A Target Group Identification Method and System Based on Graph Representation Learning

Abstract

A target group identification method based on graph representation learning is disclosed. The method includes: acquiring individual transaction data of an object to construct individual nodes; acquiring prior group data to construct prior group nodes, the prior group data including the relationship between the object and known groups; connecting each individual node and each prior group node based on the individual transaction data and the prior group data to construct a fusion graph; performing graph representation learning on the fusion graph to obtain enhanced features of each individual node; and identifying the target group based on the enhanced features of each individual node. A target group identification system based on graph representation learning and a computer-readable storage medium are also disclosed.

Description

Technical Field

[0001] This disclosure relates primarily to graph representation learning, and more particularly to target group identification methods and systems based on graph representation learning. Background Technology

[0002] Graphs are a common data structure, used to represent many real-world problems, such as social networks, search engines, advertising, and risk identification. Similarly, numerous examples of potentially risky groups exist in real life, such as those engaged in illegal financial activities. Group mining is needed to accurately identify these target groups.

[0003] Currently, graph representation learning and population mining are often treated as two independent tasks, or as two-stage tasks that are only fused at the result level. However, in many cases, known population information can help obtain better representations of graph nodes. This is especially true in security risk control scenarios, where there are naturally both large groups of risky individuals and individual risky points.

[0004] Therefore, it is desirable to provide a graph representation scheme that can better integrate individual and group information. Summary of the Invention

[0005] The following provides a brief overview of one or more aspects to offer a basic understanding of them. This overview is not an exhaustive summary of all conceived aspects, nor is it intended to identify the key or decisive elements of all aspects, nor to define the scope of any or all aspects. Its sole purpose is to present some concepts of one or more aspects in a simplified form as an introduction to the more detailed description that follows.

[0006] This disclosure proposes a target group identification method based on graph representation learning, comprising: acquiring individual transaction data of an object to construct individual nodes; acquiring prior group data to construct prior group nodes, the prior group data including the relationship between the object and known groups; connecting each individual node and each prior group node based on the individual transaction data and the prior group data to construct a fusion graph; performing graph representation learning on the fusion graph to obtain enhanced features of each individual node; and identifying the target group based on the enhanced features of each individual node.

[0007] In one embodiment of this disclosure, the edge between an individual node and a priori group node indicates that the corresponding object belongs to the corresponding known group, and the edge between different individual nodes indicates that there is a transaction relationship between the corresponding objects.

[0008] In one embodiment of this disclosure, graph representation learning of the fused graph includes: transforming the initial features of each body node and each prior group node in the fused graph to obtain corresponding transformed features; and updating the transformed features of each body node and each prior group node to obtain corresponding updated features.

[0009] In one embodiment of this disclosure, graph representation learning of the fused graph further includes: using the updated features of each individual node and each prior group node as new initial features for multiple rounds of transformation and updating, wherein the enhanced features of an individual node are the features of that individual node at the end of the iteration.

[0010] In one embodiment of this disclosure, the transformation features of each body node and each prior group node are updated using an update function based on a graph neural network.

[0011] In one embodiment of this disclosure, the graph neural network includes a convolutional neural network (GCN), a graph attention network (GAT), or a graph isomorphic network (GIN).

[0012] In one embodiment of this disclosure, the graph neural network is trained using prior population data.

[0013] In one embodiment of this disclosure, identifying a target group based on the enhanced features of each body node further includes: determining the density between each body node in the fused graph based on the enhanced features of each body node; and identifying the target group based on the density between each body node.

[0014] In one embodiment of this disclosure, identifying a target group based on the density between individual nodes further includes: determining individual nodes in the fused graph whose density meets a preset condition; and identifying the group of objects corresponding to the individual nodes that meet the preset condition as the target group.

[0015] This disclosure also proposes a target group identification system based on graph representation learning, comprising: an object data acquisition module for acquiring individual transaction data of objects to construct individual nodes; a prior data acquisition module for acquiring prior group data to construct prior group nodes, wherein the prior group data includes the relationship between the object and a known group; a fusion graph construction module for connecting each individual node and each prior group node based on the individual transaction data and the prior group data to construct a fusion graph; an enhancement feature module for performing graph representation learning on the fusion graph to obtain enhanced features of each individual node; and a target group identification module for identifying the target group based on the enhanced features of each individual node.

