WO2022143505A1 - 群组类型识别方法、装置、计算机设备及介质 - Google Patents
群组类型识别方法、装置、计算机设备及介质 Download PDFInfo
- Publication number
- WO2022143505A1 WO2022143505A1 PCT/CN2021/141553 CN2021141553W WO2022143505A1 WO 2022143505 A1 WO2022143505 A1 WO 2022143505A1 CN 2021141553 W CN2021141553 W CN 2021141553W WO 2022143505 A1 WO2022143505 A1 WO 2022143505A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- user
- user nodes
- feature
- nodes
- graph
- Prior art date
Links
- 238000000034 method Methods 0.000 title claims abstract description 72
- 230000004927 fusion Effects 0.000 claims description 119
- 238000004590 computer program Methods 0.000 claims description 60
- 230000000875 corresponding effect Effects 0.000 claims description 60
- 238000012216 screening Methods 0.000 claims description 54
- 230000006399 behavior Effects 0.000 claims description 22
- 230000008569 process Effects 0.000 claims description 12
- 238000012549 training Methods 0.000 claims description 8
- 230000002596 correlated effect Effects 0.000 claims description 7
- 238000012545 processing Methods 0.000 description 27
- 238000010586 diagram Methods 0.000 description 13
- 230000002093 peripheral effect Effects 0.000 description 10
- 230000006870 function Effects 0.000 description 7
- 239000013598 vector Substances 0.000 description 7
- 238000005516 engineering process Methods 0.000 description 6
- 239000011159 matrix material Substances 0.000 description 6
- 238000004891 communication Methods 0.000 description 5
- 238000012935 Averaging Methods 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000001914 filtration Methods 0.000 description 2
- 238000011176 pooling Methods 0.000 description 2
- 208000001613 Gambling Diseases 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000009877 rendering Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 238000012552 review Methods 0.000 description 1
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L51/00—User-to-user messaging in packet-switching networks, transmitted according to store-and-forward or real-time protocols, e.g. e-mail
- H04L51/04—Real-time or near real-time messaging, e.g. instant messaging [IM]
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L12/00—Data switching networks
- H04L12/02—Details
- H04L12/16—Arrangements for providing special services to substations
- H04L12/18—Arrangements for providing special services to substations for broadcast or conference, e.g. multicast
- H04L12/185—Arrangements for providing special services to substations for broadcast or conference, e.g. multicast with management of multicast group membership
-
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L51/00—User-to-user messaging in packet-switching networks, transmitted according to store-and-forward or real-time protocols, e.g. e-mail
- H04L51/52—User-to-user messaging in packet-switching networks, transmitted according to store-and-forward or real-time protocols, e.g. e-mail for supporting social networking services
Definitions
- the embodiments of the present application relate to the field of computer technologies, and in particular, to a group type identification method, apparatus, computer device, and medium.
- the embodiments of the present application provide a group type identification method, apparatus, computer equipment and medium, which improve the identification accuracy rate of the group type.
- the technical solution is as follows:
- a group type identification method comprising:
- the first target graph is constructed according to the association relationship between the plurality of first user nodes;
- an attention parameter of each first user node is obtained, where the attention parameter indicates the importance of the first user node in the first target graph degree;
- the attention parameters of the plurality of second user nodes are greater than the attention parameters of the unselected first user nodes
- the relationship between the two user nodes is constructed.
- a group type identification device comprising:
- a feature acquisition module configured to acquire first graph structural features of the first target graph and first user features of multiple first user nodes in the first target graph, where the first user nodes are users in the target group Identifying corresponding nodes, the first target graph is constructed according to the association relationship between the plurality of first user nodes;
- a first attention acquisition module configured to acquire attention parameters of each first user node in the first target graph based on the first graph structural feature and a plurality of first user features, and the attention parameters represent the degree of importance of the first user node in the first target graph;
- a first screening module configured to select a plurality of second user nodes from the plurality of first user nodes, the attention parameters of the plurality of second user nodes are greater than the attention parameters of the unselected first user nodes ;
- a type identification module configured to identify the group type of the target group based on the first user characteristics of the plurality of second user nodes and the second graph structure characteristics of the second target graph, where the second target graph is It is constructed according to the association relationship between the plurality of second user nodes.
- the apparatus further includes:
- a feature adjustment module configured to adjust the first user features of the multiple second user nodes based on the second graph structure feature to obtain the second user features of the multiple second user nodes;
- a second attention obtaining module configured to obtain attention parameters of each second user node in the second target graph based on the second graph structure feature and a plurality of second user features
- the second screening module is configured to select a plurality of third user nodes from the plurality of second user nodes, and the attention parameters of the plurality of third user nodes are greater than the attention parameters of the unselected second user nodes .
- the type identification module is configured to, based on the first user characteristics of the plurality of second user nodes, the second graph structure characteristics, and the first user characteristics of the plurality of third user nodes.
- the second user feature and the third graph structure feature of the third target graph identify the group type of the target group, and the third target graph is constructed according to the association relationship between the plurality of third user nodes.
- the type identification module includes:
- a first fusion unit configured to fuse the first user features and the second graph structure features of the plurality of second user nodes to obtain a first fusion feature
- a second fusion unit configured to fuse the second user features of the plurality of third user nodes and the third graph structure features to obtain second fusion features
- a type identification unit configured to identify the group type of the target group based on the first fusion feature and the second fusion feature.
- the first fusion unit is configured to:
- averaging is performed based on the first user characteristics of the plurality of second user nodes and the second graph structure characteristics to obtain the average user characteristics
- the first fusion feature is obtained by splicing the average user feature and the largest user feature among the first user features of the plurality of second user nodes.
- the type identification unit is used for:
- a group type of the target group is identified.
- the group type recognition model includes a first attention network, a first screening network and a recognition network,
- the first attention acquisition module is configured to call the first attention network to acquire each first image in the first target image based on the first image structure feature and the plurality of first user features. Attention parameters of user nodes;
- the first screening module configured to invoke the first screening network to select the plurality of second user nodes from the plurality of first user nodes;
- the type identification module is configured to invoke the identification network to identify the group type of the target group based on the first user characteristics of the plurality of second user nodes and the second graph structure characteristics.
- the group type identification model further includes a first convolutional network, a second attention network and a second screening network
- the apparatus further includes:
- a feature adjustment module configured to call the first convolutional network, adjust the first user features of the plurality of second user nodes based on the second graph structural features, and obtain the first user features of the plurality of second user nodes.
- a second attention acquisition module configured to call the second attention network, and acquire the attention of each second user node in the second target graph based on the second graph structure feature and multiple second user features force parameter;
- the second screening module is configured to call the second screening network, and select a plurality of third user nodes from the plurality of second user nodes, and the attention parameters of the plurality of third user nodes are greater than those that have not been selected.
- the attention parameter of the second user node is configured to call the second screening network, and select a plurality of third user nodes from the plurality of second user nodes, and the attention parameters of the plurality of third user nodes are greater than those that have not been selected.
- the attention parameter of the second user node is configured to call the second screening network, and select a plurality of third user nodes from the plurality of second user nodes, and the attention parameters of the plurality of third user nodes are greater than those that have not been selected.
- the type identification module configured to invoke the identification network, is based on the first user characteristics of the plurality of second user nodes, the second graph structure characteristics, the plurality of second user nodes
- the second user feature of the third user node and the third graph structure feature of the third target graph identify the group type of the target group, and the third target graph is based on the relationship between the plurality of third user nodes. relationship is built.
- the group type identification model further includes a first fusion network and a second fusion network
- the type identification module includes:
- a first fusion unit configured to invoke the first fusion network to fuse the first user features of the plurality of second user nodes and the second graph structure features to obtain the first fusion features
- a second fusion unit configured to invoke the second fusion network, and fuse the second user features of the plurality of third user nodes and the third graph structure features to obtain a second fusion feature
- a type identification unit configured to invoke the identification network to identify the group type of the target group based on the first fusion feature and the second fusion feature.
- the group type identification model further includes a splicing network, and the type identification unit is configured to:
- the identification network is invoked to identify the group type of the target group based on the splicing feature.
