nodes. Respectively constructing an information transmission network based on user post forwarding relations under different topic keywords, and customizing and dividing different topics into different categories: political, cultural, social, economic, and different topic data form a social network library. Searching a social network library according to the input topic keywords, directly ending if the topic does not exist, and taking out an information propagation network corresponding to the topic if the topic exists; and then, randomly initializing a user feature matrix, and constructing an adjacency matrix of the network according to the topic propagation condition.
Inputting an initial user feature matrix and an adjacency matrix to a user feature extraction module, wherein the user feature extraction module is mainly a new graph neural network model fusing node centrality, and aggregating and updating user information based on a message transmission mechanism of node proximity centrality to extract features of users in the network. The importance of other users to a certain user is measured by utilizing the proximity center degree of each node in the potential space, and the topological structure information in the network can be considered. The closer centrality of a node is higher, which means that the distance from the node to other points in the network is generally closer, and vice versa, the distance is farther. The specific aggregation updating user characteristic mode based on the proximity centrality is as follows (taking the node i as an example):
wherein | V | represents the number of all nodes in the graph, W and WcIs a parameter matrix, hiThe feature vector representing node i, exp represents an exponential function with a natural constant e as the base, cos (Wh)i,Whj) Measuring the distance of two nodes in a potential space by using the similarity of the nodes, wherein the closer the distance in the potential space is, the more similar the two nodes are, sigma represents a ReLU nonlinear activation function, NiRepresenting all neighbor nodes around the node i, the updated feature vector of the node is related to all the nodes adjacent to the node i, and the neighbor information is aggregated with different weights based on the centrality of each neighbor node. The user feature extraction module forms a new graph convolution layer based on the node centrality, is used for extracting user feature vectors of different levels, and then inputs the user feature vectors into the key propagation structure mining module.
The key propagation structure mining module comprises key propagation structure science consisting of a key user screening unit, a user similarity learning unit, a user connectivity analysis unit and a user centrality analysis unitThe learning module and the updating key structure are adjacent to five units of the matrix. Inputting the user characterization vectors into a key user screening unit, calculating the similarity between nodes through a distance function, analyzing the information quantity of each node containing neighbors, sorting according to the importance degree of the information quantity contained to the user, and selecting top-kiOpinion leaders, k, in social networksiIs a predefined number representing the number of nodes in the output graph. Specifically, the score of each node in the neighborhood is defined as the euclidean distance between the node representation and the neighbor representation thereof: sL=||(IL-(DL)-1AL)HL||2In which ILOne identity matrix, D, representing the L-th convolutional layerLRepresents the adjacency matrix ALDiagonal angle matrix of (H)LA feature matrix representing the nodes. I | · | purple wind2Denotes the L2 norm, sLThe Euclidean distance between the L convolutional layer and the neighbor node of each node is calculated, and the closer the distance represents that the node is similar to the neighbor distance, the more information contained in the neighbor can be contained, and the node is more important relative to other nodes. Therefore, each node is sequenced according to the calculated scores to obtain the ranking top-kiAnd (4) nodes which are opinion leader nodes in the current convolutional layer. Updating the feature matrix H in the convolutional layer according to the selected opinion leader nodesL+1And adjacency matrix AL+1In which H isL+1=HL(idx,;),AL+1=AL(idx ), idx representing the index of each node.
