Cluster detection methods, apparatus, electronic devices and computer-readable storage media
By constructing a heterogeneous network graph and extracting object features, the problem of low cluster detection accuracy in existing technologies is solved, and higher detection accuracy is achieved.
Patent Information
- Application Number
- CN202210471811.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-04-29
- Publication Date
- 2026-11-10
- Estimated Expiration
- 2042-04-29
AI Technical Summary
Existing cluster detection methods rely solely on topological information through community detection algorithms, resulting in low detection accuracy.
Construct a heterogeneous network graph, identify interaction media and extract object features, and perform clustering to filter out target clusters.
It improves the accuracy of cluster detection by combining topological information and node features, thus avoiding the low accuracy caused by single-relationship identification.
Smart Images

Figure CN117009153B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of communication technology, and more specifically to a cluster detection method, apparatus, electronic device, and computer-readable storage medium. Background Technology
[0002] In recent years, with the rapid development of Internet technology, some groups that disrupt network order have emerged, and these groups have gradually appeared in the network in a cluster pattern. Therefore, it is necessary to detect such clusters in the network. Existing cluster detection methods often use community detection algorithms to cluster suspicious objects, thereby detecting the target cluster.
[0003] In the process of researching and practicing existing technologies, the inventors of this invention discovered that community detection algorithms only target the topological structure information between the objects to be detected. Moreover, the homogeneous network graphs commonly used in community detection result in a single source of information to be detected and a low accuracy rate, thus leading to low detection accuracy in cluster detection. Summary of the Invention
[0004] This invention provides a cluster detection method, apparatus, electronic device, and computer-readable storage medium, which can improve the accuracy of cluster detection.
[0005] A cluster detection method, comprising:
[0006] Obtain the set of objects to be detected and the object interaction information of each object in the set of objects to be detected;
[0007] At least one interaction medium is identified in the object interaction information, and the interaction medium is the medium object of the object to be detected during the interaction process.
[0008] Based on the object interaction information, a heterogeneous network graph is constructed using the object to be detected and the interaction medium as nodes. The heterogeneous network graph is used to indicate the association between the object to be detected and the interaction medium.
[0009] Feature extraction is performed on the heterogeneous network graph to obtain the object features of the object to be detected;
[0010] Based on the object characteristics, the objects to be detected are clustered to obtain at least one candidate cluster, and the target cluster is selected from the candidate clusters.
[0011] Accordingly, embodiments of the present invention provide a cluster detection device, comprising:
[0012] The acquisition unit is used to acquire the set of objects to be detected and the object interaction information of each object to be detected in the set of objects to be detected;
[0013] The identification unit is used to identify at least one interaction medium in the object interaction information, wherein the interaction medium is a medium object of the object to be detected during the interaction process.
[0014] The construction unit is used to construct a heterogeneous network graph based on the object interaction information, using the object to be detected and the interaction medium as nodes. The heterogeneous network graph is used to indicate the association between the object to be detected and the interaction.
[0015] An extraction unit is used to extract features from the heterogeneous network graph to obtain the object features of the object to be detected.
[0016] The filtering unit is used to cluster the objects to be detected based on the object features to obtain at least one candidate cluster, and to filter out the target cluster from the candidate clusters.
[0017] Optionally, in some embodiments, the identification unit may be specifically used to extract the interaction attribute information of the object to be detected from the object interaction information; and to identify at least one interaction medium corresponding to the object to be detected from the interaction attribute information.
[0018] Optionally, in some embodiments, the identification unit may be specifically used to extract the interaction address information and interaction terminal information of the object to be detected from the interaction attribute information; and to determine at least one interaction medium corresponding to the object to be detected based on the interaction address information and interaction terminal information.
[0019] Optionally, in some embodiments, the construction unit may be specifically used to filter out the association information between the object to be detected and the interaction medium from the object interaction information; determine the association relationship between the object to be detected and the interaction medium based on the medium type of the interaction medium and the association information; and construct a heterogeneous network graph with the object to be detected and the interaction medium as nodes according to the association relationship.
[0020] Optionally, in some embodiments, the extraction unit may be specifically used to extract the node features of each node in the heterogeneous network graph; map the node features to a preset feature space to obtain basic node features; and aggregate the basic node features to obtain the object features of the object to be detected.
[0021] Optionally, in some embodiments, the extraction unit may be specifically used to obtain preset mapping feature dimension information and current feature dimension information of each node feature; determine the mapping parameters corresponding to the node feature based on the preset mapping feature dimension information and the current feature dimension information; and map the node feature to the node feature corresponding to the preset mapping feature dimension information according to the mapping parameters to obtain the basic node feature.
[0022] Optionally, in some embodiments, the extraction unit may be specifically used to filter out the node corresponding to the object to be detected from the nodes to obtain the target node, and sample at least one neighbor node of the target node in the heterogeneous network graph; filter out the target node feature corresponding to the target node and the neighbor node feature corresponding to the neighbor node from the basic node features; and fuse the target node feature and the neighbor node feature to obtain the object feature of the object to be detected.
[0023] Optionally, in some embodiments, the extraction unit may be specifically used to identify the number of candidate nodes adjacent to the target node in the heterogeneous network graph; determine the sampling method for the neighboring nodes of the target node based on the number of nodes; and sample at least one neighboring node from the candidate nodes based on the sampling method.
[0024] Optionally, in some embodiments, the extraction unit may be specifically used to identify the node distance between the neighboring node and the target node in the heterogeneous network graph; based on the node distance, the neighboring nodes are layered to obtain a node layer corresponding to each neighboring node, and based on the node layer, a target neighboring node is selected from the neighboring nodes; the neighboring node features corresponding to the target neighboring node are fused with the target node features to obtain the object features of the object to be detected.
[0025] Optionally, in some embodiments, the extraction unit may specifically be used to fuse the neighbor node features corresponding to the target neighbor node with the target node features to obtain the fused node features corresponding to the current node layer; when the current node layer is lower than a preset node layer threshold, the fused node features are used as the target node features, and the step of filtering out the target neighbor node from the neighbor nodes based on the node layer is returned to be executed until the current node layer is not lower than the preset node layer threshold, thereby obtaining the target fused node features of the object to be detected; the target fused node features are normalized to obtain low-dimensional node features, and the low-dimensional node features are used as the object features of the object to be detected.
[0026] Optionally, in some embodiments, the filtering unit may be specifically used to calculate the object similarity between objects to be detected in the set of objects to be detected based on the object features; construct an undirected graph using the objects to be detected as graph nodes according to the object similarity; and cluster the graph nodes in the undirected graph to obtain at least one candidate cluster.
[0027] Optionally, in some embodiments, the filtering unit may be specifically used to assign each graph node in the undirected graph to a corresponding node community and calculate the basic modularity of the node community; identify the adjacent graph nodes corresponding to each graph node in the undirected graph and assign the graph node to the target node community corresponding to the adjacent graph node; calculate the current modularity of the target node community, and based on the basic modularity and the current modularity, filter at least one target node community corresponding to the target node community from the set of objects to be detected, thereby obtaining the candidate cluster.
[0028] Optionally, in some embodiments, the filtering unit may specifically be used to calculate the difference between the basic modularity and the current modularity of the target node community to obtain a modularity difference; based on the modularity difference, update the graph nodes within the target node community to obtain an updated node community, and perform stability testing on the updated node community; when the updated node community is stable, construct an undirected graph using the updated node community as graph nodes to obtain the current undirected graph, and based on the current undirected graph, filter at least one candidate cluster corresponding to an updated node community from the set of candidates to be tested to obtain the candidate cluster.
[0029] Optionally, in some embodiments, the filtering unit may specifically be used to take the current undirected graph as the undirected graph and return to execute the step of dividing each graph node in the undirected graph into the corresponding node community until the current undirected graph does not change, thereby obtaining the target undirected graph; identify the community information of the current node community corresponding to each graph node in the target undirected graph; and based on the community information, filter at least one object to be detected within the current node community in the set of objects to be detected, thereby obtaining the candidate cluster.
[0030] Furthermore, embodiments of the present invention also provide an electronic device, including a processor and a memory, wherein the memory stores an application program, and the processor is used to run the application program in the memory to implement the cluster detection method provided in embodiments of the present invention.
[0031] Furthermore, embodiments of the present invention also provide a computer-readable storage medium storing a plurality of instructions adapted for loading by a processor to execute steps in any of the cluster detection methods provided in embodiments of the present invention.
