Search

CN-117237652-B - Object set detection method, device and storage medium

CN117237652BCN 117237652 BCN117237652 BCN 117237652BCN-117237652-B

Abstract

The application discloses a method and a device for detecting an object set and a storage medium, which can be applied to the field of maps or Internet of vehicles. The method comprises the steps of obtaining relation data of an object set to be processed and determining a relation graph network, then performing migration in nodes of the relation graph network according to preset seed objects to obtain diffusion objects associated with the preset seed objects and connecting the diffusion objects to obtain a sub-graph network, then invoking a graph neural network to perform feature extraction on the sub-graph network to obtain target graph features, and determining a target object set in the object set to be processed according to the target graph network. Therefore, the detection process of the target type object set is realized, and the seed object is adopted to diffuse the node dimension, and the object set is detected by combining the similarity of the network dimension of the graph, so that the accuracy of the object set detection is improved.

Inventors

  • YANG JI
  • CHEN WANGLIN

Assignees

  • 腾讯科技（深圳）有限公司

Dates

Publication Date
20260512
Application Date
20221028

Claims (13)

  1. 1. A method for detecting a set of objects, comprising: Acquiring corresponding relation data among all objects in an object set to be processed, wherein the relation data comprises interaction data and position data, the interaction data comprises any one or combination of a team relation, a game relation, an exchange relation, a click-killing relation and a supporting relation, and the position data comprises any one or combination of a land edge relation and a server relation; traversing based on the relevance between the interaction data and the position data to obtain relationship data; the relation data are associated in a triplet mode, so that a relation chain is obtained; integrating the relationship chains to determine a relationship graph network; Performing migration in nodes of the relational graph network according to a preset seed object to obtain a diffusion object associated with the preset seed object; connecting corresponding nodes of the diffusion objects in the relation graph network to obtain a sub-graph network; Invoking a graph neural network to perform feature extraction on the sub-graph network so as to obtain target graph features; Filtering to obtain a target graph network similar to the sub-graph network based on the target graph characteristics in the relation graph network, and determining a target object set in the object set to be processed according to the target graph network, wherein the target graph network comprises traversing in the relation graph network based on the target graph characteristics to obtain similarity parameters, filtering to obtain the target graph network similar to the sub-graph network according to the similarity parameters, predicting and weighting objects contained in the target graph network and the diffusion objects to obtain a combined suspicious degree parameter, and determining the target object set in the object set to be processed based on the combined suspicious degree parameter, wherein the object type in the target object set is similar to the type of the preset seed object.
  2. 2. The method of claim 1, wherein the wandering in a node of the relational graph network according to a preset seed object to obtain a diffuse object associated with the preset seed object comprises: performing migration in nodes of the relationship graph network according to the preset seed object to determine various migration relationships; calculating the corresponding migration parameters of each migration relation based on a restarting random migration algorithm until the migration parameters are converged, and determining the migration score vector corresponding to the migration relation; Performing correlation calculation according to the wandering score vector to obtain a suspicious degree; and if the suspicious degree reaches a preset threshold value, determining that the node corresponding to the migration relation is a diffusion object associated with the preset seed object.
  3. 3. The method according to claim 2, wherein the method further comprises: Repeatedly wandering in nodes of the relation graph network to obtain object updating parameters; and stopping wandering in the nodes of the relation graph network if the object updating parameters indicate that the diffusion objects are not increased.
  4. 4. The method of claim 1, wherein the invoking the graph neural network to perform feature extraction on the sub-graph network to obtain target graph features comprises: Marking the preset seed object and the diffusion object as black samples, and marking the objects except the preset seed object and the diffusion object in the relation graph network as white samples; invoking a preset neural network configured based on a depth map nested formula, and inputting the black sample and the white sample into the preset neural network to obtain classification loss information; Training the preset neural network by adopting a back propagation algorithm based on the classification loss information to obtain the graph neural network; And inputting the sub-graph network into the graph neural network to perform feature extraction so as to obtain the target graph features.
  5. 5. The method of any of claims 1-4, wherein the method of detecting the set of objects is applied to a blockchain device, the blockchain device being a node in a blockchain.
  6. 6. A device for detecting a set of objects, comprising: The system comprises an acquisition unit, a relation graph network and a relation graph network, wherein the acquisition unit is used for acquiring corresponding relation data among all objects in a to-be-processed object set, the relation data comprises interaction data and position data, the interaction data comprises any one or combination of a team relation, a counter relation, an exchange relation, a click-killing relation and a support relation, and the position data comprises any one or combination of a ground edge relation and a server relation; The migration unit is used for carrying out migration in the nodes of the relational graph network according to a preset seed object so as to obtain a diffusion object associated with the preset seed object; The detection unit is used for connecting the corresponding nodes of the diffusion object in the relation graph network so as to obtain a sub-graph network; the detection unit is also used for calling a graph neural network to extract characteristics of the sub-graph network so as to obtain target graph characteristics; The detection unit is further used for filtering the target graph network based on the target graph characteristics to obtain a target graph network similar to the sub-graph network, determining a target object set in the object set to be processed according to the target graph network, traversing the target graph network based on the target graph characteristics to obtain similarity parameters, filtering the target graph network similar to the sub-graph network according to the similarity parameters, performing predictive weighting according to objects contained in the target graph network and the diffusion objects to obtain a combined suspicious degree parameter, and determining the target object set in the object set to be processed based on the combined suspicious degree parameter, wherein the object type in the target object set is similar to the type of the preset seed object.
  7. 7. The device according to claim 6, characterized in that said wandering unit is specifically configured to: performing migration in nodes of the relationship graph network according to the preset seed object to determine various migration relationships; calculating the corresponding migration parameters of each migration relation based on a restarting random migration algorithm until the migration parameters are converged, and determining the migration score vector corresponding to the migration relation; Performing correlation calculation according to the wandering score vector to obtain a suspicious degree; and if the suspicious degree reaches a preset threshold value, determining that the node corresponding to the migration relation is a diffusion object associated with the preset seed object.
  8. 8. The device according to claim 7, characterized in that said wandering unit is specifically configured to: Repeatedly wandering in nodes of the relation graph network to obtain object updating parameters; and stopping wandering in the nodes of the relation graph network if the object updating parameters indicate that the diffusion objects are not increased.
  9. 9. The device according to claim 6, wherein the detection unit is specifically configured to: Marking the preset seed object and the diffusion object as black samples, and marking the objects except the preset seed object and the diffusion object in the relation graph network as white samples; invoking a preset neural network configured based on a depth map nested formula, and inputting the black sample and the white sample into the preset neural network to obtain classification loss information; Training the preset neural network by adopting a back propagation algorithm based on the classification loss information to obtain the graph neural network; And inputting the sub-graph network into the graph neural network to perform feature extraction so as to obtain the target graph features.
  10. 10. The apparatus according to any of claims 6-9, wherein the method of detecting the set of objects is applied to a blockchain device, the blockchain device being a node in a blockchain.
  11. 11. A computer device, the computer device comprising a processor and a memory: The memory being for storing program code, the processor being for executing the method of detecting a set of objects according to any one of claims 1 to 5 according to instructions in the program code.
  12. 12. A computer program product comprising computer programs/instructions which, when executed by a processor, implement the steps of the method of detection of a set of objects as claimed in any one of the preceding claims 1 to 5.
  13. 13. A computer readable storage medium, characterized in that instructions are stored in the computer readable storage medium, which instructions, when run on a computer, cause the computer to perform the method of detecting a set of objects according to any one of claims 1 to 5.