[0016] This disclosure also proposes a computer-readable storage medium storing a computer program that can be executed by a processor to perform the aforementioned target group identification method based on graph representation learning.

[0017] The technical solution disclosed herein can fuse prior group information with graph neural networks for graph construction and training, so that node representation updates can be better integrated with group information, thereby improving the performance of downstream tasks. Attached Figure Description

[0018] The features, essence, and advantages of this disclosure will become more apparent when understood in conjunction with the accompanying drawings. In the drawings, the same reference numerals are consistently used accordingly. It should be noted that the described drawings are schematic and non-limiting. In the drawings, some parts may be enlarged and are not drawn to scale for illustrative purposes.

[0019] Figure 1 An exemplary fusion diagram of an embodiment of this disclosure is shown.

[0020] Figure 2 A schematic diagram of a target group identification system based on graph representation learning according to an embodiment of the present disclosure is shown.

[0021] Figure 3 An exemplary flowchart of a target group identification method based on graph representation learning according to an embodiment of the present disclosure is shown.

[0022] Figure 4 An exemplary process for learning graph representations of a fused graph according to an embodiment of the present disclosure is shown.

[0023] Figure 5 An exemplary process for identifying a target group based on enhanced features of individual body nodes, according to an embodiment of this disclosure, is illustrated.

[0024] Figure 6 A block diagram of a target group identification system based on graph representation learning according to an embodiment of the present disclosure is shown.

[0025] Figure 7 A device block diagram of a target group identification system based on graph representation learning, according to an embodiment of the present disclosure, is shown. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of this disclosure clearer, the following detailed description is provided in conjunction with specific embodiments and the accompanying drawings. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the described exemplary embodiments. However, it will be apparent to those skilled in the art that the described embodiments can be practiced without some or all of these specific details. In other exemplary embodiments, well-known structures have not been described in detail to avoid unnecessarily obscuring the concepts of this disclosure. It should be understood that the specific embodiments described herein are merely illustrative of this disclosure and are not intended to limit it. Furthermore, the various aspects described in the embodiments can be combined arbitrarily without conflict.

[0027] Currently, graph representation learning and population mining are primarily treated as two independent tasks. Some approaches employ a two-stage learning method: first, population mining is performed, and then population information is used as node features for subsequent graph representation learning. This approach does not explicitly reveal the population relationship between two nodes during graph representation learning. In other words, population information does not propagate across the graph; only the population class features of each node are visible, thus failing to fully utilize prior population information.

[0028] The technical solution disclosed herein incorporates group information as prior information into graph neural networks for joint training, enabling node representations to better integrate group information, thereby improving the performance of downstream tasks (e.g., classification tasks, prediction tasks, etc.).

[0029] Figure 1 An exemplary fusion diagram 100 of an embodiment of this disclosure is shown.

[0030] As shown in the figure, fused graph 100 illustrates two types of nodes: individual nodes and prior group nodes. For ease of explanation, individual nodes are represented by unshaded circles in the figure, while prior group nodes are represented by shaded circles.

[0031] In one embodiment of this disclosure, each individual node can represent a single object, while each prior group node can represent a priori known group. The edges between individual nodes and prior group nodes ( Figure 1 The solid lines in the diagram represent objects belonging to a known group, while the edges between different individual nodes (…) Figure 1 The dashed lines in the diagram indicate that there is a transaction relationship between the corresponding objects.

[0032] Taking the field of financial transactions as an example, individual nodes can represent customers engaging in financial transactions, and edges between different individual nodes indicate that there is a transaction relationship between the corresponding customers. A priori known group can represent, for example, a group engaging in specific financial activities, and edges between individual nodes and a priori known group indicate that the corresponding customer belongs to that specific group.

[0033] By constructing such a fusion graph, node representations can be better integrated into group information, thereby making full use of prior group information and improving the performance of downstream tasks.

[0034] For ease of explanation, most aspects of this disclosure are described in conjunction with target group identification in the field of financial transactions. However, it should be noted that the target group identification techniques of this disclosure can be applied to target group identification in various other fields.

[0035] Figure 2 A schematic diagram 200 of a system for target group identification based on graph representation learning according to an embodiment of the present disclosure is shown.

[0036] As shown in the figure, the object to be identified and the prior group can be obtained first.