[0032] In this embodiment of the invention, after acquiring the set of objects to be detected and the object interaction information of each object in the set, at least one interaction medium is identified in the object interaction information. Based on the object interaction information, the objects to be detected and the interaction medium are used as nodes to construct a heterogeneous network graph. Then, feature extraction is performed on the heterogeneous network graph to obtain the object features of the objects to be detected. Based on the object features, the objects to be detected are clustered to obtain at least one candidate cluster, and the target cluster is selected from the candidate clusters. Since this scheme can identify at least one interaction medium in the object interaction information and construct a heterogeneous network graph based on the interaction medium, it can effectively combine the topological structure information in the interaction network with the features of the nodes themselves. In addition, using a heterogeneous network graph can also avoid the low accuracy and coverage caused by identifying the target cluster with a single relationship. Therefore, the accuracy of cluster detection can be improved. Attached Figure Description
[0033] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0034] Figure 1 This is a schematic diagram of a cluster detection method provided in an embodiment of the present invention.
[0035] Figure 2 This is a flowchart illustrating the cluster detection method provided in an embodiment of the present invention;
[0036] Figure 3 This is a flowchart illustrating the extraction of a low-dimensional vector of the object to be detected from a heterogeneous network graph, as provided in an embodiment of the present invention.
[0037] Figure 4 This is a schematic diagram of the process for performing community detection on the target object provided in an embodiment of the present invention;
[0038] Figure 5 This is a schematic diagram of the process for identifying fraud gangs in a set of objects to be detected, provided by an embodiment of the present invention;
[0039] Figure 6 This is another flowchart illustrating the cluster detection method provided in this embodiment of the invention;
[0040] Figure 7 This is a schematic diagram of the cluster detection device provided in an embodiment of the present invention;
[0041] Figure 8 This is a schematic diagram of the structure of the electronic device provided in an embodiment of the present invention. Detailed Implementation
[0042] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0043] This invention provides a cluster detection method, apparatus, electronic device, and computer-readable storage medium. The cluster detection apparatus can be integrated into an electronic device, which may be a server or a terminal, etc.
[0044] The server can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery network (CDN) acceleration services, and big data and artificial intelligence platforms. The terminal includes, but is not limited to, mobile phones, computers, smart voice interaction devices, smart home appliances, vehicle terminals, and aircraft. The terminal and server can be directly or indirectly connected via wired or wireless communication, which is not limited herein. The embodiments of this invention can be applied to various scenarios, including but not limited to cloud technology, artificial intelligence, smart transportation, and assisted driving.
[0045] For example, see Figure 1 Taking the integration of a cluster detection device into an electronic device as an example, after acquiring the set of objects to be detected and the object interaction information of each object in the set of objects to be detected, the electronic device identifies at least one interaction medium in the object interaction information. Based on the object interaction information, the objects to be detected and the interaction medium are used as nodes to construct a heterogeneous network graph. Then, feature extraction is performed on the heterogeneous network graph to obtain the object features of the objects to be detected. Based on the object features, the objects to be detected are clustered to obtain at least one candidate cluster. The target cluster is then selected from the candidate clusters, thereby improving the accuracy of cluster detection.
[0046] The identified target clusters can be stored on a cloud platform. Furthermore, for heterogeneous network graphs, object features of the objects to be detected can be extracted on the cloud platform, and clustering of these objects can also be performed on the cloud platform. A cloud platform, also known as a cloud computing platform, refers to a service that provides computing, networking, and storage capabilities based on hardware and software resources. Cloud computing is a computing model that distributes computing tasks across a resource pool composed of a large number of computers, enabling various application systems to obtain computing power, storage space, and information services as needed. The network providing these resources is called the "cloud." From the user's perspective, resources in the "cloud" are infinitely scalable, readily available, on-demand, expandable, and pay-as-you-go.
[0047] As a provider of fundamental cloud computing capabilities, a cloud resource pool (referred to as a cloud platform, generally called an IaaS (Infrastructure as a Service) platform) is established. Various types of virtual resources are deployed in the resource pool for external customers to choose from. The cloud resource pool mainly includes: computing devices (virtualized machines containing operating systems), storage devices, and network devices.
[0048] Based on logical function, a PaaS (Platform as a Service) layer can be deployed on top of the IaaS (Infrastructure as a Service) layer, and a SaaS (Software as a Service) layer can be deployed on top of the PaaS layer. Alternatively, SaaS can be deployed directly on top of IaaS. PaaS is a platform for running software, such as databases and web containers. SaaS refers to various types of business software, such as web portals and bulk SMS senders. Generally speaking, SaaS and PaaS are upper layers compared to IaaS.
[0049] It is understood that, in the specific embodiments of this application, the object interaction information and other related data involve objects. When the following embodiments of this application are applied to specific products or technologies, permission or consent is required, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.
[0050] The following sections provide detailed descriptions of each example. It should be noted that the order in which the embodiments are described is not intended to limit the preferred order of the embodiments.
[0051] This embodiment will be described from the perspective of a cluster detection device, which can be integrated into an electronic device, such as a server or a terminal. The terminal can include tablet computers, laptops, personal computers (PCs), wearable devices, virtual reality devices, or other smart devices capable of detecting clusters.
[0052] A cluster detection method, comprising:
[0053] The process involves acquiring a set of objects to be detected and object interaction information for each object in the set. At least one interaction medium is identified within this interaction information; this interaction medium is the medium object used by the object during the interaction process. Based on the object interaction information, a heterogeneous network graph is constructed using the objects to be detected and the interaction medium as nodes. This heterogeneous network graph indicates the association between the objects to be detected and the interaction medium. Feature extraction is performed on the heterogeneous network graph to obtain object features of the objects to be detected. Based on these object features, the objects to be detected are clustered to obtain at least one candidate cluster, and the target cluster is selected from the candidate clusters.
[0054] like Figure 2 As shown, the specific process of this cluster detection method is as follows:
[0055] 101. Obtain the set of objects to be detected and the object interaction information of each object to be detected in the set of objects to be detected.
[0056] The object to be detected can be any object that interacts within the interactive network. For example, if the interactive network is a social network, the object to be detected can be any object within the social network.
[0057] Object interaction information can be understood as the interaction information generated by the object to be detected in the interaction network. This interaction information may include the object information of the interactive object, as well as the interaction attribute information of the object to be detected in the interaction network, etc.
[0058] There are several ways to obtain the set of objects to be detected and the object interaction information of each object in the set of objects to be detected, as follows:
[0059] For example, one can directly obtain the set of objects to be detected sent or uploaded by the terminal and the object interaction information of each object to be detected in the set of objects to be detected. Alternatively, one can obtain a set of interaction information for a preset time period in the interaction platform, identify at least one object to be detected in the set of interaction information, obtain the set of objects to be detected, and extract the object interaction information of each object to be detected from the interaction information. Alternatively, one can receive a cluster detection request, which carries the storage address of the interaction information. Based on the storage address, one can obtain the interaction information and extract the set of objects to be detected and the object interaction information of each object to be detected from the interaction information.
[0060] 102. Identify at least one interaction medium in the object interaction information.
[0061] In this context, the interaction medium refers to the medium object used by the object to be detected during the interaction process. The object to be detected can interact with other objects through this interaction medium. There are various types of interaction media, such as interaction terminals, interaction accounts, and interaction addresses.
[0062] There are several ways to identify at least one interaction medium in object interaction information, including the following:
[0063] For example, the interaction attribute information of the object to be detected can be extracted from the object interaction information, and at least one interaction medium corresponding to the object to be detected can be identified from the interaction attribute information.
[0064] Interaction attribute information can be understood as attribute information related to the object being detected during the interaction process. There are several ways to identify at least one interaction medium corresponding to the object being detected from the interaction attribute information. For example, the interaction address information and interaction terminal information of the object being detected can be extracted from the interaction attribute information, and based on the interaction address information and interaction terminal information, at least one interaction medium corresponding to the object being detected can be determined.
[0065] The interaction address information can be the network address of the object being tested during the interaction. This network address can be its own IP address, or the IP address of the object interacting with it. The interaction terminal information can be the information of the terminal used by the object being tested during the interaction, such as a terminal identifier or device identifier. Based on the interaction address information and interaction terminal information, there are multiple ways to determine at least one interaction medium corresponding to the object being tested. For example, the terminal identifier of each interaction terminal used by the object being tested during the interaction with the interaction object can be extracted from the interaction terminal information, and each IP address of the object being tested during the interaction with the interaction object and the corresponding IP address of the interaction object can be extracted from the interaction address information, using the terminal identifier and IP address as the interaction medium.
[0066] 103. Based on the object interaction information, construct a heterogeneous network graph with the objects to be detected and the interaction medium as nodes.
[0067] Among them, heterogeneous network graphs are network graphs in which there are multiple types of nodes, used to indicate the relationship between the object to be detected and the interaction medium.
[0068] There are several ways to construct a heterogeneous network graph by using the objects to be detected and the interaction medium as nodes based on object interaction information. These methods can be as follows:
[0069] For example, the association information between the object to be detected and the interaction medium can be filtered out from the object interaction information. Based on the medium type and association information of the interaction medium, the association relationship between the object to be detected and the interaction medium can be determined. Based on the association relationship, the object to be detected and the interaction medium can be used as nodes to construct a heterogeneous network graph.