- the training process of the group type identification model includes:
- the group type recognition model is trained according to the difference between the sample type and the prediction type.
- the first screening network is used for:
- the graph structure feature includes an association degree between any two user nodes among the multiple user nodes, and the feature acquisition module is configured to:
- co-occurrence times of any two user identities in the target group where the co-occurrence times refer to publishing content in the target group based on the any two user identities within multiple reference time periods the number of times;
- the degree of association between any two user identifiers is determined, and the association degree feature is positively correlated with the co-occurrence times.
- the user features include user behavior features and user attribute features
- the feature acquisition module is configured to:
- the user social network includes multiple registered user identities
- the user social network obtain the user behavior characteristics of the multiple user identifiers
- the user attribute features of the multiple user IDs are acquired.
- a computer device in another aspect, includes a processor and a memory, the memory stores at least one computer program, and the at least one computer program is loaded and executed by the processor to implement the The operations performed in the group type identification method described in the above aspects.
- a computer-readable storage medium is provided, and at least one computer program is stored in the computer-readable storage medium, and the at least one computer program is loaded and executed by a processor to implement the above-mentioned aspects.
- a computer program product or computer program comprising computer program code, the computer program code being stored in a computer-readable storage medium, the processor of the computer device from A computer-readable storage medium reads the computer program code, and the processor executes the computer program code, so that the computer device implements the operations performed in the group type identification method described in the above aspects.
- the graph structure feature and the user node feature are taken into consideration. Compared with only acquiring user information in the prior art, the amount of information is increased, so that the obtained The attention parameter can more accurately reflect the importance of user nodes in the graph structure, so that when multiple first user nodes are screened according to the reference attention parameters, the more important user nodes can be accurately selected.
- the user characteristics and graph structure characteristics of user nodes are used to identify the group type of the target group to improve the recognition accuracy, and at the same time, unimportant user nodes are discarded to reduce the amount of data processed and improve the processing speed.
- FIG. 1 is a flowchart of a method for identifying a group type provided by an embodiment of the present application
- FIG. 2 is a schematic structural diagram of a group type identification model provided by an embodiment of the present application.
- FIG. 3 is a schematic structural diagram of another group type identification model provided by an embodiment of the present application.
- FIG. 4 is a schematic structural diagram of another group type identification model provided by an embodiment of the present application.
- FIG. 6 is a schematic structural diagram of another group type identification model provided by an embodiment of the present application.
- FIG. 7 is a schematic structural diagram of a group type identification device provided by an embodiment of the present application.
- FIG. 8 is a schematic structural diagram of another group type identification device provided by an embodiment of the present application.
- FIG. 9 is a schematic structural diagram of a terminal provided by an embodiment of the present application.
- FIG. 10 is a schematic structural diagram of a server provided by an embodiment of the present application.
- first and second used in this application may be used herein to describe various concepts, but these concepts are not limited by these terms unless otherwise specified. These terms are only used to distinguish one concept from another.
- the first user node may be referred to as the second user node
- the second user node may be referred to as the first user node.
- the terms “at least one”, “plurality”, “each”, “any one”, etc. used in this application at least one includes one, two or more, multiple includes two or more, each includes Each refers to each of the corresponding plurality, and any refers to any one of the plurality.
- the multiple user nodes include 3 user nodes, and each user node refers to each of the 3 user nodes, and any refers to any one of the 3 user nodes, and can be the third user node. One, it could be the second, it could be the third.
- the type of the group is determined according to user information corresponding to multiple user identifiers in the group.
- the user information includes user avatar, user label, age, gender and other information.
- this method of determining the group type based on the user information is based on a small amount of data, resulting in a low accuracy rate of the determined group type.
- the group type identification method provided by the embodiment of the present application is applied in a group type identification scenario.
- the group type identification method provided by the embodiment of the present application is used to identify whether the target group is a specific group, and if it is determined that the target group is a specific group group, the target group will be blocked or otherwise processed to avoid losses to users.
- the methods provided in the embodiments of the present application are executed by a computer device, where the computer device is a terminal or a server.
- the terminal is a portable, pocket-sized, hand-held and other various types of terminals, such as a mobile phone, a computer, a tablet computer, and the like.
- a server is an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or provides cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, Cloud servers for basic cloud computing services such as domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms.
- CDN Content Delivery Network
- FIG. 1 is a flowchart of a group type identification method provided by an embodiment of the present application.
- the execution body of the embodiments of the present application is a computer device. Referring to Figure 1, the method includes the following steps:
- a target group including multiple user identifiers is used as an example for description.
- the first user node is a node corresponding to the user identifier in the target group, and the first target graph is constructed according to the association relationship between a plurality of first user nodes.
- the first target graph is the initial target graph; in the case where multiple first user nodes are user nodes corresponding to some user identities in the target group, that is, multiple first user nodes have been obtained after screening, then the first target graph is screened The resulting target image.
- the first graph structural feature is used to describe the first target graph, and the first target graph includes a plurality of first user nodes and connecting lines between the plurality of first user nodes. The connecting line between them indicates that there is an association relationship between any two first user nodes.
- the first user feature is used to describe the user corresponding to the user identification, and the first user feature includes at least one of a user behavior feature or a user attribute feature.
- the attention parameter represents the importance degree of the first user node in the first target graph, that is, represents the importance degree of the user ID corresponding to the first user node in the target group.
- the importance of user identity in the target group is positively correlated with the attention parameter. For example, users such as group owners and administrators in the target group are more important in the target group, and the attention parameters of the user nodes corresponding to these users will also be larger.
- a plurality of first user nodes are screened, and a plurality of second user nodes with larger attention parameters are selected from the plurality of first user nodes, and then according to the second user nodes with larger attention parameters Nodes are used for processing, thereby discarding some unimportant information in the processing process, ensuring that while reducing the amount of data, the loss of important information is reduced.
- the second target graph is constructed according to the association relationship between multiple second user nodes, and the second graph structural feature is used to describe the second target graph.
- the second target graph is a subgraph of the first target graph.
- the second target graph includes a plurality of second user nodes and connecting lines between the plurality of second user nodes.
- the second target graph includes a plurality of second user nodes.
- the connection lines between the nodes are the same as the connection lines of the plurality of second user nodes in the first target graph.
- the group type refers to the type to which the target group belongs, and different group types can be divided according to different division standards. For example, according to the occupation of the user corresponding to the user ID in the target group, a work group and a non-work group are divided; according to whether the users corresponding to the user ID in the target group are relatives, a family group and a non-work group are divided. Family group; according to whether the behavior of the user corresponding to the user ID in the target group in the target group belongs to a specific behavior, a specific group and a non-specific group are divided. For example, certain conduct includes conduct involving pornography, gambling, fraud, etc.
- the method provided by the embodiment of the present application considers the graph structure feature and the user node feature to obtain the attention parameter of each first user node. Compared with only obtaining user information in the prior art, the amount of information is increased, so that the obtained The attention parameter can more accurately reflect the importance of user nodes in the graph structure, so that when multiple first user nodes are screened according to the reference attention parameters, the more important user nodes can be accurately selected.
- the user characteristics and graph structure characteristics of user nodes are used to identify target groups to improve the recognition accuracy. At the same time, discarding unimportant user nodes can reduce the amount of data processed and improve the processing speed.
- the computer device invokes the group type identification model to identify the group type of the target group.
- the model structure of the group type identification model is described below.
- FIG. 2 is a schematic structural diagram of a group type identification model provided by an embodiment of the present application.
- the group type recognition model includes an input network 201 , a first attention network 202 , a first screening network 203 and a recognition network 204 .
- the input network 201 is connected to the first attention network 202
- the first attention network 202 is connected to the first screening network 203
- the first screening network 203 is connected to the recognition network 204 .
- the input network 201 is used to obtain the input graph structure features and user features
- the first attention network 202 is used to obtain the attention parameters of each user node
- the first screening network 203 is used to filter according to the obtained attention parameters.
- important user nodes are selected
- the identification network 204 is configured to identify the group type according to the user characteristics of the filtered user nodes and the corresponding graph structure characteristics.