[0037] Taking the financial transaction field as an example, we can obtain candidate customers to be identified as well as previously known groups (such as known groups engaged in specific financial activities).

[0038] Subsequently, individual transaction data is obtained from the object, and then individual nodes are constructed. In addition, prior group data is obtained from the prior group, and then prior group nodes are constructed.

[0039] For example, financial transaction data of candidate customers within a specific time period can be obtained, and individual nodes can be constructed based on this data, where the information of candidate customers can be included in the node characteristics of the individual nodes.

[0040] Next, individual nodes and prior group nodes are fused based on individual transaction data and prior group data to construct a fused graph.

[0041] For example, prior group nodes can be connected to corresponding individual nodes based on prior group data to represent the affiliation between them. Similarly, individual node connections can be made based on individual transaction data to represent the transaction relationships between the individual nodes.

[0042] Then, graph representation learning can be performed on the fused graph to obtain enhanced features of individual nodes.

[0043] Then, the target group can be identified based on the enhanced features of individual nodes.

[0044] Compared to conventional techniques that separate graph representation learning from group mining / group recognition, the proposed solution integrates group relationships into graph representation learning, enabling better utilization of prior group information and improving the performance of downstream tasks. For example, in target group recognition tasks, it can significantly improve the accuracy of target group identification.

[0045] Figure 3 An exemplary flowchart of a target group identification method 300 based on graph representation learning according to an embodiment of the present disclosure is shown.

[0046] Method 300 begins at step 302. In step 302, individual transaction data of the object is obtained to construct individual nodes.

[0047] For example, transaction data of customers over a specific time period (e.g., one day, one week, one month, etc.) can be used to construct individual nodes. Each node can represent a customer, and the customer's personal information can be included in the node's characteristics.

[0048] In step 304, prior group data is obtained to construct prior group nodes, which includes the relationship between the object and known groups.

[0049] For example, group data of a known group (e.g., group transaction data, group member information, etc.) can be obtained to construct prior group nodes. Each node can represent the corresponding prior group, and group information can be included in the node features.

[0050] In step 306, each individual node and each prior group node are connected based on the individual transaction data and the prior group data to construct a fusion graph.

[0051] In one embodiment of this disclosure, the edge between an individual node and a priori group node indicates that the corresponding object belongs to the corresponding known group, and the edge between different individual nodes indicates that there is a transaction relationship between the corresponding objects.

[0052] For example, if it is known that an object (e.g., a customer) belongs to a specific known group, then in the graph, the individual node representing that object can be connected to the prior group node representing that known group. Additionally, if there is a financial transaction relationship between two objects, then in the graph, the individual nodes representing those two objects can be connected. In this way, a fusion graph can be constructed.

[0053] In one embodiment of this disclosure, the fused diagram is a bipartite diagram.

[0054] A bipartite graph, also known as a bipartite graph, is a special model in graph theory. Let G = (V, E) be an undirected graph. If the vertex set V can be partitioned into two disjoint subsets (A, B), and each edge (i, j) in the graph is associated with two vertices i and j belonging to these two distinct subsets (i in A, j in B), then graph G is called a bipartite graph. In short, it means that the vertex set V can be partitioned into two disjoint subsets, and each edge in the graph is associated with two vertices belonging to these two disjoint subsets, and the vertices within the two subsets are not adjacent.

[0055] In step 308, graph representation learning is performed on the fused graph to obtain enhanced features for each volume node.

[0056] In one embodiment of this disclosure, graph representation learning of the fused graph includes: transforming the initial features of each body node and each prior group node in the fused graph to obtain corresponding transformed features; and updating the transformed features of each body node and each prior group node to obtain corresponding updated features.

[0057] In one embodiment of this disclosure, graph representation learning of the fused graph further includes: using the updated features of each individual node and each prior group node as new initial features for multiple rounds of transformation and updating, wherein the enhanced features of an individual node are the features of that individual node at the end of the iteration.

[0058] In one embodiment of this disclosure, the transformation features of each body node and each prior group node are updated using an update function based on a graph neural network, wherein the neural network includes, but is not limited to, a convolutional neural network (GCN), a graph attention network (GAT), or a graph isomorphic network (GIN).

[0059] In one embodiment of this disclosure, the graph neural network is trained using prior population data.