Inputting the updated feature matrix and the adjacency matrix into a key propagation structure learning submodule, and reconstructing the adjacency matrix in the updated convolutional layer through the propagation structure learning submodule to learn the potential relation between the nodes. By considering the similarity between two nodes, the adjacency matrix of the nodes and the centrality of the nodes are specifically defined as:
wherein, the user similarity learning unit sigma represents a ReLU nonlinear activation function, hi,hjAnd f is a single-layer neural network, and the similarity between the two nodes is learned. User centrality analysis unitFor the importance of the user learned in the user feature extraction module, the larger the average value of the distances from the node to other nodes in the L convolutional layer is, the lower the probability of connecting edges between the node and the node j is, and the lower the probability of potential contact exists, wherein alpha and beta are two hyper-parameters. The user connectivity analysis unit AL (i, j) indicates whether a connecting edge exists between two nodes in the L convolutional layer, and if the connecting edge exists, the value of the item is 1; otherwise it is 0. ELThe ensemble measures the probability of potential edges between two nodes, and in order to make the scores comparable across different nodes, the softmax function is used for normalization:
in order to ensure the sparsity of the graph structure and facilitate training and tuning, a threshold value is set, the continuous edges of nodes above the threshold value are reserved, and the adjacent matrix in the convolutional layer is updated to obtain the optimized graph
Obtaining two levels of key propagation structures by repeatedly stacking two layers of convolution layers, and splicing node representations in each level of key structure together to obtain a final representation vector HG。
In order to train the model, an objective function is constructed by using a graph classification task, and the obtained feature representation of the whole graph is input into a three-layer perceptron to predict the label of the graph. Designing a multi-classification task cross entropy loss function to carry out model training, and after adjusting parameters, realizing interface packaging of the trained model, wherein the design of the overall loss function is as follows:
y’=softmax(MLP(HG))
wherein MLP denotes a multilayer perceptron, HGThe representation vector of each graph is obtained according to user groups of different levels, M represents the number of the social networks with the labels, and c represents the number of the categories of the social networks.
In order to achieve the above object, the present invention further provides an opinion leader identification device based on key propagation structure perception, comprising: a processor and a memory; the memory is used for the opinion leader identification method based on key propagation structure perception; the processor is connected with the memory and used for executing the method stored in the memory so as to enable the opinion leader identification device based on key propagation structure perception to execute the functional modules.
Claims (7)
1. An opinion leader identification method based on key propagation structure perception is characterized in that: the method comprises the steps of acquiring microblog contents of different topics as input, searching a social network library according to the input topics, directly ending if the topics do not exist, taking out an information propagation network formed by forwarding relations of users and postings in microblogs of the topics if the topics do not exist, and inputting the information propagation network into a feature extraction module;
the user feature extraction module is internally composed of two sequential operations of user centrality learning and aggregation updating of user features, the user centrality learning operation is executed to randomly initialize a user feature matrix, and an adjacent matrix of a network is constructed according to a topic propagation condition; then, aggregating and updating user characteristics, aggregating user information based on a message transmission mechanism of node centrality, and constructing a new graph neural network model to extract the characteristics of users in the network; the user characteristic extraction module outputs a user characterization matrix to the key propagation structure mining module;
the key propagation structure mining module comprises a key user screening unit and user similarity scienceA key propagation structure learning submodule consisting of a learning unit, a user connectivity analysis unit and a user centrality analysis unit and five units for updating a key structure adjacency matrix, wherein the key user screening unit inputs a user characterization matrix output by the user feature extraction module, and the user similarity learning unit, the user connectivity analysis unit and the user centrality analysis unit input top-k which is sorted by users and selectediEach node is used as a user representation vector and index information of an opinion leader node in the current convolutional layer; updating the output values of the user similarity learning unit, the user connectivity analysis unit and the user centrality analysis unit input by the key structure adjacency matrix, and calculating to obtain a key propagation structure in the network;
the key user screening unit sorts the importance of the users;
updating the topological structure information of the selected opinion leader nodes by the key structure adjacent matrix updating unit, updating the adjacent matrix of the leader nodes, and forming a key propagation structure in the network together with the excavated opinion leader;
and then, judging whether the user feature extraction module is more than or equal to 2, if so, outputting an identification result of the opinion leader after training and tuning, and if not, inputting a generated result into the user centrality learning operation for re-execution.
2. The opinion leader identification method based on key propagation structure perception according to claim 1, characterized in that: the step of obtaining the microblog contents of different topics comprises the following steps: the method comprises the steps of obtaining microblogs in a preset time period based on topic keywords, obtaining related microblog contents by a crawler technology through the preset time period and the keywords, if the microblog contents are in a forwarding relation, enabling a connecting edge to exist between two corresponding nodes, respectively constructing an information propagation network based on user post forwarding relations under different topic keywords, and customizing and dividing different topics into different categories: politics, culture, society, and economy.