[0070] The association information can be information about the relationship between the object to be detected and the interaction medium. For example, taking the interaction medium as a terminal identifier, the association information could include the number of interactions, interaction time, and interaction frequency of the object to be detected using the interaction terminal corresponding to the terminal identifier. Based on the medium type and association information, there are multiple ways to determine the association between the object to be detected and the interaction medium. For instance, the interaction weight between the object to be detected and the interaction medium can be determined based on the association information, and the association between them can be determined according to this interaction weight.
[0071] After determining the relationships, a heterogeneous network graph can be constructed using the objects to be detected and the interaction medium as nodes. There are several ways to construct a heterogeneous network graph. For example, based on the relationship between the objects to be detected and the interaction medium, edge information between the objects to be detected and the interaction medium can be constructed, and a heterogeneous network graph can be constructed based on this edge information.
[0072] 104. Extract features from the heterogeneous network graph to obtain the object features of the object to be detected.
[0073] Among them, object features can be feature information that represents the object to be detected, such as node representation vectors of nodes in a heterogeneous network graph.
[0074] There are several ways to extract features from heterogeneous network graphs, including the following:
[0075] For example, node features can be extracted from a heterogeneous network graph, mapped to a preset feature space to obtain basic node features, and then aggregated to obtain object features of the object to be detected. Specifically, this can be done as follows:
[0076] (1) Extract the node features of each node in the heterogeneous network graph.
[0077] For example, in a heterogeneous network, the node structure information and node attribute information of each node are extracted. The node structure information is then used to extract features to obtain node structure features, and the node attribute information is also used to extract features to obtain node attribute features. Finally, the node structure features and node attribute features are fused together to obtain node features.
[0078] (2) Map the node features to the preset feature space to obtain the basic node features.
[0079] The preset feature space can be a feature space with a fixed feature dimension that has been set in advance.
[0080] There are several ways to map node features to a preset feature space, such as the following:
[0081] For example, preset mapping feature dimension information and current feature dimension information of each node feature can be obtained. Based on the preset mapping feature dimension information and current feature dimension information, the mapping parameters corresponding to the node features can be determined. According to the mapping parameters, the node features are mapped to the node features corresponding to the preset mapping feature dimension information to obtain the basic node features.
[0082] In this context, mapping parameters can be understood as parameters used for feature mapping of node features. Feature mapping can be understood as mapping node features of different types of nodes to the same feature space, thereby aligning the feature dimensions of node features. Mapping parameters can take various forms, such as mapping matrices or other types of parameters. Based on preset mapping feature dimension information and current feature dimension information, there are several ways to determine the mapping parameters corresponding to node features. For example, the mapping feature dimension can be extracted from the preset mapping feature dimension information, and the current feature dimension of the node feature can be extracted from the current feature dimension information. The mapping feature dimension and the current feature dimension can be fused to obtain the fused feature dimension, and the mapping parameters corresponding to the node feature can be determined based on this fused feature dimension.
[0083] After determining the mapping parameters, the node features can be mapped to the node features corresponding to the preset mapping feature dimension information based on the mapping parameters. There are many mapping methods. For example, the mapping parameters can be fused with the node features to obtain the basic node features corresponding to the preset mapping feature dimension information, as shown in formula (1).
[0084]
[0085] in, Let F be the basic node characteristics of node type r. r For the current feature dimension, Let r be the node characteristics. Let r be the mapping parameter for node type r, and F be the mapping feature dimension.
[0086] (3) Aggregate the basic node features to obtain the object features of the object to be detected.
[0087] For example, the node corresponding to the object to be detected can be filtered out from the nodes to obtain the target node, and at least one neighbor node of the target node can be sampled in the heterogeneous network graph. The target node feature and the neighbor node feature corresponding to the target node and the neighbor node feature corresponding to the neighbor node can be filtered out from the basic node features. The target node feature and the neighbor node feature are fused to obtain the object feature of the object to be detected.
[0088] In this context, a neighboring node can be any node adjacent to the target node in a heterogeneous network graph. Adjacency can be direct or indirect. For example, direct adjacency means being directly connected to the target node (within one hop), while indirect adjacency means being indirectly connected (within two or more hops). There are several ways to sample at least one neighboring node of the target node in a heterogeneous network graph. For instance, one can identify the number of candidate nodes adjacent to the target node in the graph, determine the sampling method for neighboring nodes based on this number, and then sample at least one neighboring node from among the candidate nodes.
[0089] There are several ways to determine the sampling method for the neighboring nodes of the target node based on the number of nodes. For example, the number of nodes can be compared with a preset number threshold. When the number of nodes exceeds the preset number threshold, the sampling method can be determined to be sampling without replacement. When the number of nodes does not exceed the preset number threshold, the sampling method can be determined to be sampling with replacement.
[0090] After determining the sampling method for the neighboring nodes of the target node, at least one neighboring node can be sampled from the candidate nodes based on this sampling method. There are multiple sampling methods. For example, when the sampling method is sampling without replacement, the neighboring nodes corresponding to the preset number threshold can be sampled directly from the candidate nodes. When the sampling method is sampling with replacement, the neighboring nodes corresponding to the preset number threshold can be sampled from the candidate nodes using the sampling with replacement method.
[0091] There are several ways to fuse the features of the target node and the features of the neighboring nodes. For example, the distance between the neighboring nodes and the target node can be identified in the heterogeneous network graph. Based on the node distance, the neighboring nodes can be layered to obtain the node layer corresponding to each neighboring node. Based on the node layer, the target neighboring node can be selected from the neighboring nodes. The features of the neighboring nodes corresponding to the target neighboring node can be fused with the features of the target node to obtain the object features of the object to be detected.
[0092] There are several ways to identify the node distance between a neighboring node and a target node in a heterogeneous network graph. For example, one can identify the number of connection layers between each neighboring node and the target node in the heterogeneous network graph and use that number of connection layers as the node distance.
[0093] After identifying the node distance between neighboring nodes and the target node, the neighboring nodes can be layered based on the node distance. There are several ways to layer them. For example, the neighboring nodes can be layered according to the number of connection layers corresponding to each neighboring node, thus obtaining the node layer corresponding to each neighboring node. For example, taking the node layer of the target node as the first layer, the neighboring nodes with one connection layer are classified as the neighboring nodes corresponding to the second layer, and the node layer can be the second layer; the neighboring nodes with two connection layers are classified as the neighboring nodes corresponding to the third layer, and the node layer can be the third layer; the neighboring nodes with (l-1) connection layers are classified as the neighboring nodes corresponding to the l-th layer, and the node layer can be the l-th layer.
[0094] After layering the neighboring nodes, target neighboring nodes can be selected from the neighboring nodes based on the node layer. There are multiple ways to select them. For example, when selecting target neighboring nodes for the first time, the neighboring nodes corresponding to the second layer can be used as target neighboring nodes. After the neighboring nodes corresponding to the second layer have been selected, the neighboring nodes corresponding to the third layer can be used as target neighboring nodes, and so on, until no more target neighboring nodes are selected.
[0095] After selecting the target neighbor nodes, the neighbor node features corresponding to the target neighbor nodes and the target node features can be fused. There are many ways to fuse them. For example, the mean aggregator can be used as the aggregation function to concatenate the neighbor node feature vector corresponding to the target neighbor node (l-1) with the target node feature vector. Then, the average value is calculated for each dimension of the concatenated vector. The result is then subjected to a non-linear transformation to generate the nth layer feature vector of the target node. This feature vector is used as the fused node feature corresponding to the current node layer (the lth layer). Specifically, it can be shown in formula (2):
[0096]
[0097] in, For the fused node features of layer l, W l-1 For nonlinearly varying parameters, For the node features of the (l-1)th layer, The neighbor node features of the (l-1)th layer.
[0098] After fusing the neighbor node features corresponding to the target neighbor node with the target node features, the current node layer is compared with a preset node layer threshold. If the current node layer is lower than the preset node layer threshold, the fused node features are used as the target node features, and the process returns to execute the step of filtering target neighbor nodes based on the node layer until the current node layer is no lower than the preset node layer threshold, thus obtaining the target fused node features of the object to be detected. If the current node layer is not lower than the preset node layer threshold, the fused node features are used as the target fused node features corresponding to the object to be detected.
[0099] The preset node layer threshold can be set according to the actual application, and the preset node layer threshold cannot exceed the maximum number of node layers. The preset node layer threshold can be any positive integer, such as 3 layers or other numbers.
