Account detection method and device, electronic equipment and computer readable storage medium

Abstract

Embodiments of the present application provide a kind of account detection method, device, electronic equipment and computer readable storage medium, involve artificial intelligence field.The method comprises: based on fund flow information, the seed account determined is carried out relevance analysis, obtain at least one first associated account, then based on fund flow information, the at least one first associated account is carried out relevance analysis respectively, obtain at least one first candidate account, again based on the graph neural network model trained, the at least one first candidate account is detected, determine the target relevant account in the at least one first candidate account.This way, through seed account determined can be deep mining, even if the target relevant account without opinion feedback information, it can also be mined by above-mentioned way, to solve the problem that target cluster account cannot be accurately found, simultaneously, the purpose of early detection, early discovery to target relevant account is also realized.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and more specifically, to an account detection method, apparatus, electronic device, and computer-readable storage medium. Background Technology

[0002] In daily life, many individuals exhibit unusual financial behavior, often belonging to the same target group. Such groups typically stop using the same account after successfully obtaining funds several times. Those who have been successfully funded can usually only provide feedback to the target group's account. Therefore, once members of the target group change accounts, the group becomes uncontrollable. Consequently, controlling the target group solely through feedback is not only slow but also insufficiently effective.

[0003] In existing technologies, it is common practice to pre-define some rules and then use these rules to detect accounts. However, this approach has the following problems:

[0004] 1) While ensuring high accuracy, only low coverage can be achieved because manually formulated rules are inherently difficult to detect potential abnormal behavior characteristics.

[0005] 2) For target clusters that discover abnormal behavior through rules, rules need to be redefined for different abnormal behavior scenarios, which is not only time-consuming and labor-intensive, but also has poor scalability. Summary of the Invention

[0006] This application provides a method, apparatus, electronic device, and computer-readable storage medium for account detection, which can solve the problems of low coverage, time-consuming and labor-intensive processes, and poor scalability of existing detection methods. The technical solution is as follows:

[0007] According to one aspect of this application, an account detection method is provided, the method comprising:

[0008] Based on the fund flow information, a correlation analysis is performed on the identified seed accounts to obtain at least one first associated account that has fund transactions with the seed account.

[0009] Based on the fund flow information, a correlation analysis is performed on the at least one first associated account to obtain at least one first candidate account that has fund transactions with the at least one first associated account;

[0010] The trained graph neural network model is used to detect the at least one first candidate account and identify the target related account among the at least one first candidate account.

[0011] In one or more embodiments, the detection of the at least one first candidate account based on the trained graph neural network model includes:

[0012] An account graph network is constructed using the seed account and the at least one first candidate account;

[0013] The graph neural network model is used to detect the account graph network.

[0014] In one or more embodiments, constructing an account graph network using the seed account and the at least one first candidate account includes:

[0015] For the current first candidate account, perform the following first detection step: when it is detected that the seed account and any one of the current first candidate accounts have fund flow information with the same target account, generate a connection line between the seed account and any one of the current first candidate accounts;

[0016] The next first candidate account encountered is taken as the current first candidate account, and the first detection step described above is repeated until all at least one first candidate account has been traversed and detected, resulting in at least one connection line.

[0017] An account graph network is generated based on at least one of the connection lines.

[0018] In one or more embodiments, the step of performing correlation analysis on the identified seed accounts based on fund flow information to obtain at least one first associated account that has fund transactions with the seed accounts includes:

[0019] Identify at least one account that has fund flow information with the seed account, and obtain at least one first associated account that has fund transactions with the seed account;

[0020] The correlation analysis is performed on the at least one first associated account based on the fund flow information to obtain at least one first candidate account that has fund transactions with the at least one first associated account, including:

[0021] Identify at least one account that has fund flow information with at least one first associated account, and obtain at least one first candidate account that has fund transactions with the at least one first associated account.