- the group type identification model further includes a first convolutional network 205 , a second attention network 206 , a second screening network 207 and a splicing network 208 .
- the first convolutional network 205 is connected with the first screening network 203 and the second attention network 206
- the second attention network 206 is connected with the second screening network 207
- the second screening network 207 is connected with the recognition network 204
- the splicing network 208 is connected to the first screening network 203, the second screening network 207 and the identification network 204.
- the first convolutional network 205 is used to further process the user characteristics of the more important user nodes screened by the first screening network 203
- the second attention network 206 is used to obtain the attention of each screened user node parameter
- the second screening network 207 is used to further screen the screened user nodes according to the acquired attention parameters
- the splicing network 208 is used to splicing the user characteristics of the user nodes screened by the first screening network 203 and the second screening network.
- 207 User characteristics of the user nodes that are filtered out again.
- the user node is screened twice as an example for illustration.
- the user node can be screened three times, four times or even more times to obtain more information. quantity.
- three attention networks and three screening networks are used to screen user nodes three times.
- the splicing network is used to splicing the user characteristics of the user nodes screened by the three screening networks.
- a convolution network is also included before the first attention network, that is, the input user features and graph structure features are first subjected to convolution processing, and the first attention network User features and graph structure features to obtain attention parameters.
- the following describes an example of invoking a group type identification model to identify the type of a target group by using the embodiment shown in FIG. 5 .
- FIG. 5 is a flowchart of a group type identification method provided by an embodiment of the present application.
- the execution body of the embodiments of the present application is a computer device. Referring to Figure 5, the method includes the following steps:
- the embodiments of the present application are described by taking the first target graph as an initially constructed target graph as an example.
- corresponding user nodes are constructed according to multiple user identifiers in the target group, and then multiple user nodes are connected together according to the association relationship between multiple user identifiers to form a first target graph.
- the first target graph is an isomorphic graph, that is, the first target graph includes only one type of node, the user node, and the user nodes are connected according to the same type of association relationship, that is, the connection in the first target graph line is the same type of connection line.
- association relationship between any two user nodes is determined according to the association degree feature between the any two user nodes, and the association degree feature represents the degree of intimacy between the users corresponding to the any two user nodes.
- the co-occurrence times of any two user identities in the target group are obtained, and the correlation feature between any two user identities is determined according to the co-occurrence times.
- the correlation feature can be called the correlation degree
- the number of co-occurrences refers to the number of times that any two user IDs jointly publish content in the target group within multiple reference time periods, and the correlation degree is positively correlated with the number of co-occurrences. , that is, the greater the number of co-occurrences, the greater the degree of association between the two user identities, and the smaller the number of co-occurrences, the smaller the degree of association between the two user identities.
- the number of co-occurrences is directly used as the degree of association.
- the reference time period is 5 minutes.
- the number of co-occurrences is incremented by 1, and in the case that only one user speaks, or when neither user speaks, the number of co-occurrences does not increase.
- the degree of association can be determined according to the number of co-occurrences.
- the number of co-occurrences is set to 0, thus avoiding chance cases. For example, if a group owner in a target group publishes a message, all users in the target group except the group owner may reply to the message. At this time, many users in the target group will speak at the same time, but these may not have a close relationship between users.
- the first target graph is an homogeneous graph for description.
- the first target graph is a heterogeneous graph.
- the first target graph includes the user A node and a user type node, the user nodes are connected according to the same type of association relationship, and the user node and the user type node are connected according to the type of the user corresponding to the user node.
- the first graph structural feature represents a plurality of user nodes in the first target graph and the association relationship between the plurality of user nodes, and the first graph structural feature includes a relationship between any two user nodes in the plurality of user nodes. Correlation.
- the first user feature includes a user behavior feature and a user attribute feature, the user behavior feature represents the user's historical behavior, and the user attribute feature represents the user's own attribute.
- a user social network is obtained, and according to the user social network, user behavior characteristics identified by multiple users in the target group are obtained.
- the user social network includes multiple registered user identifiers.
- the user behavior characteristics identified by the user are acquired according to the user's social network in a graph embedding manner.
- the graph embedding is used to represent each node in the graph as a dense vector in a low-dimensional space, and the obtained dense vector is used as the feature information of the node.
- the core idea of the graph embedding is to preserve the intrinsic structural properties of the graph structure, that is, Keep nodes connected in a graph close to each other in a vector space.
- graph embedding methods include DeepWalk (a method for generating node representations in a network) and Node2Vec (a model for generating node vectors in a network) and other node embedding methods.
- Deepwalk node embedding method As an example, according to the user's social network, starting from each user node in the graph, according to the user's social network and the user's connection weight, randomly walk multiple trajectories, and use all the trajectories as the corpus input. To the word2vec word vector embedding model, the word2vec word vector embedding model is used for processing, and finally the user behavior characteristics of each user node are obtained.
- user attribute features of multiple user IDs in the target group are acquired according to user portrait information corresponding to multiple user IDs in the target group.
- the user portrait information includes information such as user portrait, dynamic information published by the user, user age, gender, and geographic location where the user is located.
- a vector is used to represent the first user characteristics
- a matrix is used to represent multiple first user characteristics, that is, a matrix is formed by splicing the multiple first user characteristics together. For example, a certain row or a certain column in the matrix represents the first user characteristic of a first user node.
- a matrix is used to represent the structural feature of the first graph, and an element of each position in the matrix represents an association relationship between corresponding two user nodes.
- the elements in the third row and the fourth column in the matrix represent the association relationship between the third user node and the fourth user node.
- the following formula is used to obtain the association relationship between any two first user nodes:
- a ij log(C ij );
- a ij represents the association relationship between the i-th first user node and the j-th user node
- C ij represents the common relationship between the user ID corresponding to the i-th first user node and the user ID corresponding to the j-th user node The number of occurrences.
- the first attention network is at least one GNN (Graph Neural Network, graph neural network). Use any of the following formulas to obtain the attention parameter:
- Z represents the attention parameter
- X represents multiple first user features
- A represents the first graph structure feature
- ⁇ is the reference value
- GNN( ) represents the convolution processing of the first user feature and the first graph structure feature
- m denotes the mth GNN
- M denotes the number of GNNs.
- One GNN is used in the first and second formula above, two GNNs are used in the third formula above, and M GNNs are used in the fourth formula above.
- the two GNNs are connected in sequence, that is, the first GNN is used for one processing, and then the second GNN is used for the second processing based on the first processing to obtain the attention parameters.
- the M GNNs process the first user feature and the first graph structure feature respectively, and average the M attention parameters obtained by processing to obtain the final attention parameter.
- Invoke the first screening network select a plurality of second user nodes from the plurality of first user nodes, and construct a second target graph according to the association relationship among the plurality of second user nodes.
- the attention parameters of the plurality of second user nodes are greater than the attention parameters of the unselected first user nodes.
- the first screening network is called, the number of multiple first user nodes is multiplied by the reference ratio to obtain the reference number, and the attention parameters of the multiple first user nodes are sorted in descending order. Arrange in order, select the reference number of attention parameters arranged in the front, and use the first user node corresponding to the selected attention parameter as the second user node. After the plurality of second user nodes are selected, a second target graph is formed according to the association relationship between the selected plurality of second user nodes.
- the following formula is used to obtain the first user features of multiple second user nodes and the second graph structure features of the second target graph:
- X′ represents the first user features of multiple second user nodes
- X idx represents the first user feature of the second user node selected from the first user features of multiple first user nodes
- a l represents the first user feature of the second user node.
- the second graph structural feature, A idx, idx represents the association relationship corresponding to a plurality of second user nodes selected from the first graph structural feature.
- the following formula is used to adjust the first user characteristics of multiple second user nodes to obtain the adjusted first user characteristics:
- X l represents the adjusted first user feature
- Z represents the attention parameters of multiple second user nodes
- ⁇ represents the bitwise product, that is, the first user feature of each second user node is multiplied by the attention parameter .
- the embodiment of the present application is only described by taking multiple screening of user nodes as an example.
- a plurality of second After the user node, a plurality of second After the user node, the group type of the target group is identified directly based on the user features of the plurality of second user nodes and the second graph structure feature for constructing the second target graph, and subsequent steps are not performed.