[0100] When the target fused node features of the object to be detected are obtained, the target fused node features can be normalized. There are many ways to normalize them. For example, when the target fused node features are used as feature vectors, the L2 norm of each dimension of the feature vector can be calculated, and the ratio between the feature vector and the L2 norm can be calculated to obtain low-dimensional node features. The low-dimensional features are then used as the object features of the object to be detected. Specifically, it can be shown in formula (3):
[0101]
[0102] Among them, z v The object features of the object to be detected. The target is the node features after fusion. Let L2 norm be the node features after target fusion.
[0103] It should be noted that the low-dimensional vector of the object to be detected extracted from the heterogeneous network graph can be a low-dimensional vector obtained by embedding the node structure information and node attribute information of each node in the heterogeneous network graph together in a low-dimensional manner, as shown in the following example. Figure 3 As shown, the node features are aligned in terms of feature dimensions to obtain basic node features. Node neighbors are sampled in the heterogeneous network graph, and the node features are aggregated to obtain node vectors. When the current node layer of the node vector is not lower than the preset node layer threshold, the node vector is normalized, thereby outputting a low-dimensional vector, which is then used as the object feature of the object to be detected.
[0104] 105. Based on object features, cluster the objects to be detected to obtain at least one candidate cluster, and select the target cluster from the candidate clusters.
[0105] In this context, a candidate cluster can be understood as a group or cluster of objects in the set of objects to be detected that have an inherent relationship.
[0106] There are several ways to cluster objects to be detected based on their features, as follows:
[0107] For example, based on object features, the object similarity between objects in the set of objects to be detected can be calculated. Based on the object similarity, the objects to be detected can be used as graph nodes to construct an undirected graph. The graph nodes in the undirected graph can be clustered to obtain at least one candidate cluster.
[0108] Object similarity can be understood as the degree of similarity between objects in a set of objects to be detected. Based on object features, there are various ways to calculate the object similarity between objects in a set of objects to be detected. For example, one can calculate the feature distance between the object features of the objects in the set of objects to be detected, and determine the object similarity between the objects based on the feature distance, using the Euclidean distance d. ij For example, the object similarity between the objects to be detected can be determined as shown in formula (4):
[0109]
[0110] Among them, S ij Let d be the object similarity between object i and object j to be detected. ij Let σ be the Euclidean distance between object i and object j to be detected, and σ be the similarity parameter.
[0111] After calculating the object similarity, an undirected graph can be constructed using the objects to be detected as graph nodes based on the object similarity. An undirected graph is a network graph with edges that have no direction. There are several ways to construct an undirected graph. For example, the object similarity can be compared with a preset similarity threshold. When the object similarity exceeds the preset similarity threshold, an edge can be constructed between the objects to be detected corresponding to that object similarity. The process of comparing the object similarity with the preset similarity threshold is repeated until every object to be detected in the set of objects to be detected is connected to at least one edge, thus obtaining an undirected graph. Each graph node in the undirected graph corresponds to an object to be detected.
[0112] After constructing the undirected graph, the graph nodes in the undirected graph can be clustered. There are various clustering methods. For example, each graph node in the undirected graph can be assigned to a corresponding node community and the basic modularity of the node community can be calculated. The adjacent graph nodes corresponding to each graph node in the undirected graph can be identified, and the graph node can be assigned to a target node community corresponding to at least the adjacent graph nodes. The current modularity of the target node community can be calculated. Based on the basic modularity and the current modularity, at least one target node community corresponding to the target node community can be selected from the set of objects to be detected to obtain candidate clusters.
[0113] There are several ways to assign each graph node in an undirected graph to its corresponding community. For example, each graph node can be assigned to a different community, i.e., each graph node is treated as a community. After assigning a node to a community, the basic modularity of that community can be calculated. The basic modularity can be understood as the modularity of the community before attempting to assign a node in that community to the community of its neighboring graph nodes. Modularity can also be called a modularity metric, which is often used to measure the strength of the network community structure. There are several ways to calculate the basic modularity. For example, the target weight in the undirected graph can be obtained, the weights inside the community can be extracted from the target weight to obtain the internal weight, the edge weights of the edges connecting the points inside the community can be extracted from the target weight, the ratio between the internal weight and the target weight can be calculated to obtain the internal weight ratio, and the ratio between the edge weight and the target weight can be calculated to obtain the edge weight ratio. The internal weight ratio and the edge weight ratio can be fused to obtain the basic modularity of the community, as shown in formula (5).
[0114]
[0115] Where Q is the basic modularity, m is the target weight, and Σ in For internal weights, Σ tot The edge weight is denoted as .
[0116] The method for calculating the current modularity of the target node community is the same as the method for calculating the basic modularity of the node community. For details, please refer to the above text, and we will not repeat them here.
[0117] After calculating the current modularity of the target node community, at least one target node community can be selected from the set of objects to be detected based on the base modularity and the current modularity, thus obtaining a candidate cluster. There are several ways to select the target node community. For example, the difference between the base modularity and the current modularity of the target node community can be calculated to obtain the modularity difference. Based on the modularity difference, the graph nodes in the target node community can be updated to obtain the updated node community. The stability of the updated node community can then be tested. When the updated node community is stable, the updated node community can be used as the graph nodes to construct an undirected graph. Based on the current undirected graph, at least one target node community corresponding to the updated node community can be selected from the set of objects to be detected, thus obtaining a candidate cluster.
[0118] There are several ways to update graph nodes within a target node community based on the modularity difference. For example, the modularity difference can be compared with a preset difference threshold. When the modularity difference exceeds the preset threshold, the graph node is assigned to the target node community corresponding to an adjacent graph node, resulting in an updated node community. When the modularity difference does not exceed the preset difference threshold, the assignment operation is abandoned, and the target node community is used as the updated node community. Taking a preset difference threshold of 0 as an example, it can be determined whether the modularity difference ΔQ before and after the partition is positive. If it is positive, the partition is accepted; if it is negative, the partition is abandoned, resulting in an updated node community.
[0119] After updating the graph nodes within the target node community, a stability test can be performed on the updated node community. There are several ways to perform a stability test. For example, it can be determined whether all graph nodes have been attempted to be assigned to the target node community and whether the modularity can no longer be increased. When the above conditions are met, it can be determined that the updated node community is stable. Otherwise, the process returns to the step of assigning the graph node to the target node community corresponding to the adjacent graph node, until the updated node community is stable.
[0120] Once the updated node community is stable, there are several ways to construct an undirected graph by treating the updated node community as a graph node. For example, all graph nodes within the updated node community can be merged into one node, that is, each updated node community can be regarded as a graph node. Then, the edges between communities remain unchanged, thereby constructing the current undirected graph.
[0121] After constructing the current undirected graph, at least one node community corresponding to the target object can be selected from the set of target objects based on the current undirected graph. There are multiple ways to select the target object. For example, the current undirected graph can be used as the target undirected graph, and the steps of assigning each graph node in the undirected graph to the corresponding node community can be returned to be executed until the current undirected graph does not change, thus obtaining the target undirected graph. In the target undirected graph, the community information of the current node community corresponding to each graph node can be identified. Based on the community information, at least one target object in the current node community can be selected from the set of target objects to obtain the candidate cluster.
[0122] When the current undirected graph remains unchanged, it indicates that the node communities in the current undirected graph are stable. Therefore, the community information of the current node community corresponding to each graph node can be identified in the obtained target undirected graph. The community information can be the node identifier of the graph nodes contained in the current node community. Based on the node identifier, the objects to be detected contained in the current node community can be filtered out from the set of objects to be detected, and the objects to be detected in the current node community can be regarded as a candidate cluster. Thus, the candidate cluster corresponding to each current node community in the target undirected graph is obtained.
[0123] It's important to clarify that clustering detection objects based on object features can be understood as cluster detection based on the object vectors of the objects to be detected. This primarily leverages the similarity of vectors, using a similarity formula to calculate the similarity between pairs of nodes. First, the Euclidean distance d between the features of each pair of points is calculated. ij Redefining similarity S ij When the similarity is greater than 0.9, an edge is constructed between the two points, thus building an undirected graph. Finally, a community detection algorithm, such as the fast unfolding algorithm, is used for clustering the constructed undirected graph to identify suspicious fraud groups. Specifically, this can be done as follows: Figure 4 As shown,
[0124] S41. Divide each graph node in the undirected graph into different communities, and each graph node is a node community.
[0125] S42. For each graph node, try to assign each node to the community where its neighboring nodes are located, calculate the modularity at this time, and determine whether the difference in modularity before and after the assignment ΔQ is positive. If it is positive, accept the assignment; if it is negative, abandon the assignment.
[0126] S43. Determine whether all graph nodes have been partitioned into node communities and whether the modularity can no longer be increased. If not, proceed to step S42; otherwise, proceed to step S44.
[0127] S44. Merge all graph nodes in each updated node community in the undirected graph into one node, that is, treat each updated node community as a graph node, and keep the edges between communities unchanged, and construct the current undirected graph.