[0060] The detailed process of learning graph representations from fused graphs will be discussed below. Figure 4 Further description is required.

[0061] Finally, in step 310, the target group is identified based on the enhanced features of each individual node.

[0062] For example, the density between individual nodes in the fusion graph can be determined based on the enhanced features of each individual node, and the target group can be identified based on the density.

[0063] The detailed process of identifying target groups based on the enhanced features of individual nodes will be discussed below. Figure 5 Further description is required.

[0064] Figure 4An exemplary process 400 for learning graph representations of a fused graph according to an embodiment of the present disclosure is shown.

[0065] like Figure 4 As shown, after constructing a fusion graph based on individual transaction data and prior group data, the initial features of each individual node and each prior group node in the fusion graph can be transformed to obtain the corresponding transformed features.

[0066] In one embodiment of this disclosure, the initial features of each volume node and each prior group node can be linearly transformed based on the projection matrix of the fused graph to obtain the corresponding transformed features.

[0067] For example, individual nodes can be transformed using the following formula:

[0068]

[0069] Where x i W represents the individual node before the transformation. n Represents the projection matrix. This represents the transformed individual node.

[0070] The prior population nodes can be transformed using the following formula:

[0071]

[0072] Where g i W represents the prior group of nodes before the transformation. g Represents the projection matrix. This represents the transformed prior group nodes.

[0073] In one embodiment of this disclosure, the transformation process may further include a representation transformation of the edges associated with the nodes. For example, the edges of individual nodes and the edges of prior group nodes can be transformed using a corresponding edge projection matrix.

[0074] After obtaining the transformation features of each individual node and each prior group node, these transformation features can be updated to obtain the corresponding updated features.

[0075] For example, an individual node can be updated using the following formula:

[0076]

[0077] Prior group nodes can be updated using the following formula:

[0078]

[0079] Where N(v; A) represents the neighboring nodes of node v in the fusion graph, which include two types: group nodes and individual nodes. This represents the edge projection matrix. UPDATE represents the update function, and AGGREGATE represents the aggregation function.

[0080] Aggregation functions are used to aggregate the features of a target node's neighbors, including the target node's own features, the features of its neighboring nodes, and the edge features between the target node and its neighboring nodes, and then form a message vector that is passed to the target node.

[0081] The update function is used to update the node features of the next layer node, which combines the features of the current layer node with the message obtained from the aggregation function.

[0082] In one embodiment of this disclosure, UPDATE and AGGREGATE can be implemented using update functions and aggregation functions of a graph neural network.

[0083] Graph Neural Networks (GNNs) are a framework that has emerged in recent years, utilizing deep learning to directly learn from graph-structured data. By formulating certain strategies on the nodes and edges of a graph, GNNs transform graph-structured data into a standardized representation, which is then fed into various neural networks for training, achieving excellent results in tasks such as node classification, edge information propagation, and graph clustering.

[0084] In one example of this disclosure, a graph neural network may include a graph convolutional network (GCN), a graph attention network (GAT), or a graph isomorphic network (GIN).

[0085] Graph Convolutional Networks (GCNs) are convolutional neural networks that can directly operate on graphs and utilize their structural information. GCNs extend convolutional neural networks to the graph domain, enabling end-to-end learning of node feature and structural information. The convolutional network structure is determined through a local first-order approximation of spectral graph convolution, and the network model is trained using partially labeled node data from graph structure data, allowing the network model to further classify unlabeled data.

[0086] Graph Attention Network (GAT) is a novel neural network architecture based on graph-structured data. It utilizes hidden self-attention layers to address the shortcomings of previous graph convolutional or approximation-based methods. By stacking layers that allow nodes to process features of their neighborhood, different weights can be implicitly assigned to different nodes in the neighborhood without any costly matrix operations or reliance on prior knowledge of the graph structure.

[0087] Graph Isomorphism Networks (GINs) aim to solve the problem of classifying graphs using graph neural networks. Using GNNs such as GCNs, we can obtain the embedding vector representation of each node. By concatenating the vector representations of all nodes in the graph, we can obtain the entire graph's embedding. Graph representation learning based on GINs first computes the node representations, and then performs graph pooling on these node representations to obtain the final graph representation.

[0088] It should be noted that although three specific graph neural networks are described in this disclosure, this disclosure is not limited thereto. In practical implementations, those skilled in the art can employ other suitable graph neural networks as needed.