- the first convolutional network is GCN (Graph Convolutional Networks, graph convolutional network), and the following formula is used to determine the second user feature after convolution processing:
- X l+1 represents the second user features of multiple second user nodes
- X l represents the first user features of multiple second user nodes
- a l represents the second graph structure feature
- W l+1 represents the first user feature of the multiple second user nodes.
- the model parameters in the convolutional network, ⁇ is the reference value.
- Invoke the second screening network select a plurality of third user nodes from the plurality of second user nodes, and construct a third target graph according to the association relationship among the plurality of third user nodes.
- steps 506 to 507 are the same as the implementations of the above-mentioned steps 503 to 504, and are not repeated here.
- first user features and second graph structure features of multiple second user nodes are first fused to obtain first fusion features; multiple third users are fused.
- the second user feature and the third graph structure feature of the node are used to obtain the second fusion feature; the identification network is invoked to identify the group type of the target group based on the first fusion feature and the second fusion feature.
- an average processing is performed on the first user characteristics and the second graph structure characteristics of the plurality of second user nodes, that is, based on the first user characteristics of the plurality of second user nodes
- the average user feature and the second graph structure feature are averaged to obtain the average user feature corresponding to the multiple second user nodes; the average user feature and the largest user feature among the first user features of the multiple second user nodes are spliced to obtain the first fusion feature.
- the second graph structure feature includes the association relationship between multiple second user nodes, for any second user node, the second user node and other second user nodes can be determined from the second graph structure feature.
- the relationship between user nodes, when calculating the mean value, the first user feature of the second user node and the relationship between the second user node and other second user nodes are taken as a whole.
- the first user characteristics of the plurality of second user nodes and the plurality of association relationships are averaged.
- s represents the first fusion feature
- N represents the number of second user nodes
- x i represents the first user feature of the ith second user node and the relationship between the ith second user node and other second user nodes.
- represents the splicing of the features before
- the average processing is performed on the second user feature and the third graph structure feature of multiple third user nodes. , that is, based on the average of the second user features and the third graph structure features of multiple third user nodes, the average user features corresponding to multiple third user nodes are obtained; The largest user feature among the user features is obtained to obtain the second fusion feature.
- the first fusion feature and the second fusion feature are spliced to obtain a splicing feature corresponding to the target group, and based on the splicing feature, a group type corresponding to the target group is identified.
- the splicing feature is the feature representing the target group.
- the group type identification model further includes a first fusion network, a second fusion network, and a splicing network, that is, the first fusion network is called, and a plurality of second fusion networks are called.
- the first user feature and the second graph structure feature of the user node are used to obtain the first fusion feature;
- the second fusion network is called to fuse the second user feature and the third graph structure feature of multiple third user nodes to obtain the second fusion feature. ;
- the recognition network is a classifier
- the classifier includes a multi-layer perceptron
- the output of the recognition network is 0 or 1.
- the output of the network is 0, it indicates that the target group is not a specific group, and when the output of the recognition network is 1, it indicates that the target group is a specific group.
- E represents the splicing feature
- MLP( ⁇ ) means that MLP (Multi-Layer Perceptron) is used to process the splicing feature.
- the output of the identification network is a probability.
- the output probability is greater than the reference probability, it indicates that the target group is a specific group, and when the output probability is not greater than the reference probability, it indicates the target group.
- Groups are not specific groups.
- the computer device can directly Structural features and multiple first user features, obtain the attention parameters of each first user node, select multiple second user nodes from multiple first user nodes, and according to the association relationship between multiple second user nodes Construct a second target graph, adjust the first user features of multiple second user nodes according to the second graph structural features of the second target graph, and obtain the second user features of multiple second user nodes, based on the second graph structural features and multiple second user features, obtain the attention parameters of each second user node, select multiple third user nodes from multiple second user nodes, and construct the first user node according to the association relationship between multiple third user nodes.
- Three target graphs identifying target groups based on first user features of multiple second user nodes, second graph structural features, second user features of multiple third user nodes, and third graph structural features of the third target graph group type.
- the graph structure feature and the user node feature are taken into consideration. Compared with only acquiring user information in the prior art, the amount of information is increased, so that the obtained The attention parameter can more accurately reflect the importance of user nodes in the graph structure, so that when multiple first user nodes are screened according to the reference attention parameters, the more important user nodes can be accurately selected.
- the user characteristics and graph structure characteristics of user nodes are used to identify the group type of the target group to improve the recognition accuracy, and at the same time, unimportant user nodes are discarded to reduce the amount of data processed and improve the processing speed.
- the user nodes are screened multiple times to obtain user features and graph structure features at different levels.
- the user features and graph structure features at different levels are considered, which further improves the accuracy of identification. Rate.
- the group type identification model needs to be trained first.
- the training process of the group type identification model includes: obtaining the sample type of the sample group and the sample graph structure of the sample graph Features and sample user characteristics of multiple sample user nodes in the sample target graph; call the group type identification model to identify the prediction type of the sample group based on the sample graph structural characteristics and the sample user characteristics of multiple sample user nodes; based on the sample type and predict the difference between the types, training the cohort type recognition model.
- the sample user node is a node corresponding to the sample user ID, and the sample graph is constructed according to the association relationship between multiple sample user IDs in the sample group.
- the target group can be used to continue training the group type identification model.
- the above training process is only an example of one training, and in another embodiment, the group type identification model can be iteratively trained for multiple times.
- a keyword filtering technology is used to determine whether the content published by the user includes a specific word, and if it includes a specific word, then It is considered that the target group belongs to a specific group, but if the user uses other non-specific vocabulary to replace the corresponding specific vocabulary, the keyword filtering technology cannot be used to detect it, and it is impossible to determine whether it is a specific vocabulary. Accuracy is low.
- the method of user reporting is adopted, that is, after the user reports the target group, the technician will manually review it to determine whether the target group belongs to a specific group, but this method depends on the user Affected by the user's reporting behavior, the identification accuracy of the target group is also low.
- the method provided by the embodiment of the present application is not affected by the vocabulary used by the user and the reporting behavior of the user, and can directly The user characteristics of the node identify the group type of the target group, which improves the identification accuracy.
- the pooling function is used to read the information of all user nodes in the target group, but when there are many user nodes, it is difficult to read all the information by using the pooling function, which will lead to loss of a lot of information , and the more user nodes, the slower the processing speed.
- user nodes can be screened according to the attention parameter, and only the information of the screened user nodes can be read out, and the information of important nodes can be guaranteed to be retained according to the size of the attention parameter, which will not cause A large amount of information is lost, and the processing speed is improved while ensuring the amount of information.
- FIG. 7 is a schematic structural diagram of a group type identification device provided by an embodiment of the present application. Referring to Figure 7, the device includes:
- the feature acquisition module 701 is used to acquire the first graph structure feature of the first target graph and the first user features of a plurality of first user nodes in the first target graph, where the first user nodes are corresponding to the user IDs in the target group. node, and the first target graph is constructed according to the association relationship between a plurality of first user nodes;
- the first attention obtaining module 702 is configured to obtain the attention parameter of each first user node in the first target graph based on the first graph structure feature and the plurality of first user features, where the attention parameter indicates that the first user node is in The degree of importance in the first objective graph;
- the first screening module 703 is configured to select a plurality of second user nodes from a plurality of first user nodes according to the obtained plurality of attention parameters, and the attention parameters of the plurality of second user nodes are greater than the unselected first user nodes.
- the type identification module 704 is used to identify the group type of the target group based on the first user characteristics of the plurality of second user nodes and the second graph structure characteristics of the second target graph, and the second target graph is based on the plurality of second target graphs.
- the association relationship between user nodes is constructed.
- the graph structure feature and the user node feature are considered. Compared with only acquiring user information in the prior art, the amount of information is increased, so that the obtained The attention parameter can more accurately reflect the importance of user nodes in the graph structure, so that when multiple first user nodes are screened according to the reference attention parameters, the more important user nodes can be accurately selected.
- the information of user nodes is used to identify the group type of the target group to improve the recognition accuracy, and at the same time, unimportant user nodes are discarded to reduce the amount of data processed and improve the processing speed.