[0128] S45. Determine whether the current undirected graph has changed before and after merging nodes. If it has changed, jump to step S42 and continue partitioning the graph nodes; if the current graph has not changed, jump to step S46.
[0129] S46. The communities in the target undirected graph have stabilized. Output the graph node partitioning results in the target undirected graph to obtain at least one candidate cluster.
[0130] After identifying candidate clusters, target clusters can be selected from them. There are several ways to select target clusters. For example, type detection can be performed on the object features of each object to be detected in the candidate cluster to obtain the object type of each object. When at least one object in the candidate cluster has the object type of the target object type, the candidate cluster can be identified as the target cluster. Alternatively, object type features can be extracted from the object features of each object to be detected in the candidate cluster, and the corresponding object type of the object to be detected can be determined based on these features. When at least one object in the candidate cluster has the object type of the target object type, the candidate cluster can be identified as the target cluster. When the target object type is the fraud object, if at least one object in the candidate cluster is a fraud object, the candidate cluster can be identified as a fraud gang or a criminal gang, and the candidate cluster can be used as the target cluster.
[0131] Optionally, in one embodiment, the cluster detection device further includes storing the target cluster on a blockchain.
[0132] Taking the target cluster as an example of a fraud gang, the process of identifying the fraud gang in the set of objects to be detected in this solution can be as follows: Figure 5 As shown, the specific details are as follows:
[0133] S51. Data preparation.
[0134] For example, information such as the interaction behavior of relevant interactive objects and devices within a certain period of a social network can be collected to construct a set of objects to be detected and the object interaction information of each object in the set of objects to be detected.
[0135] S52. Use heterogeneous graph neural networks to embed nodes to obtain low-dimensional object features.
[0136] For example, a heterogeneous network graph can be constructed using the relationship between the object to be detected and the interaction. The interaction mediums such as the object to be detected and devices / IPs are represented by nodes, and the relationships between the object to be detected and devices / IPs are represented by edges. Then, a heterogeneous graph neural network algorithm is used to perform low-dimensional embedding of the structural and attribute information of each node in the heterogeneous network graph, thereby obtaining low-dimensional object features. For example, the HGraphSAGE algorithm can be used to obtain low-dimensional vectors that effectively represent the structural and attribute information of nodes. Nodes with similar structural and attribute information will also obtain similar vectors.
[0137] S53. Detect fraud gangs based on object characteristics.
[0138] For example, the similarity of vectors can be utilized to calculate the similarity between pairs of nodes using a similarity formula. When the similarity is greater than 0.9, an edge is constructed between the two points, thus building an undirected graph. The fast unfolding algorithm is then used to cluster the constructed undirected graph, ultimately identifying suspected fraud groups. Fraud groups are then selected from among these suspected fraud groups.
[0139] In this scheme, a graph neural network algorithm is used to combine network structure information and node attribute information to perform low-dimensional node representation of the object to be detected, rather than directly using node features as representation vectors. This provides additional information gain, removes redundant information, and makes the representation vector of the object to be detected more accurate, thereby improving the accuracy of the object features. Furthermore, using a heterogeneous graph algorithm allows for the simultaneous utilization of multiple structural relationships, making the connections between the objects to be detected closer, thus improving the accuracy and coverage of target cluster detection.
[0140] As can be seen from the above, in this embodiment of the application, after obtaining the set of objects to be detected and the object interaction information of each object to be detected in the set of objects to be detected, at least one interaction medium is identified in the object interaction information. Based on the object interaction information, the objects to be detected and the interaction medium are used as nodes to construct a heterogeneous network graph. Then, feature extraction is performed on the heterogeneous network graph to obtain the object features of the objects to be detected. Based on the object features, the objects to be detected are clustered to obtain at least one candidate cluster, and the target cluster is selected from the candidate clusters. Since this scheme can identify at least one interaction medium in the object interaction information and construct a heterogeneous network graph based on the interaction medium, it can effectively combine the topological structure information in the interaction network with the features of the nodes themselves. In addition, the use of a heterogeneous network graph can also avoid the low accuracy and coverage caused by identifying the target cluster with a single relationship. Therefore, the accuracy of cluster detection can be improved.
[0141] Based on the method described in the above embodiments, the following examples will provide further detailed explanations.
[0142] In this embodiment, the cluster detection device will be specifically integrated into an electronic device, with the electronic device serving as a server and the target cluster representing a fraud gang, as an example for illustration.
[0143] like Figure 6 As shown, a cluster detection method has the following specific process:
[0144] 201. The server obtains the set of objects to be detected and the object interaction information of each object to be detected in the set of objects to be detected.
[0145] For example, the server can directly obtain the set of objects to be detected sent or uploaded by the terminal and the object interaction information of each object to be detected in the set of objects to be detected. Alternatively, it can obtain the set of interaction information for a preset time period in the interaction platform, identify at least one object to be detected in the set of interaction information, obtain the set of objects to be detected, and extract the object interaction information of each object to be detected from the interaction information. Alternatively, it can also receive a cluster detection request, which carries the storage address of the interaction information. Based on the storage address, it can obtain the interaction information and extract the set of objects to be detected and the object interaction information of each object to be detected from the interaction information.
[0146] 202. The server identifies at least one interaction medium in the object interaction information.
[0147] For example, the server can extract the interaction attribute information of the object to be detected from the object interaction information, extract the interaction address information and interaction terminal information of the object to be detected from the interaction attribute information, extract the terminal identifier of each interaction terminal used by the object to be detected when interacting with the interaction object from the interaction terminal information, extract each IP address of the object to be detected and the corresponding IP address of the interaction object when interacting with the interaction object from the interaction address information, and so on, using the terminal identifier and IP address as the interaction medium.
[0148] 203. Based on the object interaction information, the server constructs a heterogeneous network graph using the objects to be detected and the interaction medium as nodes.
[0149] For example, the server can filter the association information between the object to be detected and the interaction medium from the object interaction information. Based on the association information, it can determine the interaction weight between the object to be detected and the interaction medium, and determine the association relationship between the object to be detected and the interaction medium according to the interaction weight. Based on the relationship between the object to be detected and the interaction medium, it can construct the edge information between the object to be detected and the interaction medium, and construct a heterogeneous network graph based on the edge information.
[0150] 204. The server extracts the node features of each node in the heterogeneous network graph.
[0151] For example, the server extracts the node structure information and node attribute information of each node in a heterogeneous network, performs feature extraction on the node structure information to obtain node structure features, and performs feature extraction on the node attribute information to obtain node attribute features. The node structure features and node attribute features are then fused to obtain node features.
[0152] 205. The server maps the node features to a preset feature space to obtain the basic node features.
[0153] For example, the server can obtain preset mapping feature dimension information and the current feature dimension information of each node feature. It can extract the mapping feature dimension from the preset mapping feature dimension information and the current feature dimension of the node feature from the current feature dimension information. The server can then fuse the mapping feature dimension and the current feature dimension to obtain the fused feature dimension. Based on this fused feature dimension, the server can determine the mapping parameters corresponding to the node feature. By fusing the mapping parameters with the node feature, the server can obtain the basic node feature corresponding to the preset mapping feature dimension information, as shown in formula (1).
[0154] 206. The server aggregates the basic node features to obtain the object features of the object to be detected.
[0155] For example, the server can filter out the nodes corresponding to the object to be detected from the nodes to obtain the target node. In the heterogeneous network graph, it can identify the number of candidate nodes adjacent to the target node and compare the number of nodes with a preset number threshold. When the number of nodes exceeds the preset number threshold, the sampling method can be determined to be sampling without replacement. When the number of nodes does not exceed the preset number threshold, the sampling method can be determined to be sampling with replacement. When the sampling method is sampling without replacement, the neighboring nodes corresponding to the preset number threshold can be directly sampled from the candidate nodes. When the sampling method is sampling with replacement, the neighboring nodes corresponding to the preset number threshold can be sampled from the candidate nodes using the sampling with replacement method.
[0156] The server filters the target node features and neighbor node features from the basic node features to identify the target node and its neighbors. In the heterogeneous network graph, the number of connection layers between each neighbor and the target node is identified and used as the node distance. Neighbors are then stratified based on their connection layer number to obtain the node layer for each neighbor. For example, taking the target node's node layer as the first layer, neighbors with one connection layer are classified as second-layer neighbors; neighbors with two connection layers are classified as third-layer neighbors; and neighbors with (l-1) connection layers are classified as l-layer neighbors. When initially filtering target neighbors, the second-layer neighbors can be selected. Once the second-layer neighbors are selected, the third-layer neighbors can be selected, and so on, until no further target neighbors are selected. The mean aggregator is used as the aggregation function to concatenate the neighbor feature vectors corresponding to the target neighbor node (l-1) with the target node feature vector. Then, the average value is calculated for each dimension of the concatenated vector. The result is then subjected to a non-linear transformation to generate the nth layer feature vector of the target node. This feature vector is used as the fused node feature corresponding to the current node layer (lth layer), as shown in formula (2). The current node layer is compared with the preset node layer threshold. When the current node layer is lower than the preset node layer threshold, the fused node feature is used as the target node feature, and the process of selecting the target neighbor node based on the node layer is returned until the current node layer is no lower than the preset node layer threshold, thus obtaining the target fused node feature of the object to be detected. When the current node layer is no lower than the preset node layer threshold, the fused node feature is used as the target fused node feature corresponding to the object to be detected. When the node features after target fusion are used as feature vectors, the L2 norm of each dimension of the feature vector can be calculated, and the ratio between the feature vector and the L2 norm can be calculated to obtain low-dimensional node features. The low-dimensional features are used as object features of the object to be detected, as shown in formula (3).