[0022] In one or more embodiments, the step of using the graph neural network model to detect the account graph network includes:

[0023] Perform the following second detection step: For any first candidate account in the account graph network, identify at least one target seed account that has a connection line with any first candidate account;

[0024] The graph neural network model is used to perform forward propagation on the at least one target seed account to obtain the deep feature representation of any first candidate account;

[0025] The activation function in the graph neural network model is used to predict the probability value of the deep feature representation to obtain the prediction result of any first candidate account, and the prediction result is used to determine whether any first candidate account is a target-related account.

[0026] Repeat the second detection step described above until all at least one first candidate account has been detected.

[0027] In one or more embodiments, the trained graph neural network model is trained in the following manner:

[0028] Perform the following training steps: conduct correlation analysis on the sample seed account to obtain at least one second associated account that has financial transactions with the sample seed account;

[0029] Based on the fund flow information, a correlation analysis is performed on the at least one first associated account to obtain at least one second candidate account;

[0030] A preset original graph neural network model is used to detect any one of the at least one second candidate accounts, and the detection result of any one of the second candidate accounts that has a financial transaction with the at least one second associated account is obtained.

[0031] Based on the detection results and tags of any of the second candidate accounts, the forward propagation error is calculated using the loss function in the original graph neural network model to obtain the error value;

[0032] The error value is used for backpropagation to update the weight matrix of the original graph neural network model, resulting in an updated graph neural network model.

[0033] Using the updated graph neural network model as the current original graph neural network, repeat the above training steps until the loss function converges to its minimum value, thus obtaining the trained graph neural network model.

[0034] According to another aspect of this application, an account detection apparatus is provided, the apparatus comprising:

[0035] The first mining module is used to perform correlation analysis on the identified seed accounts based on fund flow information to obtain at least one first associated account that has fund transactions with the seed account.

[0036] The second mining module is used to perform correlation analysis on the at least one first associated account based on the fund flow information, and obtain at least one first candidate account that has fund transactions with the at least one first associated account.

[0037] The detection module is used to detect the at least one first candidate account based on the trained graph neural network model, and to determine the target related account among the at least one first candidate account.

[0038] In one or more embodiments, the detection module includes:

[0039] A construction submodule is used to construct an account graph network using the seed account and the at least one first candidate account;

[0040] The detection submodule is used to detect the account graph network using the graph neural network model.

[0041] In one or more embodiments, the construction submodule includes:

[0042] A connection line generation unit is used to generate a connection line between the seed account and any of the current first candidate accounts when it is detected that the seed account and any of the current at least one first candidate accounts have fund flow information with the same target account.

[0043] The next first candidate account encountered during the traversal is taken as the current first candidate account, and the connection line generation unit is called repeatedly until all at least one first candidate account has been traversed and detected, resulting in at least one connection line.

[0044] An account graph network generation unit is used to generate an account graph network based on the at least one connection line.

[0045] In one or more embodiments, the first mining module is specifically used for:

[0046] Identify at least one account that has fund flow information with the seed account, and obtain at least one first associated account that has fund transactions with the seed account;

[0047] The second mining module is specifically used for:

[0048] Identify at least one account that has fund flow information with at least one first associated account, and obtain at least one first candidate account that has fund transactions with the at least one first associated account.

[0049] In one or more embodiments, the detection submodule includes:

[0050] The first processing unit is configured to determine, for any first candidate account in the account graph network, at least one target seed account that has a connection line with any first candidate account.

[0051] The second processing unit is used to perform forward propagation on the at least one target seed account using the graph neural network model to obtain a deep feature representation of any first candidate account.

[0052] The third processing unit is used to perform probability value prediction on the deep feature representation using the activation function in the graph neural network model to obtain the prediction result of any first candidate account, and to determine whether any first candidate account is a target-related account based on the prediction result.

[0053] The first processing unit, the second processing unit, and the third processing unit are repeatedly called until all at least one first candidate account has been detected.