[0089] In one embodiment of this disclosure, the graph neural network can be trained using prior population data.

[0090] After obtaining the updated features of each individual node and each prior group node, it can be determined whether the iteration termination condition is met. For example, if the updated features of each individual node and each prior group node meet a preset condition (e.g., downstream task convergence), then the iteration termination condition can be considered met. At this point, the graph representation learning of the fused graph ends, and the features of the individual nodes are output as enhanced features.

[0091] If the updated features of each individual node and each prior group node do not meet the preset conditions, then the updated features of each individual node and each prior group node can be used as new initial features, and the process can proceed to the next iteration. When the iteration terminates, the features of the individual nodes are the output enhanced features.

[0092] Figure 5 An exemplary process 500 for identifying a target group based on enhanced features of individual body nodes, according to an embodiment of this disclosure, is shown.

[0093] Process 500 begins at step 502. In step 502, graph representation learning is performed on the fused graph to obtain enhanced features for each volume node.

[0094] Step 502 corresponds to Figure 3 Step 308 will not be repeated here.

[0095] In step 504, the density between the individual volume nodes in the fusion graph is determined based on the enhanced features of each volume node.

[0096] In this disclosure, the closeness between individual nodes can be represented in various ways. For example, the closeness between individual nodes can be represented by distance, similarity, correlation, etc. For instance, the closer the distance between individual nodes, the closer the closeness between them. Alternatively, if the similarity or correlation between individual nodes is higher, the closer the closeness between them is also considered.

[0097] For example, the feature similarity between the augmented features of each individual node can be calculated, and the density between individual nodes can be determined based on this feature similarity.

[0098] In step 506, individual nodes in the fusion graph whose density meets the preset conditions are identified.

[0099] After determining the density between individual nodes, it can be determined whether these densities meet preset conditions, thereby identifying individual nodes that meet the preset conditions.

[0100] For example, the density between individual nodes can be compared with a preset density threshold to determine the individual nodes that meet the density threshold.

[0101] In practice, preset conditions (e.g., preset density thresholds) can be set based on experience, through training or experimental processes, or through theoretical calculations.

[0102] In step 508, the group of objects corresponding to individual nodes that meet the preset conditions is identified as the target group.

[0103] After identifying individual nodes that meet the preset conditions, the objects corresponding to these nodes can be found, and the group composed of these objects can be identified as the target group.

[0104] Figure 6 A block diagram of a target group identification system 600 based on graph representation learning according to an embodiment of the present disclosure is shown.

[0105] See Figure 6 The system 600 may include an object data acquisition module 602, a prior data acquisition module 604, a fusion graph construction module 606, an enhanced feature module 608, and a target group identification module 610. Each of these modules may be directly or indirectly connected to or communicate with each other on one or more buses 612.

[0106] The object data acquisition module 602 can acquire individual transaction data of an object to construct individual nodes.

[0107] The prior data acquisition module 604 can acquire prior group data to construct prior group nodes, which includes the relationship between the object and the known group.

[0108] In one embodiment of this disclosure, the edge between an individual node and a priori group node indicates that the corresponding object belongs to the corresponding known group, and the edge between different individual nodes indicates that there is a transaction relationship between the corresponding objects.

[0109] The fusion graph construction module 606 can connect individual nodes and prior group nodes based on individual transaction data and prior group data to construct a fusion graph.

[0110] The enhanced feature module 608 can perform graph representation learning on the fused graph to obtain enhanced features for each volume node.

[0111] In one embodiment of this disclosure, the enhanced feature module 608 may also be configured to: transform the initial features of each volume node and each prior group node in the fusion graph to obtain corresponding transformed features; and update the transformed features of each volume node and each prior group node to obtain corresponding updated features.

[0112] In one embodiment of this disclosure, the enhanced feature module 608 may also be configured to: perform multiple rounds of transformation and updating using the updated features of each individual node and each prior group node as new initial features, wherein the enhanced feature of an individual node is the feature of that individual node at the end of the iteration.

[0113] In one embodiment of this disclosure, the transformation features of each body node and each prior group node are updated using an update function based on a graph neural network.

[0114] In one embodiment of this disclosure, the graph neural network includes a convolutional neural network, a graph attention network, or a graph isomorphic network.