[0157] 207. The server clusters the objects to be detected based on object features to obtain at least one candidate cluster.
[0158] For example, the server can calculate the feature distance between object features of objects in the set of objects to be detected, and determine the object similarity between the objects based on the feature distance, using the feature distance as the Euclidean distance d. ij For example, the similarity between objects to be detected can be determined as shown in formula (4).
[0159] The server compares the object similarity with a preset similarity threshold. When the object similarity exceeds the preset similarity threshold, an edge is constructed between the objects to be detected corresponding to that object similarity. The server then returns to the step of comparing the object similarity with the preset similarity threshold until each object to be detected in the set of objects to be detected is connected to at least one edge, thus obtaining an undirected graph. Each node in the undirected graph corresponds to an object to be detected.
[0160] The server can divide each graph node in the undirected graph into different communities, that is, each graph node is a node community. Obtain the target weight in the undirected graph, extract the weight inside the node community from the target weight to obtain the internal weight, extract the edge weight of the edge connecting the node to the node inside the community from the target weight, calculate the ratio between the internal weight and the target weight to obtain the internal weight ratio, and calculate the ratio between the edge weight and the target weight to obtain the edge weight ratio. Combine the internal weight ratio and the edge weight ratio to obtain the basic modularity of the node community, as shown in formula (5).
[0161] The server identifies the neighboring nodes of each graph node in the undirected graph and assigns the graph node to a target node community corresponding to at least its neighboring graph nodes, calculating the current modularity of the target node community. It calculates the difference between the base modularity and the current modularity of the target node community, obtaining the modularity difference. If the modularity difference is positive, it accepts the partitioning operation to assign the graph node to the target node community corresponding to its neighboring graph nodes, resulting in an updated node community. If the modularity difference is negative, it abandons the partitioning operation to assign the graph node to the target node community corresponding to its neighboring graph nodes, and uses the target node community as the updated node community. It determines whether all graph nodes have been attempted to be assigned to target node communities and whether the modularity can no longer be increased. If these conditions are met, the updated node community is considered stable; otherwise, it returns to the step of assigning the graph node to the target node community corresponding to its neighboring graph nodes, continuing until the updated node community is stable.
[0162] The server merges all graph nodes within the updated node community into a single node, treating each updated node community as a single graph node. The edges between communities remain unchanged, thus constructing the current undirected graph. This current undirected graph is then used as the target undirected graph, and the process of assigning each graph node in the target undirected graph to its corresponding node community is repeated until the current undirected graph remains unchanged. The target undirected graph is then obtained by identifying the community information of each graph node's current node community. Node identifiers containing graph nodes are then filtered from this community information. Based on these node identifiers, the objects to be detected within the current node community are selected from the set of objects to be detected. These objects within the current node community are then considered as a candidate cluster, thus obtaining the candidate clusters corresponding to each current node community in the target undirected graph.
[0163] 208. The server filters out fraud gangs from the candidate clusters.
[0164] For example, the server performs type detection on the object features of each object to be detected in the candidate cluster to obtain the object type of each object to be detected. When there is at least one object to be detected in the candidate cluster whose object type is fraudulent, the candidate cluster can be identified as a fraud gang. Alternatively, the object type features can be extracted from the object features of each object to be detected in the candidate cluster, and the object type of the corresponding object to be detected can be determined based on the object type features. When there is at least one object to be detected in the candidate cluster whose object type is fraudulent, the candidate cluster can be identified as a fraud gang.
[0165] As can be seen from the above, in this embodiment, after the server obtains the set of objects to be detected and the object interaction information of each object to be detected in the set of objects to be detected, it identifies at least one interaction medium in the object interaction information. Based on the object interaction information, it constructs a heterogeneous network graph with the objects to be detected and the interaction medium as nodes. Then, it extracts features from the heterogeneous network graph to obtain the object features of the objects to be detected. Based on the object features, it clusters the objects to be detected to obtain at least one candidate cluster, and then filters out fraud gangs from the candidate clusters. Since this scheme can identify at least one interaction medium in the object interaction information and construct a heterogeneous network graph based on the interaction medium, it can effectively combine the topological structure information in the interaction network with the features of the nodes themselves. In addition, using a heterogeneous network graph can also avoid the low accuracy and coverage caused by identifying target clusters with a single relationship. Therefore, it can improve the accuracy of cluster detection.
[0166] To better implement the above methods, embodiments of the present invention also provide a cluster detection device, which can be integrated into electronic devices, such as servers or terminals, and the terminals can include tablet computers, laptop computers and / or personal computers.
[0167] For example, such as Figure 7 As shown, the cluster detection device may include an acquisition unit 301, an identification unit 302, a construction unit 303, an extraction unit 304, and a filtering unit 305, as follows:
[0168] (1) Obtain unit 301;
[0169] The acquisition unit 301 is used to acquire the set of objects to be detected and the object interaction information of each object to be detected in the set of objects to be detected.
[0170] For example, the acquisition unit 301 can be specifically used to acquire the set of objects to be detected sent or uploaded by the terminal and the object interaction information of each object to be detected in the set of objects to be detected. Alternatively, it can acquire a set of interaction information for a preset time period in the interaction platform, identify at least one object to be detected in the set of interaction information, obtain the set of objects to be detected, and extract the object interaction information of each object to be detected from the interaction information. Alternatively, it can also receive a cluster detection request, which carries the storage address of the interaction information, acquire the interaction information based on the storage address, and extract the set of objects to be detected and the object interaction information of each object to be detected from the interaction information.
[0171] (2) Identification unit 302;
[0172] The identification unit 302 is used to identify at least one interaction medium in the object interaction information, wherein the interaction medium is the medium object of the object to be detected during the interaction process.
[0173] For example, the identification unit 302 can be used to extract the interaction attribute information of the object to be detected from the object interaction information, extract the interaction address information and interaction terminal information of the object to be detected from the interaction attribute information, and determine at least one interaction medium corresponding to the object to be detected based on the interaction address information and interaction terminal information.
[0174] (3) Constructing unit 303;
[0175] The construction unit 303 is used to construct a heterogeneous network graph based on the object interaction information, using the object to be detected and the interaction medium as nodes. The heterogeneous network graph is used to indicate the association between the object to be detected and the interaction.
[0176] For example, the construction unit 303 can be used to filter out the association information between the object to be detected and the interaction medium in the object interaction information, determine the association relationship between the object to be detected and the interaction medium based on the medium type and association information of the interaction medium, and construct a heterogeneous network graph with the object to be detected and the interaction medium as nodes according to the association relationship.
[0177] (4) Extraction unit 304;
[0178] Extraction unit 304 is used to extract features from heterogeneous network graphs to obtain object features of the object to be detected.
[0179] For example, extraction unit 304 can be used to extract the node features of each node in a heterogeneous network graph, map the node features to a preset feature space to obtain basic node features, filter out the node corresponding to the object to be detected from the nodes to obtain the target node, sample at least one neighbor node of the target node in the heterogeneous network graph, filter out the target node features corresponding to the target node and the neighbor node features corresponding to the neighbor node from the basic node features, and fuse the target node features and the neighbor node features to obtain the object features of the object to be detected.
[0180] (5) Filtering unit 305;
[0181] The filtering unit 305 is used to cluster the objects to be detected based on object features to obtain at least one candidate cluster, and to filter out the target cluster from the candidate clusters.
[0182] For example, the filtering unit 305 can be used to calculate the object similarity between objects in the set of objects to be detected based on object features. Based on the object similarity, the objects to be detected are used as graph nodes to construct an undirected graph. The graph nodes in the undirected graph are then clustered to obtain at least one candidate cluster. The target cluster is then selected from the candidate clusters.
[0183] In practice, each of the above units can be implemented as an independent entity or can be arbitrarily combined to be implemented as the same or several entities. For the specific implementation of each of the above units, please refer to the previous method embodiments, which will not be repeated here.