[0054] In one or more embodiments, it further includes:

[0055] The first mining module is also used to perform correlation analysis on the sample seed account to obtain at least one second associated account that has financial transactions with the sample seed account;

[0056] The second mining module is further configured to perform correlation analysis on the at least one first associated account based on the fund flow information, and obtain at least one second candidate account that has fund transactions with the at least one second associated account;

[0057] The detection module is further configured to use a preset original graph neural network model to detect any one of the at least one second candidate accounts, and obtain the detection result of any one of the second candidate accounts;

[0058] The calculation module is used to calculate the forward propagation error based on the detection result and label of any second candidate account, using the loss function in the original graph neural network model, and obtain the error value.

[0059] The update module is used to perform backpropagation using the error value to update the weight matrix of the original graph neural network model, thereby obtaining the updated graph neural network model.

[0060] Using the updated graph neural network model as the current original graph neural network, the first mining module, the second mining module, the detection module, the calculation module, and the update module are repeatedly called until the minimum value of the loss function converges, thus obtaining the trained graph neural network model.

[0061] According to another aspect of this application, an electronic device is provided, the electronic device comprising:

[0062] One or more processors;

[0063] Memory;

[0064] One or more computer programs, wherein the one or more computer programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs being configured to perform operations corresponding to the account detection method as shown in the first aspect of this application.

[0065] According to another aspect of this application, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the account detection method shown in the first aspect of this application.

[0066] 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 various alternative implementations of any of the above aspects.

[0067] The beneficial effects of the technical solution provided in this application are:

[0068] Based on fund flow information, correlation analysis is performed on the identified seed accounts to obtain at least one first associated account with fund transactions with the seed accounts. Then, correlation analysis is performed on each of the at least one first associated account based on fund flow information to obtain at least one first candidate account with fund transactions with the at least one first associated account. Finally, a trained graph neural network model is used to detect the at least one first candidate account to determine the target related accounts among the at least one first candidate account. In this way, deep mining can be carried out using the identified seed accounts. Even target related accounts without feedback information can be mined using the above method, thus solving the problem of not being able to accurately discover target cluster accounts, and also achieving the goal of early detection and discovery of target related accounts. Attached Figure Description

[0069] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments of this application will be briefly introduced below.

[0070] Figure 1 A schematic diagram of the application environment for implementing an account detection method provided in this application embodiment;

[0071] Figure 2 A flowchart illustrating an account detection method provided in an embodiment of this application;

[0072] Figure 3 A schematic diagram of correlation analysis provided for embodiments of this application. Figure 1 ;

[0073] Figure 4 A schematic diagram of correlation analysis provided for embodiments of this application. Figure 2 ;

[0074] Figure 5 The diagram shows the effect of generating connecting lines provided in the embodiments of this application;

[0075] Figure 6 A schematic diagram of an account graph network provided in an embodiment of this application;

[0076] Figure 7 A schematic diagram of fund flow provided for an embodiment of this application;

[0077] Figure 8 A flowchart illustrating the training method for the trained graph neural network model provided in this application embodiment;

[0078] Figure 9 This is a schematic diagram of the structure of an account detection device provided in an embodiment of this application;

[0079] Figure 10 This is a schematic diagram of the structure of an electronic device for account detection provided in an embodiment of this application. Detailed Implementation

[0080] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain this application, and should not be construed as limiting this application.

[0081] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this application means the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. It should be understood that when we say an element is “connected” or “coupled” to another element, it can be directly connected or coupled to the other element, or there may be intermediate elements. Furthermore, “connected” or “coupled” as used herein can include wireless connections or wireless coupling. The term “and / or” as used herein includes all or any units and all combinations of one or more associated listed items.

[0082] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0083] First, let's introduce and explain several terms used in this application:

[0084] Artificial intelligence (AI) is the theory, methods, technology, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. In other words, AI is a comprehensive technology within computer science that attempts to understand the essence of intelligence and produce a new kind of intelligent machine that can react in a way similar to human intelligence. AI studies the design principles and implementation methods of various intelligent machines, enabling them to possess the functions of perception, reasoning, and decision-making.

[0085] Artificial intelligence (AI) is a comprehensive discipline encompassing a wide range of fields, including both hardware and software technologies. Fundamental AI technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies primarily include computer vision, speech processing, natural language processing, and machine learning / deep learning.