3. The opinion leader identification method based on key propagation structure perception according to claim 2, characterized in that: the operation of the user feature extraction module for aggregating and updating the user features adopts a new graph convolution neural network model based on node centrality, the feature matrix X of the user nodes in the social network library and the relation matrix A among the user nodes are input, the user feature matrix is initialized at random, constructing an adjacency matrix of the network according to the topic propagation condition, then aggregating user information based on a message transmission mechanism of node centrality, constructing a new graph neural network model to extract the characteristics of users in the network, wherein the importance of other users to a certain user is measured by the proximity center of each node in the potential space, meanwhile, the topological structure information in the network can be considered, the proximity centrality of one point is higher, which shows that the distance from the point to other points in the network is closer overall, otherwise, the distance is farther, and the message aggregation mode is as follows:
where i is a node, | V | represents the number of all nodes in the graph, W and WcIs a parameter matrix, hiThe feature vector representing node i, exp represents an exponential function with a natural constant e as the base, cos (Wh)i,Whj) Measuring the distance of two nodes in a potential space by using the similarity of the nodes, wherein the closer the distance in the potential space is, the more similar the two nodes are, sigma represents a ReLU nonlinear activation function, NiAnd representing all neighbor nodes around the node i, wherein the updated feature vector of the node is related to all the adjacent nodes, neighbor information is aggregated by different weights based on the centrality of each neighbor node, and the user feature extraction module forms a new graph convolution layer based on the centrality of the node and is used for extracting user feature vectors of different levels and inputting the user feature vectors into the key propagation structure mining module.
4. The opinion leader identification method based on key propagation structure perception according to claim 3, characterized in that:the specific step of calculating the node score is that the score of each node in the neighborhood is defined as the Euclidean distance between the node representation and the neighbor representation thereof: sL=||(IL-(DL)-1AL)HL||2In which ILOne identity matrix, D, representing the L-th convolutional layerLRepresents the adjacency matrix ALDiagonal angle matrix of (H)LA feature matrix representing nodes, | · | > non-woven phosphor2Denotes the L2 norm, sLThe Euclidean distance between the L convolutional layer and the neighbor node of each node is calculated, and the closer the distance represents that the node is similar to the neighbor distance, the more information contained in the neighbor can be contained, and the node is more important relative to other nodes.
5. The opinion leader identification method based on key propagation structure perception according to claim 4, characterized in that: the user characterization vectors of the opinion leader nodes are used to update a feature matrix H in convolutional layersL+1And adjacency matrix AL+1。
6. The opinion leader identification method based on key propagation structure perception according to claim 5, characterized in that: selected kiThe nodes can cause the disconnection of the highly relevant nodes in the subgraph, thereby losing the integrity of the graph structure information and further hindering the message transmission process, and meanwhile, potential relations exist among the nodes, and k selected by the updated key structure adjacency matrix learning is selectediThe potential connection between each node reconstructs an adjacent matrix in the convolutional layer, and is realized by considering the similarity between two nodes, wherein the adjacent matrix of the node and the centrality of the node are specifically defined as follows:
where σ denotes the ReLU nonlinear activation function, hi,hjA feature vector representing nodes i, j, f being oneA single-layer neural network for learning the similarity between two nodes, for the importance of the user learned in the user feature extraction module, the larger the average value of the distances from the node to other nodes in the L convolutional layer is, the lower the probability of connecting edges between the node and the node j is, and the lower the probability of potential connection exists, wherein alpha and beta are two hyper-parameters, AL(i, j) indicates whether there is a connecting edge between two nodes in the L convolutional layer, if so, the value of the item is 1; otherwise 0, ELThe probability of potential continuous edges of two nodes is measured integrally, in order to enable scores to be compared on different nodes, a softmax function is used for normalization, meanwhile, a threshold value is set, continuous edges of nodes above the threshold value are reserved, and an adjacency matrix in a convolutional layer is updated.