[0115] In one embodiment of this disclosure, the graph neural network is trained using prior population data.

[0116] The target group identification module 610 can identify the target group based on the enhanced features of each individual node.

[0117] In one embodiment of this disclosure, the target group identification module 610 is further configured to: determine the density between individual body nodes in the fusion graph based on the enhanced features of each body node; and identify the target group based on the density between each body node.

[0118] In one embodiment of this disclosure, the target group identification module 610 is further configured to: determine individual nodes in the fusion graph whose density meets preset conditions; and identify the group of objects corresponding to the individual nodes that meet the preset conditions as the target group.

[0119] Although Figure 6The diagram illustrates specific modules of system 600, but it should be understood that these modules are exemplary and not limiting. In different implementations, one or more of these modules may be combined, split, removed, or additional modules may be added. For example, in some implementations, object data acquisition module 602 and prior data acquisition module 604 may be merged into a single module. In some implementations, system 600 may also include additional modules. For example, system 600 may also include an output module (not shown) to output target group identification results.

[0120] Figure 7 A block diagram of a device 700 including a target group identification system based on graph representation learning, according to an embodiment of the present disclosure, is shown.

[0121] This device illustrates a general hardware environment in which the present disclosure can be applied according to exemplary embodiments thereof.

[0122] Now refer to Figure 7 Device 700 is described as an exemplary embodiment of a hardware device that can be applied to various aspects of this disclosure. Device 700 can be any machine configured to perform processing and / or computation, and can be, but is not limited to, a workstation, server, desktop computer, laptop computer, tablet computer, personal digital assistant (PDA), smartphone, or any combination thereof.

[0123] Device 700 may include components that can be connected to or communicate with bus 712 via one or more interfaces. For example, device 700 may include bus 712, processor 702, memory 704, input device 708, and output device 710, etc.

[0124] Processor 702 can be any type of processor and may include, but is not limited to, general-purpose processors and / or dedicated processors (e.g., special-purpose processing chips), intelligent hardware devices (e.g., general-purpose processors, DSPs, CPUs, microcontrollers, ASICs, FPGAs, programmable logic devices, discrete gate or transistor logic components, discrete hardware components, or any combination thereof). In some cases, processor 702 may be configured to use a memory controller to operate a memory array. In other cases, a memory controller (not shown) may be integrated into processor 702. Processor 702 may be responsible for managing the bus and general processing, including executing software stored in memory. Processor 702 may also be configured to perform various functions described herein related to target group identification based on graph representation learning. For example, processor 702 may be configured to: acquire individual transaction data of an object to construct individual nodes; acquire prior group data to construct prior group nodes, the prior group data including the relationship between the object and a known group; connect each individual node and each prior group node based on the individual transaction data and the prior group data to construct a fusion graph; perform graph representation learning on the fusion graph to obtain enhanced features of each individual node; and identify the target group based on the enhanced features of each individual node.

[0125] Memory 704 can be any storage device capable of storing data. Memory 704 may include, but is not limited to, disk drives, optical storage devices, solid-state storage, floppy disks, hard disks, magnetic tape or any other magnetic media, optical discs or any other optical media, ROM (read-only memory), RAM (random access memory), cache memory and / or any other memory chip or cartridge, and / or any other medium from which a computer can read data, instructions and / or code. Memory 704 may store computer-executable software 706 including computer-readable instructions that, when executed, cause a processor to perform the various functions described herein related to target group identification based on graph representation learning.

[0126] Input device 708 can be any type of device that can be used to input information.

[0127] The output device 710 can be any type of device used for outputting information. In one case, the output device 710 can be any type of output device capable of displaying information.

[0128] The technical solution disclosed herein integrates prior group information with graph neural networks, enabling node representation updates to better incorporate group information and greatly improving the performance of downstream tasks.

[0129] The detailed description above, in conjunction with the accompanying drawings, describes examples but does not represent all examples that can be implemented or fall within the scope of the claims. The terms "example" and "exemplary" are used in this specification to mean "serving as an example, instance, or illustration" and do not imply "superiority or superiority over other examples."