[0086] Machine learning (ML) is a multidisciplinary field involving probability theory, statistics, approximation theory, convex analysis, and algorithm complexity theory. It specifically studies how computers can simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to endow computers with intelligence; its applications span all areas of artificial intelligence. Machine learning and deep learning typically include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and instructional learning.

[0087] In this application, training the graph neural network model requires the use of machine learning techniques from artificial intelligence.

[0088] In daily life, individuals exhibiting unusual financial behavior emerge. These individuals belong to the same target group, which typically stops using the same account after successfully borrowing money several times. Those who have successfully obtained funds can usually only provide feedback on the target group's account. Therefore, once the target group changes accounts, it becomes impossible to control the group. Consequently, controlling the target group solely through feedback and other information is not only slow but also insufficiently effective.

[0089] In existing technologies, it is common practice to pre-define some rules and then use these rules to detect accounts. However, this approach has the following problems:

[0090] 1) While ensuring high accuracy, only low coverage can be achieved because manually formulated rules are inherently difficult to discover potential features;

[0091] 2) When mining target clusters through rules, rules need to be redefined for different scenarios, which is not only time-consuming and labor-intensive, but also has poor scalability.

[0092] The account detection method, apparatus, electronic device, and computer-readable storage medium provided in this application are intended to solve the above-mentioned technical problems of the prior art.

[0093] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0094] This invention provides an application environment for implementing an account detection method. See [link to relevant documentation]. Figure 1The application environment includes a first device 101 and a second device 102. The first device 101 and the second device 102 are connected via a network. The first device 101 is the accessing device, and the second device 102 is the accessed device. The first device 101 can be a terminal, and the second device 102 can be a server. The terminal may have the following characteristics:

[0095] (1) In terms of hardware, the device has a central processing unit, memory, input components, and output components. In other words, the device is often a microcomputer device with communication capabilities. In addition, it can have multiple input methods, such as keyboard, mouse, touch screen, microphone, and camera, and the input can be adjusted as needed. At the same time, the device often has multiple output methods, such as receiver, display screen, etc., which can also be adjusted as needed;

[0096] (2) In terms of software, the device must have an operating system, such as Windows Mobile, Symbian, Palm, Android, iOS, etc. At the same time, these operating systems are becoming increasingly open, and personalized computer programs developed based on these open operating system platforms are emerging in large numbers, such as address books, calendars, notepads, calculators, and various games, which greatly meet the needs of personalized users;

[0097] (3) In terms of communication capabilities, the device has flexible access methods and high bandwidth communication performance, and can automatically adjust the selected communication method according to the selected service and the environment, thus facilitating user use. The device can support mobile communication based on 3GPP (3rd Generation Partnership Project), 4GPP (4th Generation Partnership Project), 5GPP (5th Generation Partnership Project), LTE (Long Term Evolution), and WIMAX (World Interoperability for Microwave Access), computer network communication based on TCP / IP (Transmission Control Protocol / Internet Protocol) and UDP (User Datagram Protocol), as well as short-range wireless transmission based on Bluetooth and infrared transmission standards. It not only supports voice services, but also a variety of wireless data services.

[0098] (4) In terms of functionality, the equipment places greater emphasis on humanization, personalization, and multi-functionality. With the development of computer technology, the equipment has shifted from a "device-centric" model to a "human-centric" model, integrating embedded computing, control technology, artificial intelligence technology, and biometric authentication technology, fully embodying the principle of human-centeredness. Due to the development of software technology, the equipment can be customized according to individual needs, becoming more personalized. At the same time, the equipment itself integrates numerous software and hardware components, making its functions increasingly powerful.

[0099] A server can be a standalone physical server, a server cluster or distributed system consisting of multiple physical servers, or a cloud server providing cloud computing services. Multiple servers can form a blockchain, with each server acting as a node on the blockchain.