7. The utility model provides an opinion leader recognition device based on perception of key propagation structure which characterized in that: use of the opinion leader identification method based on key propagation structure perception according to any of the claims 1-6.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111638795.5A CN114297498A (en) | 2021-12-29 | 2021-12-29 | Opinion leader identification method and device based on key propagation structure perception |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111638795.5A CN114297498A (en) | 2021-12-29 | 2021-12-29 | Opinion leader identification method and device based on key propagation structure perception |
Publications (1)
Publication Number | Publication Date |
---|---|
CN114297498A true CN114297498A (en) | 2022-04-08 |
Family
ID=80970882
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202111638795.5A Pending CN114297498A (en) | 2021-12-29 | 2021-12-29 | Opinion leader identification method and device based on key propagation structure perception |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114297498A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2024098516A1 (en) * | 2022-11-07 | 2024-05-16 | 中电科大数据研究院有限公司 | Social network key node mining method and device and storage medium |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20070168533A1 (en) * | 2003-12-30 | 2007-07-19 | Geoffrey Canright | Method for managing networks by analyzing connectivity |
CN106204297A (en) * | 2016-07-11 | 2016-12-07 | 深圳市中北明夷科技有限公司 | A kind of recognition methods closing social propagation opinion leader and device |
CN112269922A (en) * | 2020-10-14 | 2021-01-26 | 西华大学 | Community public opinion key character discovery method based on network representation learning |
-
2021
- 2021-12-29 CN CN202111638795.5A patent/CN114297498A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20070168533A1 (en) * | 2003-12-30 | 2007-07-19 | Geoffrey Canright | Method for managing networks by analyzing connectivity |
CN106204297A (en) * | 2016-07-11 | 2016-12-07 | 深圳市中北明夷科技有限公司 | A kind of recognition methods closing social propagation opinion leader and device |
CN112269922A (en) * | 2020-10-14 | 2021-01-26 | 西华大学 | Community public opinion key character discovery method based on network representation learning |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2024098516A1 (en) * | 2022-11-07 | 2024-05-16 | 中电科大数据研究院有限公司 | Social network key node mining method and device and storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110263280B (en) | Multi-view-based dynamic link prediction depth model and application | |
CN114117220B (en) | Deep reinforcement learning interactive recommendation system and method based on knowledge enhancement | |
CN112529168B (en) | GCN-based attribute multilayer network representation learning method | |
Chiroma et al. | Progress on artificial neural networks for big data analytics: a survey | |
Natesan Ramamurthy et al. | Model agnostic multilevel explanations | |
CN109255586B (en) | Online personalized recommendation method for e-government affairs handling | |
Tan et al. | A coevolutionary algorithm for rules discovery in data mining | |
Lu et al. | GLR: A graph-based latent representation model for successive POI recommendation | |
CN111008337B (en) | Deep attention rumor identification method and device based on ternary characteristics | |
CN113378913A (en) | Semi-supervised node classification method based on self-supervised learning | |
CN112559764A (en) | Content recommendation method based on domain knowledge graph | |
CN112580902A (en) | Object data processing method and device, computer equipment and storage medium | |
Kose et al. | Fair contrastive learning on graphs | |
CN115687760A (en) | User learning interest label prediction method based on graph neural network | |
Zhang et al. | Consumer credit risk assessment: A review from the state-of-the-art classification algorithms, data traits, and learning methods | |
Fu et al. | Robust representation learning for heterogeneous attributed networks | |
CN118071400A (en) | Application method and system based on graph computing technology in information consumption field | |
CN114297498A (en) | Opinion leader identification method and device based on key propagation structure perception | |
Kumar et al. | Graph Convolutional Neural Networks for Link Prediction in Social Networks | |
Chen et al. | Improving the forecasting and classification of extreme events in imbalanced time series through block resampling in the joint predictor-forecast space | |
CN114265954B (en) | Graph representation learning method based on position and structure information | |
CN115238075B (en) | Text sentiment classification method based on hypergraph pooling | |
CN113159976B (en) | Identification method for important users of microblog network | |
CN115086179A (en) | Detection method for community structure in social network | |
Rezaeipanah et al. | Providing a new method for link prediction in social networks based on the meta-heuristic algorithm |
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 |