[0184] As can be seen from the above, in this embodiment, after the acquisition unit 301 acquires the set of objects to be detected and the object interaction information of each object to be detected in the set of objects to be detected, the identification unit 302 identifies at least one interaction medium in the object interaction information. The construction unit 303 constructs a heterogeneous network graph with the objects to be detected and the interaction medium as nodes based on the object interaction information. Then, the extraction unit 304 extracts features from the heterogeneous network graph to obtain the object features of the objects to be detected. The filtering unit 305 clusters the objects to be detected based on the object features to obtain at least one candidate cluster and filters out the target cluster from the candidate clusters. Since this scheme can identify at least one interaction medium in the object interaction information and construct a heterogeneous network graph based on the interaction medium, it can effectively combine the topological structure information in the interaction network with the features of the nodes themselves. In addition, the use of a heterogeneous network graph can also avoid the low accuracy and coverage caused by identifying the target cluster with a single relationship. Therefore, the accuracy of cluster detection can be improved.
[0185] This invention also provides an electronic device, such as... Figure 8 As shown, it illustrates a structural schematic diagram of the electronic device involved in an embodiment of the present invention, specifically:
[0186] The electronic device may include components such as a processor 401 with one or more processing cores, a memory 402 with one or more computer-readable storage media, a power supply 403, and an input unit 404. Those skilled in the art will understand that... Figure 8 The electronic device structure shown does not constitute a limitation on the electronic device and may include more or fewer components than shown, or combine certain components, or have different component arrangements. Wherein:
[0187] The processor 401 is the control center of the electronic device, connecting various parts of the device via various interfaces and lines. It executes software programs and / or modules stored in the memory 402, and calls data stored in the memory 402, to perform various functions and process data. Optionally, the processor 401 may include one or more processing cores; preferably, the processor 401 may integrate an application processor and a modem processor, wherein the application processor mainly handles the operating system, user interface, and applications, and the modem processor mainly handles wireless communication. It is understood that the modem processor may not be integrated into the processor 401.
[0188] The memory 402 can be used to store software programs and modules. The processor 401 executes various functional applications and data processing by running the software programs and modules stored in the memory 402. The memory 402 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, application programs required for at least one function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the electronic device, etc. In addition, the memory 402 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device. Accordingly, the memory 402 may also include a memory controller to provide the processor 401 with access to the memory 402.
[0189] The electronic device also includes a power supply 403 that supplies power to the various components. Preferably, the power supply 403 can be logically connected to the processor 401 through a power management system, thereby enabling functions such as charging, discharging, and power consumption management through the power management system. The power supply 403 may also include one or more DC or AC power supplies, recharging systems, power fault detection circuits, power converters or inverters, power status indicators, and other arbitrary components.
[0190] The electronic device may also include an input unit 404, which can be used to receive input digital or character information, and generate keyboard, mouse, joystick, optical or trackball signal inputs related to user settings and function control.
[0191] Although not shown, the electronic device may also include a display unit, etc., which will not be described in detail here. Specifically, in this embodiment, the processor 401 in the electronic device loads the executable files corresponding to the processes of one or more applications into the memory 402 according to the following instructions, and the processor 401 runs the applications stored in the memory 402 to realize various functions, as follows:
[0192] The process involves acquiring a set of objects to be detected and object interaction information for each object in the set. At least one interaction medium is identified within this interaction information; this interaction medium is the medium object used by the object during the interaction process. Based on the object interaction information, a heterogeneous network graph is constructed using the objects to be detected and the interaction medium as nodes. This heterogeneous network graph indicates the association between the objects to be detected and the interaction medium. Feature extraction is performed on the heterogeneous network graph to obtain object features of the objects to be detected. Based on these object features, the objects to be detected are clustered to obtain at least one candidate cluster, and the target cluster is selected from the candidate clusters.
[0193] For example, an electronic device can acquire a set of objects to be detected sent or uploaded by a terminal, along with the object interaction information of each object in that set. Alternatively, it can acquire a set of interaction information over a preset time period from an interaction platform, identify at least one object to be detected within that set, obtain a set of objects to be detected, and extract the object interaction information of each object from the interaction information. Alternatively, it can receive a cluster detection request carrying the storage address of the interaction information, acquire the interaction information based on that address, and extract the set of objects to be detected and the object interaction information of each object from the interaction information. From the object interaction information, it can extract the interaction attribute information of the objects to be detected, and from the interaction attribute information, it can extract the interaction address information and interaction terminal information of the objects to be detected. Based on the interaction address information and interaction terminal information, it can determine at least one interaction medium corresponding to the object to be detected. It can then filter out the association information between the objects to be detected and the interaction medium from the object interaction information, determine the association relationship between the objects to be detected and the interaction medium based on the medium type and association information, and construct a heterogeneous network graph using the objects to be detected and the interaction medium as nodes according to the association relationship. The node features of each node in the heterogeneous network graph are extracted and mapped to a preset feature space to obtain basic node features. Nodes corresponding to the objects to be detected are selected from these basic node features to obtain target nodes. At least one neighboring node of the target node is sampled from the heterogeneous network graph. Target node features and neighboring node features corresponding to the target node and the neighboring node features are then selected from the basic node features and fused to obtain the object features of the objects to be detected. Based on these object features, the object similarity between objects in the set of objects to be detected is calculated. Based on the object similarity, an undirected graph is constructed using the objects to be detected as graph nodes. The graph nodes in the undirected graph are clustered to obtain at least one candidate cluster. The target cluster is then selected from the candidate clusters.
[0194] For details on the implementation of each of the above operations, please refer to the previous examples, which will not be repeated here.
[0195] As can be seen from the above, after obtaining the set of objects to be detected and the object interaction information of each object in the set of objects to be detected, this embodiment of the invention identifies at least one interaction medium in the object interaction information. Based on the object interaction information, the objects to be detected and the interaction medium are used as nodes to construct a heterogeneous network graph. Then, feature extraction is performed on the heterogeneous network graph to obtain the object features of the objects to be detected. Based on the object features, the objects to be detected are clustered to obtain at least one candidate cluster, and the target cluster is selected from the candidate clusters. Since this scheme can identify at least one interaction medium in the object interaction information and construct a heterogeneous network graph based on the interaction medium, it can effectively combine the topological structure information in the interaction network with the features of the nodes themselves. In addition, using a heterogeneous network graph can also avoid the low accuracy and coverage caused by identifying the target cluster with a single relationship. Therefore, the accuracy of cluster detection can be improved.
[0196] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be performed by instructions, or by instructions controlling related hardware. These instructions can be stored in a computer-readable storage medium and loaded and executed by a processor.
[0197] To this end, embodiments of the present invention provide a computer-readable storage medium storing a plurality of instructions that can be loaded by a processor to execute steps in any of the cluster detection methods provided in the embodiments of the present invention. For example, the instructions can execute the following steps:
[0198] The process involves acquiring a set of objects to be detected and object interaction information for each object in the set. At least one interaction medium is identified within this interaction information; this interaction medium is the medium object used by the object during the interaction process. Based on the object interaction information, a heterogeneous network graph is constructed using the objects to be detected and the interaction medium as nodes. This heterogeneous network graph indicates the association between the objects to be detected and the interaction medium. Feature extraction is performed on the heterogeneous network graph to obtain object features of the objects to be detected. Based on these object features, the objects to be detected are clustered to obtain at least one candidate cluster, and the target cluster is selected from the candidate clusters.
[0199] For example, the system can acquire a set of objects to be detected sent or uploaded by the terminal, along with the object interaction information of each object in the set. Alternatively, it can acquire a set of interaction information over a preset time period from the interaction platform, identify at least one object to be detected within this set, obtain the set of objects to be detected, and extract the object interaction information of each object from the interaction information. Another approach is to receive a cluster detection request carrying the storage address of the interaction information, acquire the interaction information based on this address, and extract the set of objects to be detected and the object interaction information of each object from the interaction information. The system then extracts the interaction attribute information of the objects to be detected from the object interaction information, and further extracts the interaction address information and interaction terminal information of the objects to be detected from the interaction attribute information. Based on the interaction address information and interaction terminal information, it determines at least one interaction medium corresponding to the object to be detected. Finally, it filters the association information between the objects to be detected and the interaction medium from the object interaction information, determines the association relationship between the objects to be detected and the interaction medium based on the medium type and association information, and constructs a heterogeneous network graph using the objects to be detected and the interaction medium as nodes according to the association relationship. The node features of each node in the heterogeneous network graph are extracted and mapped to a preset feature space to obtain basic node features. Nodes corresponding to the objects to be detected are selected from these basic node features to obtain target nodes. At least one neighboring node of the target node is sampled from the heterogeneous network graph. Target node features and neighboring node features corresponding to the target node and the neighboring node features are then selected from the basic node features and fused to obtain the object features of the objects to be detected. Based on these object features, the object similarity between objects in the set of objects to be detected is calculated. Based on the object similarity, an undirected graph is constructed using the objects to be detected as graph nodes. The graph nodes in the undirected graph are clustered to obtain at least one candidate cluster. The target cluster is then selected from the candidate clusters.