[0100] In the above application environment, an account verification method can be implemented, such as... Figure 2 As shown, the method includes:

[0101] Step S201: Based on the fund flow information, perform correlation analysis on the identified seed accounts to obtain at least one first associated account that has fund transactions with the seed accounts.

[0102] In this embodiment of the invention, the seed account can be a pre-determined target account, but the specific method of determining the seed account is not limited. That is to say, all methods that can determine the seed account are applicable to this embodiment of the invention.

[0103] After obtaining the seed account, you can perform correlation analysis on the seed account based on the fund flow information to find at least one primary associated account that has fund transactions with the seed account.

[0104] Step S202: Based on the fund flow information, perform correlation analysis on at least one first associated account to obtain at least one first candidate account that has fund transactions with at least one first associated account;

[0105] In practical applications, the receiving account of the target cluster may also conduct fund transactions with other target accounts, but there are no fund transactions between multiple target accounts. Therefore, after obtaining the first associated account, it is necessary to further analyze the correlation of the first associated account to obtain at least one first candidate account that has fund transactions with at least one first associated account.

[0106] Step S203: Detect at least one first candidate account based on the trained graph neural network model, and identify the target related account among at least one first candidate account.

[0107] After obtaining at least one first candidate account through correlation analysis, a trained graph neural network model can be used to detect each of the at least one first candidate account, thereby determining whether the target account and the receiving account exist among the at least one first candidate account.

[0108] In this embodiment of the invention, correlation analysis is performed on the identified seed accounts based on fund flow information to obtain at least one first associated account that has fund transactions with the seed accounts. Then, correlation analysis is performed on each of the at least one first associated account based on fund flow information to obtain at least one first candidate account that has fund transactions with the at least one first associated account. Finally, a trained graph neural network model is used to detect the at least one first candidate account to determine the target related accounts among the at least one first candidate account. In this way, deep mining can be carried out using the identified seed accounts. Even target related accounts without feedback information can be mined through the above method, thereby solving the problem of not being able to accurately discover accounts in the target cluster. At the same time, it also achieves the purpose of early detection and discovery of target related accounts.

[0109] In another embodiment, for example Figure 2 The steps of an account detection method are explained in detail below.

[0110] Step S201: Based on the fund flow information, perform correlation analysis on the identified seed accounts to obtain at least one first associated account that has fund transactions with the seed accounts.

[0111] In this embodiment of the invention, any type of account has, but is not limited to, the following functions: social communication and financial transactions. Social communication includes, but is not limited to, sending and receiving messages; financial transactions include, but are not limited to, transferring funds out and transferring funds in.

[0112] In practical applications, target clusters typically use target-related accounts for transactions. These target-related accounts are mainly divided into two types: communication accounts and receiving accounts. Communication accounts are primarily used for social communication, but also for financial transactions; while receiving accounts are primarily used for financial transactions. For example, a communication account may communicate with a victim's account on social media. When the victim successfully makes a financial transaction to the communication account, the communication account will transfer the received funds to the receiving account.

[0113] In other words, multiple communication accounts conduct fund transactions with the same receiving account. Furthermore, fund transactions typically flow from the communication account to the receiving account. Therefore, based on this, the seed account in this embodiment of the invention can be a pre-determined communication account.

[0114] After obtaining the seed account, you can perform correlation analysis on the seed account based on the fund flow information to find at least one primary associated account that has fund transactions with the seed account.

[0115] It should be noted that, in the embodiments of the present invention, the seed account can be a pre-determined communication account, but the specific method of determining the seed account is not limited. That is to say, all methods that can determine the seed account are applicable to the embodiments of the present invention.

[0116] Specifically, based on fund flow information, correlation analysis is performed on the identified seed accounts to obtain at least one primary associated account that has fund transactions with the seed accounts, including:

[0117] Identify at least one account with fund flow information that has a connection with the seed account, and obtain at least one first associated account that has fund transactions with the seed account.

[0118] Specifically, since fund transactions typically involve funds flowing from a communication account to a receiving account, the flow of funds is relatively simple and exhibits clear transaction characteristics. Therefore, these transaction characteristics can be used to identify accounts with fund flow information related to the seed account and designate these accounts as the primary associated accounts. For example, if seed account A transferred funds to account B, then account B can be designated as the primary associated account, i.e., the receiving account.