Description

Object set detection method, device and storage medium Technical Field The present application relates to the field of computer technologies, and in particular, to a method and apparatus for detecting an object set, and a storage medium. Background With the rapid development of internet technology, various network products are presented in the life of people, but there may be non-compliant scenes and objects in the network products, such as black-product group partners, etc. Generally, for black-producing group partners, a current common detection mode is to construct a player relationship network, cut off the relationship connection lower than a threshold value by setting a weight threshold value of the degree among player nodes, cut the network into a plurality of small networks, give the behavior characteristics of players in the small networks, and judge whether the network belongs to the black-producing group partners or not through rules. However, with the continuous change of black-producing groups, the situation that detection is missed may occur when the rule of the fixed rule is judged, and the accuracy of object set detection is affected. Disclosure of Invention In view of this, the present application provides a method for detecting an object set, which can effectively improve the accuracy of object set detection. The first aspect of the present application provides a method for detecting an object set, which may be applied to a system or a program including a function of detecting an object set in a terminal device, and specifically includes: Acquiring relationship data corresponding to an object set to be processed, so as to determine a relationship graph network based on the relationship data; Performing migration in nodes of the relational graph network according to a preset seed object to obtain a diffusion object associated with the preset seed object; connecting corresponding nodes of the diffusion objects in the relation graph network to obtain a sub-graph network; Invoking a graph neural network to perform feature extraction on the sub-graph network so as to obtain target graph features; filtering in the relation graph network based on the target graph characteristics to obtain a target graph network similar to the sub-graph network, and determining a target object set in the object set to be processed according to the target graph network, wherein the object type in the target object set is similar to the type of the preset seed object. Optionally, in some possible implementations of the present application, the obtaining relationship data corresponding to the set of objects to be processed to determine a relationship graph network based on the relationship data includes: acquiring corresponding interaction data and position data among all objects in the object set to be processed; traversing based on the relevance between the interaction data and the position data to obtain the relation data; the relation data are associated in a triplet mode, so that a relation chain is obtained; And integrating the relation chain to determine the relation graph network. Optionally, in some possible implementations of the present application, the performing the walk according to a preset seed object in a node of the relational graph network to obtain a diffusion object associated with the preset seed object includes: performing migration in nodes of the relationship graph network according to the preset seed object to determine various migration relationships; calculating the corresponding migration parameters of each migration relation based on a restarting random migration algorithm until the migration parameters are converged, and determining the migration score vector corresponding to the migration relation; Performing correlation calculation according to the wandering score vector to obtain a suspicious degree; and if the suspicious degree reaches a preset threshold value, determining that the node corresponding to the migration relation is a diffusion object associated with the preset seed object. Optionally, in some possible implementations of the present application, the method further includes: Repeatedly wandering in nodes of the relation graph network to obtain object updating parameters; and stopping wandering in the nodes of the relation graph network if the object updating parameters indicate that the diffusion objects are not increased. Optionally, in some possible implementations of the present application, the invoking graph neural network performs feature extraction on the sub-graph network to obtain target graph features, including: Marking the preset seed object and the diffusion object as black samples, and marking the objects except the preset seed object and the diffusion object in the relation graph network as white samples; invoking a preset neural network configured based on a depth map nested formula, and inputting the black sample and the white sample into the preset neura