- the apparatus further includes:
- a feature adjustment module 705, configured to adjust the first user features of a plurality of second user nodes based on the structural features of the second graph to obtain second user features of a plurality of second user nodes;
- the second attention acquisition module 706 is configured to process based on the second graph structure feature and a plurality of second user features, and obtain the attention parameter of each second user node;
- the second screening module 707 is configured to select a plurality of third user nodes from the plurality of second user nodes, and the attention parameters of the plurality of third user nodes are greater than the attention parameters of the unselected second user nodes.
- the type identification module 704 is configured to, based on the first user characteristics of the plurality of second user nodes, the second graph structure characteristics, the second user characteristics of the plurality of third user nodes, and the third target
- the third graph structural feature of the graph identifies the group type of the target group, and the third target graph is constructed according to the association relationship between a plurality of third user nodes.
- the type identification module 704 includes:
- a first fusion unit 7041 configured to fuse first user features and second graph structure features of multiple second user nodes to obtain first fusion features
- a second fusion unit 7042 configured to fuse the second user features and the third graph structure features of multiple third user nodes to obtain second fusion features
- the type identification unit 7043 is configured to identify the group type of the target group based on the first fusion feature and the second fusion feature.
- the first fusion unit 7041 is used for:
- a first fusion feature is obtained by splicing the average user feature and the largest user feature among the first user features of the plurality of second user nodes.
- the type identification unit 7043 is used for:
- the group type of the target group is identified.
- the group type recognition model includes a first attention network, a first screening network and a recognition network,
- the first attention obtaining module 702 is used to call the first attention network, and obtain the attention parameters of each first user node based on the first graph structure feature and the plurality of first user features;
- a first screening module 703, configured to invoke a first screening network to select a plurality of second user nodes from a plurality of first user nodes;
- the type identification module 704 is configured to invoke the identification network to identify the group type of the target group based on the first user characteristics and the second graph structure characteristics of the plurality of second user nodes.
- the group type identification model further includes a first convolutional network, a second attention network, and a second screening network.
- the apparatus further includes:
- the feature adjustment module 705 is configured to call the first convolutional network, and based on the second graph structure feature, adjust the first user features of the plurality of second user nodes to obtain the second user features of the plurality of second user nodes;
- the second attention obtaining module 706 is configured to call the second attention network, and obtain the attention parameters of each second user node in the second target graph based on the second graph structure feature and the plurality of second user features;
- the second screening module 707 is configured to invoke the second screening network, and select a plurality of third user nodes from the plurality of second user nodes, and the attention parameters of the plurality of third user nodes are greater than those of the unselected second user nodes. attention parameter.
- the type identification module 704 is configured to invoke the identification network, based on the first user characteristics of the plurality of second user nodes, the second graph structure characteristics, and the second users of the plurality of third user nodes.
- the feature and the third graph structure feature of the third target graph identify the group type of the target group, and the third target graph is constructed according to the association relationship between a plurality of third user nodes.
- the group type identification model further includes a first fusion network and a second fusion network.
- the type identification module 704 includes:
- the first fusion unit 7041 is configured to call the first fusion network, and fuse the first user features and the second graph structure features of multiple second user nodes to obtain the first fusion features;
- the second fusion unit 7042 is configured to call the second fusion network, and fuse the second user features and the third graph structure features of a plurality of third user nodes to obtain the second fusion features;
- the type identification unit 7043 is configured to invoke the identification network to identify the group type of the target group based on the first fusion feature and the second fusion feature.
- the group type identification model further includes a splicing network.
- the type identification unit 7043 is used for:
- the recognition network is called to identify the group type of the target group based on the splicing features.
- the training process of the group type recognition model includes:
- the sample user node is the node corresponding to the sample user ID, and the sample graph is based on the The association relationship between multiple sample user IDs is constructed;
- the cohort type recognition model is trained based on the difference between sample type and prediction type.
- the first screening network 703 is used for:
- the graph structure feature includes a correlation feature between any two user nodes among the multiple user nodes.
- the feature acquisition module 701 is configured to:
- co-occurrence times of any two user identities in the target group and the co-occurrence times refer to the number of times that content is jointly published in the target group based on any two user identities within multiple reference time periods;
- the degree of association between any two user IDs is determined, and the degree of association is positively correlated with the number of co-occurrences.
- the user features include user behavior features and user attribute features.
- the feature acquisition module 701 is used for:
- the user's social network includes multiple registered user IDs
- the user attribute features of the multiple user IDs are acquired.
- the group type identification device when the group type identification device provided in the above embodiment identifies the group type, only the division of the above functional modules is used as an example for illustration. In practical applications, the above functions can be assigned to different functions Module completion, that is, dividing the internal structure of the computer device into different functional modules to complete all or part of the functions described above.
- the group type identification device and the group type identification method embodiments provided by the above embodiments belong to the same concept, and the specific implementation process thereof is detailed in the method embodiments, which will not be repeated here.
- the embodiment of the present application also provides a computer device, the computer device includes a processor and a memory, and at least one computer program is stored in the memory, and the at least one computer program is loaded and executed by the processor to implement the following steps:
- the first target graph is constructed according to the association relationship between the plurality of first user nodes;
- an attention parameter of each first user node is obtained, where the attention parameter indicates the importance of the first user node in the first target graph degree;
- the attention parameters of the plurality of second user nodes are greater than the attention parameters of the unselected first user nodes
- the relationship between the two user nodes is constructed.
- the at least one computer program is loaded and executed by the processor to implement the following steps:
- a plurality of third user nodes are selected from the plurality of second user nodes, and the attention parameters of the plurality of third user nodes are greater than the attention parameters of the unselected second user nodes.
- the at least one computer program is loaded and executed by the processor to implement the following steps:
- the second graph structure feature, the second user feature of the plurality of third user nodes, and the third graph structure feature of the third target graph identify the the group type of the target group, and the third target graph is constructed according to the association relationship between the plurality of third user nodes.
- the at least one computer program is loaded and executed by the processor to implement the following steps:
- a group type of the target group is identified.
- the at least one computer program is loaded and executed by a processor to implement the following steps: averaging based on the first user characteristics of the plurality of second user nodes and the second graph structure characteristics, get the average user characteristics;
- the first fusion feature is obtained by splicing the average user feature and the largest user feature among the first user features of the plurality of second user nodes.
- the at least one computer program is loaded and executed by the processor to implement the following steps:
- a group type of the target group is identified.
- the group type recognition model includes a first attention network, a first screening network and a recognition network, and the at least one computer program is loaded and executed by the processor to implement the following steps:
- the selecting multiple second user nodes from the multiple first user nodes includes:
- the identification network is invoked to identify the group type of the target group based on the first user characteristics of the plurality of second user nodes and the second graph structure characteristics.
- the group type identification model further includes a first convolution network, a second attention network and a second screening network, and the at least one computer program is loaded and executed by the processor to implement the following steps :
- the attention parameters of the plurality of third user nodes are greater than the attention of the unselected second user nodes parameter.
- the at least one computer program is loaded and executed by the processor to implement the following steps:
- the graph structure feature identifies the group type of the target group, and the third target graph is constructed according to the association relationship between the plurality of third user nodes.
- the group type identification model further includes a first fusion network and a second fusion network
- the at least one computer program is loaded and executed by the processor to implement the following steps:
- the identification network is invoked to identify the group type of the target group based on the first fusion feature and the second fusion feature.
- the group type identification model further includes a splicing network, and the at least one computer program is loaded and executed by the processor to implement the following steps:
- the identification network is invoked to identify the group type of the target group based on the splicing feature.
- the at least one computer program is loaded and executed by the processor to implement the following steps:
- the group type recognition model is trained according to the difference between the sample type and the prediction type.
- the graph structure feature includes the degree of association between any two user nodes in the plurality of user nodes, and the at least one computer program is loaded and executed by the processor to implement the following steps:
- co-occurrence times of any two user identities in the target group where the co-occurrence times refer to publishing content in the target group based on the any two user identities within multiple reference time periods the number of times;
- the degree of association between any two user identifiers is determined, and the association degree is positively correlated with the co-occurrence times.