[0119] In such Figure 3 In the correlation analysis diagram shown, each smallest dot represents a seed account, and the other dots represent first associated accounts. The size of a first associated account represents the number of associated seed accounts. For example, if first associated account B has financial transactions with seed accounts A and C, and first associated account D has financial transactions with seed accounts E, F, G, and J, then the size of D is greater than the size of B (not shown in the diagram).

[0120] The fund flow information can be detailed information about fund transactions, including but not limited to the transaction counterparty, the amount of funds, and the transaction time. Fund flow information is generated after a fund transaction is completed and can be stored on a terminal, a server, or other locations. When performing correlation analysis on seed accounts, fund flow information can be obtained first, and then correlation analysis can be performed based on that information.

[0121] Step S202: Based on the fund flow information, perform correlation analysis on at least one first associated account to obtain at least one first candidate account that has fund transactions with at least one first associated account;

[0122] In practical applications, the receiving account of the target cluster may also conduct financial transactions with other communication accounts, but there may be no financial transactions between the multiple communication accounts. For example, receiving account A may simultaneously receive transfers from communication accounts B, C, and D, but there are no financial transactions between communication accounts B, C, and D. Therefore, directly performing correlation analysis on any one of communication accounts B, C, or D will not reveal any other communication accounts. However, if correlation analysis is first performed on any one of communication accounts B, C, or D to obtain receiving account A, and then correlation analysis is performed on receiving account A, it may be possible to find other communication accounts. Therefore, after obtaining the first associated account, further correlation analysis is needed to obtain at least one first candidate account that has financial transactions with at least one first associated account.

[0123] In practical applications, the first candidate account is not necessarily the target-related account. For example, if the first associated account purchases an item on a shopping platform, then the first associated account has a financial transaction with the shopping platform's account. In other words, the shopping platform's account is the first candidate account, but clearly, it is not the target-related account.

[0124] Specifically, based on fund flow information, correlation analysis is performed on at least one primary associated account to obtain at least one primary candidate account that has fund transactions with at least one primary associated account, including:

[0125] Identify at least one account that has fund flow information with at least one first associated account, and obtain at least one first candidate account that has fund transactions with at least one first associated account.

[0126] Specifically, since fund transactions typically involve funds flowing from a communication account to a receiving account, the flow of funds is relatively simple and exhibits clear transaction characteristics. Therefore, these transaction characteristics can be used to identify accounts with fund flow information related to the receiving account and these accounts can be considered as primary candidate accounts. For example, if associated account A receives a transfer from account B, then account B can be considered as a primary candidate account. In this context, primary candidate accounts are those that are likely related to the target account.

[0127] like Figure 4 The diagram showing the correlation analysis is in Figure 3 This is a further correlation analysis based on the correlation analysis diagram shown, where the dots within the shaded area (including...) Figure 3 The seed account and the receiving account in the system are the first candidate accounts.

[0128] Step S203: Detect at least one first candidate account based on the trained graph neural network model, and identify the target related account among at least one first candidate account.

[0129] After obtaining at least one first candidate account through correlation analysis, a trained graph neural network model can be used to detect each of the at least one first candidate account, thereby determining whether there is a communication account and a payment account among the at least one first candidate account.

[0130] In one embodiment of the present invention, detecting at least one first candidate account based on a trained graph neural network model includes:

[0131] An account graph network is constructed using seed accounts and at least one first candidate account.

[0132] A graph neural network model is used to detect account graph networks.

[0133] Specifically, when using a trained graph neural network model to detect at least one first candidate account, an account graph network can be constructed first using seed accounts and at least one first candidate account. The account graph network is unstructured graph data, including at least one seed account and at least one first candidate account, as well as the relationships between each seed account and each first candidate account.