[0130] Throughout this specification, the terms "an embodiment" or "an embodiment" mean that a particular feature, structure, or characteristic described in connection with that embodiment is included in at least one embodiment of this disclosure. Therefore, the use of these phrases may refer to more than one embodiment. Furthermore, the described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0131] The preceding description is provided to enable any person skilled in the art to practice the various aspects described herein. Various modifications to these aspects will readily be understood by those skilled in the art, and the universal principles defined herein can be applied to other aspects. Therefore, the claims are not intended to be limited to the aspects shown herein, but are to be granted the full scope consistent with the language of the claims, wherein references to the singular form of an element, unless specifically stated otherwise, are not intended to mean “one and only one,” but rather “one or more.” Unless specifically stated otherwise, the term “some” refers to one or more. All structural and functional equivalents of the elements of the various aspects described throughout this disclosure, whether now or hereafter known to a person skilled in the art, are expressly incorporated herein by reference and are intended to be covered by the claims.

[0132] It should also be noted that these embodiments may be described as processes depicted as flowcharts, flow diagrams, structure diagrams, or block diagrams. Although a flowchart may describe the operations as a sequential process, many of these operations can be executed in parallel or concurrently. Furthermore, the order of these operations can be rearranged.

[0133] While various embodiments have been described and illustrated, it should be understood that the embodiments are not limited to the precise configurations and components described above. Various modifications, substitutions, and improvements that will be apparent to those skilled in the art can be made to the arrangement, operation, and details of the apparatus disclosed herein without departing from the scope of the claims.

Claims

1. A target group identification method based on graph representation learning, comprising: Obtain individual transaction data of objects to construct individual nodes; Acquire prior group data to construct prior group nodes, wherein the prior group data includes the relationship between the object and a known group; Based on the individual transaction data and the prior group data, connect each individual node and each prior group node to construct a fusion graph; Graph representation learning is performed on the fused graph to obtain enhanced features for each volume node; as well as Identifying target groups based on the enhanced features of each individual node. The graph representation learning of the fused graph includes: The initial features of each volume node and each prior group node in the fusion graph are transformed to obtain the corresponding transformed features; The transformation features of each individual node and each prior group node are updated to obtain the corresponding updated features; and The updated features of each individual node and each prior group node are used as new initial features for multiple rounds of transformation and updating. The enhanced features of an individual node are the features of that individual node at the end of the iteration.

2. As described in claim 1, the edge between an individual node and a priori group node indicates that the corresponding object belongs to the corresponding known group, and the edge between different individual nodes indicates that there is a transaction relationship between the corresponding objects.

3. The method as described in claim 1, wherein the transformation features of each individual node and each prior group node are updated using an update function based on a graph neural network.

4. The method of claim 3, wherein the graph neural network includes a convolutional neural network (GCN), a graph attention network (GAT), or a graph isomorphic network (GIN).

5. The method of claim 3, wherein the graph neural network is trained using the prior population data.

6. The method of claim 1, wherein identifying the target group based on the enhanced features of each body node further comprises: The density between the individual nodes in the fusion graph is determined based on the enhanced features of each individual node. as well as Identify target groups based on the density between individual nodes.

7. The method of claim 6, wherein identifying the target group based on the density between individual nodes further comprises: Identify individual nodes in the fusion graph whose density meets preset conditions; The group of objects corresponding to individual nodes that meet the preset conditions is identified as the target group.

8. A target group identification system based on graph representation learning, comprising: The object data acquisition module acquires individual transaction data of objects to construct individual nodes; The prior data acquisition module acquires prior group data to construct prior group nodes, wherein the prior group data includes the relationship between the object and a known group; The fusion graph construction module connects each individual node and each prior group node based on the individual transaction data and the prior group data to construct a fusion graph. The enhanced feature module performs graph representation learning on the fused graph to obtain enhanced features for each volume node; as well as The target group identification module identifies the target group based on the enhanced features of each individual node. The enhanced feature module is configured to perform graph representation learning on the fused graph through the following operations: The initial features of each volume node and each prior group node in the fusion graph are transformed to obtain the corresponding transformed features; The transformation features of each individual node and each prior group node are updated to obtain the corresponding updated features; as well as The updated features of each individual node and each prior group node are used as new initial features for multiple rounds of transformation and updating. The enhanced features of an individual node are the features of that individual node at the end of the iteration.

9. A computer-readable storage medium storing a computer program that can be executed by a processor to perform the method as described in any one of claims 1-7.

Images