- the at least one computer program is loaded and executed by the processor to implement the following steps:
- the user characteristics include user behavior characteristics and user attribute characteristics
- the at least one computer program is loaded and executed by the processor to implement the following steps:
- the user social network includes multiple registered user identities
- the user social network obtain the user behavior characteristics of the multiple user identifiers
- the user attribute features of the multiple user IDs are acquired.
- FIG. 9 is a schematic structural diagram of a terminal 900 provided by an embodiment of the present application.
- the terminal 900 includes: a processor 901 and a memory 902 .
- the processor 901 may include one or more processing cores, such as a 4-core processor, an 8-core processor, and the like.
- the processor 901 may also include a main processor and a co-processor.
- the main processor is a processor used to process data in the wake-up state, also called CPU (Central Processing Unit, central processing unit); the co-processor is A low-power processor for processing data in a standby state.
- the processor 901 may be integrated with a GPU (Graphics Processing Unit, image processor), and the GPU is used for rendering and drawing the content that needs to be displayed on the display screen.
- GPU Graphics Processing Unit, image processor
- Memory 902 may include one or more computer-readable storage media, which may be non-transitory.
- the non-transitory computer-readable storage medium in the memory 902 is used to store at least one computer program, and the at least one computer program is used to be executed by the processor 901 to implement the methods provided by the method embodiments in this application. Group type identification method.
- the terminal 900 may optionally further include: a peripheral device interface 903 and at least one peripheral device.
- the processor 901, the memory 902 and the peripheral device interface 903 may be connected through a bus or a signal line.
- Each peripheral device can be connected to the peripheral device interface 903 through a bus, a signal line or a circuit board.
- the peripheral device includes: at least one of a radio frequency circuit 904 , a display screen 905 , a camera assembly 906 , an audio circuit 907 and a power supply 908 .
- the peripheral device interface 903 may be used to connect at least one peripheral device related to I/O (Input/Output) to the processor 901 and the memory 902 .
- processor 901, memory 902, and peripherals interface 903 are integrated on the same chip or circuit board; in some other embodiments, any one of processor 901, memory 902, and peripherals interface 903 or The two can be implemented on a separate chip or circuit board, which is not limited in this embodiment.
- the radio frequency circuit 904 is used for receiving and transmitting RF (Radio Frequency, radio frequency) signals, also called electromagnetic signals.
- the radio frequency circuit 904 communicates with the communication network and other communication devices through electromagnetic signals.
- the radio frequency circuit 904 converts electrical signals into electromagnetic signals for transmission, or converts received electromagnetic signals into electrical signals.
- the display screen 905 is used for displaying UI (User Interface, user interface).
- the UI can include graphics, text, icons, video, and any combination thereof.
- the display screen 905 also has the ability to acquire touch signals on or above the surface of the display screen 905 .
- the touch signal may be input to the processor 901 as a control signal for processing.
- the camera assembly 906 is used to capture images or video.
- the camera assembly 906 includes a front camera and a rear camera.
- Audio circuitry 907 may include a microphone and speakers.
- the microphone is used to collect the sound waves of the user and the environment, convert the sound waves into electrical signals, and input them to the processor 901 for processing, or to the radio frequency circuit 904 to realize voice communication.
- Power supply 908 is used to power various components in terminal 900 .
- the power source 908 may be alternating current, direct current, primary batteries, or rechargeable batteries.
- FIG. 9 does not constitute a limitation on the terminal 900, and may include more or less components than shown, or combine some components, or adopt different component arrangements.
- the computer device is provided as a server.
- 10 is a schematic structural diagram of a server provided by an embodiment of the present application.
- the server 1000 may vary greatly due to different configurations or performance, and may include one or more processors (Central Processing Units, CPU) 1001 and a Or more than one memory 1002, wherein, at least one computer program is stored in the memory 1002, and the at least one computer program is loaded and executed by the processor 1001 to implement the methods provided by the above method embodiments.
- the server may also have components such as a wired or wireless network interface, a keyboard, and an input/output interface for input and output, and the server may also include other components for implementing device functions, which will not be described here.
- Embodiments of the present application further provide a computer-readable storage medium, where at least one computer program is stored in the computer-readable storage medium, and the at least one computer program is loaded and executed by a processor to implement the following steps:
- the first target graph is constructed according to the association relationship between the plurality of first user nodes;
- an attention parameter of each first user node is obtained, where the attention parameter indicates the importance of the first user node in the first target graph degree;
- the attention parameters of the plurality of second user nodes are greater than the attention parameters of the unselected first user nodes
- the relationship between the two user nodes is constructed.
- the at least one computer program is loaded and executed by the processor to implement the following steps:
- a plurality of third user nodes are selected from the plurality of second user nodes, and the attention parameters of the plurality of third user nodes are greater than the attention parameters of the unselected second user nodes.
- the at least one computer program is loaded and executed by the processor to implement the following steps:
- the second graph structure feature, the second user feature of the plurality of third user nodes, and the third graph structure feature of the third target graph identify the the group type of the target group, and the third target graph is constructed according to the association relationship between the plurality of third user nodes.
- the at least one computer program is loaded and executed by the processor to implement the following steps:
- a group type of the target group is identified.
- the at least one computer program is loaded and executed by a processor to implement the following steps: averaging based on the first user characteristics of the plurality of second user nodes and the second graph structure characteristics, get the average user characteristics;
- the first fusion feature is obtained by splicing the average user feature and the largest user feature among the first user features of the plurality of second user nodes.
- the at least one computer program is loaded and executed by the processor to implement the following steps:
- a group type of the target group is identified.
- the group type recognition model includes a first attention network, a first screening network and a recognition network, and the at least one computer program is loaded and executed by the processor to implement the following steps:
- the selecting multiple second user nodes from the multiple first user nodes includes:
- the identification network is invoked to identify the group type of the target group based on the first user characteristics of the plurality of second user nodes and the second graph structure characteristics.
- the group type identification model further includes a first convolution network, a second attention network and a second screening network, and the at least one computer program is loaded and executed by the processor to implement the following steps :
- the attention parameters of the plurality of third user nodes are greater than the attention of the unselected second user nodes parameter.
- the at least one computer program is loaded and executed by the processor to implement the following steps:
- the graph structure feature identifies the group type of the target group, and the third target graph is constructed according to the association relationship between the plurality of third user nodes.
- the group type identification model further includes a first fusion network and a second fusion network
- the at least one computer program is loaded and executed by the processor to implement the following steps:
- the identification network is invoked to identify the group type of the target group based on the first fusion feature and the second fusion feature.
- the group type identification model further includes a splicing network, and the at least one computer program is loaded and executed by the processor to realize the following steps:
- the identification network is invoked to identify the group type of the target group based on the splicing feature.
- the at least one computer program is loaded and executed by the processor to implement the following steps:
- the group type recognition model is trained according to the difference between the sample type and the prediction type.
- the graph structure feature includes the degree of association between any two user nodes in the plurality of user nodes, and the at least one computer program is loaded and executed by the processor to implement the following steps:
- co-occurrence times of any two user identities in the target group where the co-occurrence times refer to publishing content in the target group based on the any two user identities within multiple reference time periods the number of times;
- the degree of association between any two user identifiers is determined, and the association degree is positively correlated with the co-occurrence times.
- the at least one computer program is loaded and executed by the processor to implement the following steps:
- the user characteristics include user behavior characteristics and user attribute characteristics
- the at least one computer program is loaded and executed by the processor to implement the following steps:
- the user social network includes multiple registered user identities
- the user social network obtain the user behavior characteristics of the multiple user identifiers
- the user attribute features of the multiple user IDs are acquired.
- Embodiments of the present application also provide a computer program product or computer program, where the computer program product or computer program includes computer program code, and the computer program code is stored in a computer-readable storage medium.
- the processor of the computer device reads the computer program code from the computer-readable storage medium, and the processor executes the computer program code, so that the computer device implements the operations performed in the group type identification method of the foregoing embodiment.