[0200] For details on the implementation of each of the above operations, please refer to the previous examples, which will not be repeated here.
[0201] The computer-readable storage medium may include: read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.
[0202] Since the instructions stored in the computer-readable storage medium can execute the steps in any of the cluster detection methods provided in the embodiments of the present invention, the beneficial effects that any of the cluster detection methods provided in the embodiments of the present invention can achieve can be realized, as detailed in the preceding embodiments, and will not be repeated here.
[0203] According to one aspect of this application, a computer program product or computer program is provided, comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the methods provided in the various alternative implementations of the above-described cluster detection or fraud gang detection aspects.
[0204] The above provides a detailed description of a cluster detection method, apparatus, electronic device, and computer-readable storage medium provided by embodiments of the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, those skilled in the art will recognize that there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.
Claims
1. A cluster detection method, characterized in that, include: Obtain the set of objects to be detected and the object interaction information of each object in the set of objects to be detected; At least one interaction medium is identified in the object interaction information, and the interaction medium is the medium object of the object to be detected during the interaction process. Based on the object interaction information, a heterogeneous network graph is constructed using the object to be detected and the interaction medium as nodes. The heterogeneous network graph is used to indicate the association between the object to be detected and the interaction medium. Extract the node features of each node from the heterogeneous network graph. The node features are mapped to a preset feature space to obtain basic node features; The node corresponding to the object to be detected is selected from the nodes to obtain the target node, and at least one neighbor node of the target node is sampled in the heterogeneous network graph. Filter out the target node features corresponding to the target node and the neighbor node features corresponding to the neighbor nodes from the basic node features; The target node features and neighboring node features are fused to obtain the object features of the object to be detected; Based on the object characteristics, the objects to be detected are clustered to obtain at least one candidate cluster, and the target cluster is selected from the candidate clusters.
2. The cluster detection method according to claim 1, characterized in that, The process of identifying at least one interaction medium in the object interaction information includes: Extract the interaction attribute information of the object to be detected from the object interaction information; Identify at least one interactive medium corresponding to the object to be detected from the interactive attribute information.
3. The cluster detection method according to claim 2, characterized in that, The step of identifying at least one interactive medium corresponding to the object to be detected from the interactive attribute information includes: The interaction address information and interaction terminal information of the object to be detected are extracted from the interaction attribute information; Based on the interaction address information and interaction terminal information, at least one interaction medium corresponding to the object to be detected is determined.
4. The cluster detection method according to claim 2, characterized in that, The step of constructing a heterogeneous network graph by using the object to be detected and the interaction medium as nodes based on the object interaction information includes: The association information between the object to be detected and the interaction medium is filtered out from the object interaction information; Based on the medium type of the interaction medium and the association information, the association relationship between the object to be detected and the interaction medium is determined; Based on the aforementioned relationships, a heterogeneous network graph is constructed using the objects to be detected and the interaction medium as nodes.
5. The cluster detection method according to claim 1, characterized in that, The step of mapping the node features to a preset feature space to obtain basic node features includes: Obtain the preset mapping feature dimension information and the current feature dimension information of each node feature; Based on the preset mapping feature dimension information and the current feature dimension information, determine the mapping parameters corresponding to the node features; Based on the mapping parameters, the node features are mapped to the node features corresponding to the preset mapping feature dimension information to obtain the basic node features.
6. The cluster detection method according to claim 1, characterized in that, The step of sampling at least one neighboring node of the target node in the heterogeneous network graph includes: The number of candidate nodes adjacent to the target node is identified in the heterogeneous network graph; Based on the number of nodes, determine the sampling method for the neighboring nodes of the target node; Based on the sampling method, at least one neighboring node is sampled from the candidate nodes.
7. The cluster detection method according to claim 1, characterized in that, The process of fusing the target node features and neighboring node features to obtain the object features of the object to be detected includes: The node distance between the neighboring node and the target node is identified in the heterogeneous network graph; Based on the node distance, the neighboring nodes are layered to obtain a node layer corresponding to each neighboring node, and based on the node layer, target neighboring nodes are selected from the neighboring nodes. The features of the neighboring nodes corresponding to the target neighboring nodes are fused with the features of the target node to obtain the object features of the object to be detected.
8. The cluster detection method according to claim 7, characterized in that, The step of fusing the neighbor node features corresponding to the target neighbor node with the target node features to obtain the object features of the object to be detected includes: The neighbor node features corresponding to the target neighbor node are fused with the target node features to obtain the fused node features corresponding to the current node layer; When the current node layer is lower than the preset node layer threshold, the fused node features are used as the target node features, and the step of filtering out the target neighbor node from the neighbor nodes based on the node layer is returned to be executed until the current node layer is not lower than the preset node layer threshold, so as to obtain the target fused node features of the object to be detected. The fused node features of the target are normalized to obtain low-dimensional node features, which are then used as the object features of the object to be detected.
9. The cluster detection method according to any one of claims 1 to 4, characterized in that, The step of clustering the objects to be detected based on the object features to obtain at least one candidate cluster includes: Based on the object features, calculate the object similarity between the objects to be detected in the set of objects to be detected; Based on the object similarity, the objects to be detected are used as graph nodes to construct an undirected graph; Cluster the graph nodes in the undirected graph to obtain at least one candidate cluster.
10. The cluster detection method according to claim 9, characterized in that, The step of clustering the graph nodes in the undirected graph to obtain at least one candidate cluster includes: Each graph node in the undirected graph is assigned to a corresponding node community, and the basic modularity of the node community is calculated. In the undirected graph, the adjacent graph nodes corresponding to each graph node are identified, and the graph node is assigned to the target node community corresponding to the adjacent graph node. Calculate the current modularity of the target node community, and based on the basic modularity and the current modularity, select at least one target node community corresponding to the target node community from the set of objects to be detected, thereby obtaining the candidate cluster.
11. The cluster detection method according to claim 10, characterized in that, The process of selecting at least one target node community from the set of objects to be detected based on the basic modularity and the current modularity to obtain the candidate cluster includes: Calculate the difference between the base modularity and the current modularity of the target node community to obtain the modularity difference; Based on the modularity difference, the graph nodes within the target node community are updated to obtain the updated node community, and the stability of the updated node community is then tested. When the updated node community is stable, the updated node community is used as a graph node to construct an undirected graph to obtain the current undirected graph. Based on the current undirected graph, at least one object corresponding to the updated node community is selected from the set of objects to be detected to obtain the candidate cluster.
12. The cluster detection method according to claim 11, characterized in that, The step of selecting at least one candidate cluster corresponding to an updated node community from the set of candidate objects based on the current undirected graph to obtain the candidate cluster includes: The current undirected graph is used as the undirected graph, and the step of assigning each graph node in the undirected graph to the corresponding node community is returned to be executed until the current undirected graph does not change, thus obtaining the target undirected graph; In the target undirected graph, the community information of the current node community corresponding to each graph node is identified; Based on the community information, at least one target object within the current node's community is selected from the set of target objects to be detected, thus obtaining the candidate cluster.
13. A cluster detection method, characterized in that, include: The acquisition unit is used to acquire the set of objects to be detected and the object interaction information of each object to be detected in the set of objects to be detected; The identification unit is used to identify at least one interaction medium in the object interaction information, wherein the interaction medium is a medium object of the object to be detected during the interaction process. The construction unit is used to construct a heterogeneous network graph based on the object interaction information, using the object to be detected and the interaction medium as nodes. The heterogeneous network graph is used to indicate the association between the object to be detected and the interaction. An extraction unit is configured to extract node features of each node in the heterogeneous network graph, map the node features to a preset feature space to obtain basic node features, filter out the node corresponding to the object to be detected from the nodes to obtain the target node, sample at least one neighbor node of the target node in the heterogeneous network graph, filter out the target node features corresponding to the target node and the neighbor node features corresponding to the neighbor node from the basic node features, and fuse the target node features and the neighbor node features to obtain the object features of the object to be detected. The filtering unit is used to cluster the objects to be detected based on the object features to obtain at least one candidate cluster, and to filter out the target cluster from the candidate clusters.
14. An electronic device, characterized in that, It includes a processor and a memory, the memory storing an application program, and the processor running the application program within the memory to perform the steps of the cluster detection method according to any one of claims 1 to 12.
15. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the cluster detection method according to any one of claims 1 to 12.
16. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a plurality of instructions adapted for loading by a processor to perform the steps of the cluster detection method according to any one of claims 1 to 12.
Citation Information
Patent Citations
Junk account detection method, device and equipment
CN110119860A
Method for determining unified account and equipment thereof
CN111701247A