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Signal Processing (AREA)
- Computer Networks & Wireless Communication (AREA)
- Computing Systems (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
Description
Claims (18)
- 一种群组类型识别方法,由计算机设备执行,所述方法包括:获取第一目标图的第一图结构特征和所述第一目标图中多个第一用户节点的第一用户特征,所述第一用户节点为目标群组中的用户标识对应的节点,所述第一目标图为根据所述多个第一用户节点之间的关联关系构建的;基于所述第一图结构特征和多个第一用户特征,获取每个第一用户节点的注意力参数,所述注意力参数表示所述第一用户节点在所述第一目标图中的重要程度;从所述多个第一用户节点中选取多个第二用户节点,所述多个第二用户节点的注意力参数大于未被选取的第一用户节点的注意力参数;基于所述多个第二用户节点的第一用户特征和第二目标图的第二图结构特征,识别所述目标群组的群组类型,所述第二目标图为根据所述多个第二用户节点之间的关联关系构建的。
- 根据权利要求1所述的方法,其中,所述从所述多个第一用户节点中选取多个第二用户节点之后,所述方法还包括:基于所述第二图结构特征,调整所述多个第二用户节点的第一用户特征,得到所述多个第二用户节点的第二用户特征;基于所述第二图结构特征和多个第二用户特征,获取每个第二用户节点的注意力参数;从所述多个第二用户节点中选多个第三用户节点,所述多个第三用户节点的注意力参数大于未被选取的第二用户节点的注意力参数。
- 根据权利要求2所述的方法,其中,所述基于所述多个第二用户节点的第一用户特征和第二目标图的第二图结构特征,识别所述目标群组的群组类型,包括:基于所述多个第二用户节点的第一用户特征、所述第二图结构特征、所述多个第三用户节点的第二用户特征和第三目标图的第三图结构特征,识别所述目标群组的群组类型,所述第三目标图为根据所述多个第三用户节点之间的关联关系构建的。
- 根据权利要求3所述的方法,其中,所述基于所述多个第二用户节点的第一用户特征、所述第二图结构特征、所述多个第三用户节点的第二用户特征和第三目标图的第三图结构特征,识别所述目标群组的群组类型,包括:融合所述多个第二用户节点的第一用户特征和所述第二图结构特征,得到第一融合特征;融合所述多个第三用户节点的第二用户特征和所述第三图结构特征,得到第二融合特征;基于所述第一融合特征和所述第二融合特征,识别所述目标群组的群组类型。
- 根据权利要求4所述的方法,其中,所述融合所述多个第二用户节点的第一用户特征和所述第二图结构特征,得到第一融合特征,包括:基于所述多个第二用户节点的第一用户特征和所述第二图结构特征求均值,得到平均用户特征;拼接所述平均用户特征与所述多个第二用户节点的第一用户特征中的最大用户特征,得到所述第一融合特征。
- 根据权利要求4所述的方法,其中,所述基于所述第一融合特征和所述第二融合特征,识别所述目标群组的群组类型,包括:拼接所述第一融合特征和所述第二融合特征,得到所述目标群组对应的拼接特征;基于所述拼接特征,识别所述目标群组的群组类型。
- 根据权利要求1所述的方法,其中,群组类型识别模型包括第一注意力网络、第一筛选网络和识别网络,所述基于所述第一图结构特征和多个第一用户特征,获取每个第一用户节点的注意力参数,包括:调用所述第一注意力网络,基于所述第一图结构特征和所述多个第一用户特征,获取所述每个第一用户节点的注意力参数;所述从所述多个第一用户节点中选取多个第二用户节点,包括:调用所述第一筛选网络,从所述多个第一用户节点中选取所述多个第二用户节点;所述基于所述多个第二用户节点的第一用户特征和第二目标图的第二图结构特征,识别所述目标群组的群组类型,包括:调用所述识别网络,基于所述多个第二用户节点的第一用户特征和所述第二图结构特征,识别所述目标群组的群组类型。
- 根据权利要求7所述的方法,其中,所述群组类型识别模型还包括第一卷积网络、第二注意力网络和第二筛选网络,所述调用所述第一筛选网络,从所述多个第一用户节点中选取所述多个第二用户节点之后,所述方法还包括:调用所述第一卷积网络,基于所述第二图结构特征,调整所述多个第二用户节点的第一用户特征,得到所述多个第二用户节点的第二用户特征;调用所述第二注意力网络,基于所述第二图结构特征和多个第二用户特征,获取每个第二用户节点的注意力参数;调用所述第二筛选网络,从所述多个第二用户节点中选取多个第三用户节点,所述多个第三用户节点的注意力参数大于未被选取的第二用户节点的注意力参数。
- 根据权利要求8所述的方法,其中,所述调用所述识别网络,基于所述多个第二用户节点的第一用户特征和所述第二图结构特征,识别所述目标群组的群组类型,包括:调用所述识别网络,基于所述多个第二用户节点的第一用户特征、所述第二图结构特征、所述多个第三用户节点的第二用户特征和第三目标图的第三图结构特征,识别所述目标群组的群组类型,所述第三目标图为根据所述多个第三用户节点之间的关联关系构建的。
- 根据权利要求9所述的方法,其中,所述群组类型识别模型还包括第一融合网络和第二融合网络,所述调用所述识别网络,基于所述多个第二用户节点的第一用户特征、所述第二图结构特征、所述多个第三用户节点的第二用户特征和第三目标图的第三图结构特征,识别所述目标群组的群组类型,包括:调用所述第一融合网络,融合所述多个第二用户节点的第一用户特征和所述第二图结构特征,得到第一融合特征;调用所述第二融合网络,融合所述多个第三用户节点的第二用户特征和所述第三图结构特征,得到第二融合特征;调用所述识别网络,基于所述第一融合特征和所述第二融合特征,识别所述目标群组的群组类型。
- 根据权利要求10所述的方法,其中,所述群组类型识别模型还包括拼接网络,所述调用所述识别网络,基于所述第一融合特征和所述第二融合特征,识别所述目标群组的群组类型,包括:调用所述拼接网络,拼接所述第一融合特征和所述第二融合特征,得到所述目标群组对应的拼接特征;调用所述识别网络,基于所述拼接特征,识别所述目标群组的群组类型。
- 根据权利要求7-11任一项所述的方法,其中,所述群组类型识别模型的训练过程包括:获取样本群组的样本类型、样本图的样本图结构特征和所述样本目标图中多个样本用户节点的样本用户特征,所述样本用户节点为所述样本用户标识对应的节点,所述样本图为根据样本群组中的多个样本用户标识之间的关联关系构建的;调用所述群组类型识别模型,基于所述样本图结构特征和所述多个样本用户节点的样本用户特征,识别所述样本群组的预测类型;根据所述样本类型和所述预测类型之间的差异,训练所述群组类型识别模型。
- 根据权利要求1所述的方法,其中,图结构特征包括多个用户节点中任两个用户节点之间的关联度,所述获取第一目标图的图结构特征,包括:获取所述目标群组中任两个用户标识的共同出现次数,所述共同出现次数是指在多个参考时间段内,基于所述任两个用户标识共同在所述目标群组中发布内容的次数;基于所述共同出现次数,确定所述任两个用户标识之间的关联度,所述关联度与所述共同出现次数呈正相关关系。
- 根据权利要求1所述的方法,其中,所述从所述多个第一用户节点中选取多个第二用户节点,包括:将所述多个第一用户节点的数量与参考比例相乘,得到参考数量;将所述多个第一用户节点的注意力参数按照从大到小的顺序排列,选取排列在前面的参考数量个注意力参数,将选取的注意力参数对应的第一用户节点作为所述第二用户节点。
- 根据权利要求1所述的方法,其中,用户特征包括用户行为特征和用户属性特征,获取所述第一目标图中多个第一用户节点的第一用户特征,包括:获取用户社交网络,所述用户社交网络包括已注册的多个用户标识;根据所述用户社交网络,获取所述多个用户标识的用户行为特征;根据所述多个用户标识对应的用户画像信息,获取所述多个用户标识的用户属性特征。
- 一种群组类型识别装置,其特征在于,所述装置包括:特征获取模块,用于获取第一目标图的第一图结构特征和所述第一目标图中多个第一用户节点的第一用户特征,所述第一用户节点为目标群组中的用户标识对应的节点,所述第一目标图为根据所述多个第一用户节点之间的关联关系构建的;第一注意力获取模块,用于基于所述第一图结构特征和多个第一用户特征,获取所述第一目标图中每个第一用户节点的注意力参数,所述注意力参数表示所述第一用户节点在所述第一目标图中的重要程度;第一筛选模块,用于从所述多个第一用户节点中选取多个第二用户节点,所述多个第二用户节点的注意力参数大于未被选取的第一用户节点的注意力参数;类型识别模块,用于基于所述多个第二用户节点的第一用户特征和第二目标图的第二图结构特征,识别所述目标群组的群组类型,所述第二目标图为根据所述多个第二用户节点之间的关联关系构建的。
- 一种计算机设备,其特征在于,所述计算机设备包括处理器和存储器,所述存储器中存储有至少一条计算机程序,所述至少一条计算机程序由所述处理器加载并执行,以实现如权利要求1至15任一权利要求所述的群组类型识别方法中所执行的操作。
- 一种计算机可读存储介质,其特征在于,所述计算机可读存储介质中存储有至少一条 计算机程序,所述至少一条计算机程序由处理器加载并执行,以实现如权利要求1至15任一权利要求所述的群组类型识别方法中所执行的操作。
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2023519589A JP2023544022A (ja) | 2021-01-04 | 2021-12-27 | グループタイプの識別方法、装置、コンピュータ機器及びコンピュータプログラム |
US17/963,919 US11916853B2 (en) | 2021-01-04 | 2022-10-11 | Group type identification method and apparatus, computer device, and medium |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110002127.7A CN114764480A (zh) | 2021-01-04 | 2021-01-04 | 群组类型识别方法、装置、计算机设备及介质 |
CN202110002127.7 | 2021-01-04 |
Related Child Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
US17/963,919 Continuation US11916853B2 (en) | 2021-01-04 | 2022-10-11 | Group type identification method and apparatus, computer device, and medium |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2022143505A1 true WO2022143505A1 (zh) | 2022-07-07 |
Family
ID=82260211
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/CN2021/141553 WO2022143505A1 (zh) | 2021-01-04 | 2021-12-27 | 群组类型识别方法、装置、计算机设备及介质 |
Country Status (4)
Country | Link |
---|---|
US (1) | US11916853B2 (zh) |
JP (1) | JP2023544022A (zh) |
CN (1) | CN114764480A (zh) |
WO (1) | WO2022143505A1 (zh) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116680633A (zh) * | 2023-05-06 | 2023-09-01 | 国网四川省电力公司广安供电公司 | 基于多任务学习的异常用户检测方法、***及存储介质 |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20170235812A1 (en) * | 2016-02-16 | 2017-08-17 | Microsoft Technology Licensing, Llc | Automated aggregation of social contact groups |
CN107770049A (zh) * | 2017-10-23 | 2018-03-06 | 林楚莲 | 一种被邀请用户获取群组信息的方法及*** |
CN110083791A (zh) * | 2019-05-05 | 2019-08-02 | 北京三快在线科技有限公司 | 目标群组检测方法、装置、计算机设备及存储介质 |
CN111738628A (zh) * | 2020-08-14 | 2020-10-02 | 支付宝(杭州)信息技术有限公司 | 一种风险群组识别方法及装置 |
Family Cites Families (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20130055099A1 (en) * | 2011-08-22 | 2013-02-28 | Rose Yao | Unified Messaging System with Integration of Call Log Data |
US10305758B1 (en) * | 2014-10-09 | 2019-05-28 | Splunk Inc. | Service monitoring interface reflecting by-service mode |
US10565229B2 (en) * | 2018-05-24 | 2020-02-18 | People.ai, Inc. | Systems and methods for matching electronic activities directly to record objects of systems of record |
-
2021
- 2021-01-04 CN CN202110002127.7A patent/CN114764480A/zh active Pending
- 2021-12-27 WO PCT/CN2021/141553 patent/WO2022143505A1/zh active Application Filing
- 2021-12-27 JP JP2023519589A patent/JP2023544022A/ja active Pending
-
2022
- 2022-10-11 US US17/963,919 patent/US11916853B2/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20170235812A1 (en) * | 2016-02-16 | 2017-08-17 | Microsoft Technology Licensing, Llc | Automated aggregation of social contact groups |
CN107770049A (zh) * | 2017-10-23 | 2018-03-06 | 林楚莲 | 一种被邀请用户获取群组信息的方法及*** |
CN110083791A (zh) * | 2019-05-05 | 2019-08-02 | 北京三快在线科技有限公司 | 目标群组检测方法、装置、计算机设备及存储介质 |
CN111738628A (zh) * | 2020-08-14 | 2020-10-02 | 支付宝(杭州)信息技术有限公司 | 一种风险群组识别方法及装置 |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116680633A (zh) * | 2023-05-06 | 2023-09-01 | 国网四川省电力公司广安供电公司 | 基于多任务学习的异常用户检测方法、***及存储介质 |
CN116680633B (zh) * | 2023-05-06 | 2024-01-26 | 国网四川省电力公司广安供电公司 | 基于多任务学习的异常用户检测方法、***及存储介质 |
Also Published As
Publication number | Publication date |
---|---|
US20230031045A1 (en) | 2023-02-02 |
CN114764480A (zh) | 2022-07-19 |
JP2023544022A (ja) | 2023-10-19 |
US11916853B2 (en) | 2024-02-27 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110363346B (zh) | 点击率预测方法、预测模型的训练方法、装置及设备 | |
WO2020156389A1 (zh) | 信息推送方法和装置 | |
WO2020093289A1 (zh) | 资源推荐方法、装置、电子设备及存储介质 | |
CN114036398B (zh) | 内容推荐和排序模型训练方法、装置、设备以及存储介质 | |
CN107273415A (zh) | 通过基于即时消息接发的主题索引的可搜索对等*** | |
WO2021068764A1 (zh) | 信息处理方法和装置 | |
US20230252070A1 (en) | Method and apparatus for training retrieval model, retrieval method and apparatus, device and medium | |
US11470032B2 (en) | Method for recommending groups and related electronic device | |
CN116775915A (zh) | 资源推荐方法、推荐预测模型训练方法、装置及设备 | |
WO2022143505A1 (zh) | 群组类型识别方法、装置、计算机设备及介质 | |
CN110097004B (zh) | 面部表情识别方法和装置 | |
CN110245684B (zh) | 数据处理方法、电子设备和介质 | |
CN113052246B (zh) | 用于训练分类模型及图像分类的方法和相关装置 | |
CN112989134B (zh) | 节点关系图的处理方法、装置、设备及存储介质 | |
WO2019227633A1 (zh) | 建立用户画像和建立状态信息分析模型的方法和装置 | |
CN111340574B (zh) | 风险用户的识别方法、装置和电子设备 | |
CN111915689A (zh) | 用于生成目标函数的方法、装置、电子设备和计算机可读介质 | |
CN114358102A (zh) | 数据分类方法、装置、设备及存储介质 | |
CN113762306A (zh) | 模型训练方法、装置、电子设备和计算机可读介质 | |
CN112036418A (zh) | 用于提取用户特征的方法和装置 | |
CN111786937B (zh) | 用于识别恶意请求的方法、装置,电子设备和可读介质 | |
CN115131023A (zh) | 一种数据处理方法、装置、服务器以及存储介质 | |
CN117391490A (zh) | 金融业务的评价信息处理方法、装置和计算机设备 | |
CN117807993A (zh) | 分词方法、装置、计算机设备及存储介质 | |
CN117332298A (zh) | 对象识别方法、装置、终端、存储介质及程序产品 |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 21914242 Country of ref document: EP Kind code of ref document: A1 |
|
ENP | Entry into the national phase |
Ref document number: 2023519589 Country of ref document: JP Kind code of ref document: A |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
32PN | Ep: public notification in the ep bulletin as address of the adressee cannot be established |
Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 16.11.2023) |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 21914242 Country of ref document: EP Kind code of ref document: A1 |