[0134] Because account graph networks are a type of unstructured graph data, and graph data has irregular or incomplete data structures and lacks a predefined data model, it is inconvenient to represent them using two-dimensional logical tables in a database, for example... Figure 3 , 4 The diagram illustrating the correlation analysis is also graph data. If represented using a two-dimensional logical table in a database, the representation would be less effective. Therefore, using traditional neural networks or machine learning methods to detect account graph networks makes it difficult to fully utilize the unstructured information in the graph network. Graph neural network models, however, can better utilize this unstructured information. Therefore, this embodiment of the invention can use a graph neural network model to detect account graph networks.

[0135] The account graph network is constructed using a seed account and at least one first candidate account, including:

[0136] For the current first candidate account, perform the following first detection step: when it is detected that the seed account and any one of the current first candidate accounts have fund flow information with the same target account, generate a connection line between the seed account and any one of the current first candidate accounts;

[0137] Take the next first candidate account encountered during the traversal as the current first candidate account, and repeat the first detection step described above until at least one first candidate account has been completely traversed and detected, resulting in at least one connection line.

[0138] Generate an account graph network based on at least one connection line.

[0139] Specifically, from at least one primary candidate account, a current primary candidate account is selected. It is then checked whether this primary candidate account and the seed account both have fund flow information with the same target account. If so, a connection is generated between the seed account and the primary candidate account. For example, ... Figure 5 As shown, both seed account A and first candidate account B have transferred funds to target account C, so a connection line between A and B can be generated.

[0140] Then, from at least one first candidate account, the next first candidate account is selected as the current first candidate account. The first detection step described above is repeated until all first candidate accounts in at least one first candidate account have been traversed and detected, resulting in at least one connection line. Then, an account graph network is generated based on at least one connection line.

[0141] For example, if there are three candidate accounts, A, B, and C, firstly, B is selected from the three candidate accounts, and the first detection step is performed on B. When it is detected that both the seed account and B have fund flow information with the same target account, a connection line is generated between the seed account and B. Then, A is traversed from A and C, A is selected, and the first detection step is performed on A. When it is detected that neither the seed account nor A has fund flow information with the same target account, C is traversed from C, C is selected, and the first detection step is performed on C. When it is detected that both the seed account and C have fund flow information with the same target account, a connection line is generated between the seed account and C. Thus, two connection lines are obtained.

[0142] In an account graph network, because a seed account and multiple first-candidate accounts may have financial transactions with the same target account, a seed account may be connected to multiple first-candidate accounts; similarly, multiple seed accounts may have financial transactions with the same target account, so a first-candidate account may also be connected to multiple seed accounts. Figure 6 As shown, the light-colored dots represent the first candidate accounts, and the dark-colored dots represent the seed accounts.

[0143] It should be noted that selecting a current first candidate account from at least one first candidate account can be done randomly or in a certain order. For example, continuing from the previous example, when selecting from A, B, and C, the selection can be done randomly or in the order of A, B, C. Of course, other methods can also be used. In practical applications, the settings can be configured according to actual needs, and this embodiment of the invention does not impose any restrictions on this.

[0144] Furthermore, the account graph network is generated using connectors because the generated account graph network is unstructured graph data. Since graph data can take many forms besides connectors, in this embodiment of the invention, all methods that can generate account graph networks are applicable, and adjustments can be made according to actual needs in practical applications.

[0145] Among them, the use of a graph neural network model to detect account graph networks includes:

[0146] Perform the following second detection step: For any first candidate account in the account graph network, identify at least one target seed account that has a connection line with any first candidate account;

[0147] A graph neural network model is used to perform forward propagation on at least one target seed account to obtain a deep feature representation of any first candidate account;

[0148] The activation function in the graph neural network model is used to predict the probability value of the deep feature representation to obtain the prediction result of any first candidate account, and the prediction result is used to determine whether any first candidate account is a target-related account.

[0149] Repeat the second detection step described above until at least one first candidate account has been completely detected.

[0150] Specifically, when using a graph neural network model to detect accounts in a graph network, one can first select any first candidate account from the account graph network and determine at least one target seed account that has a connection with any first candidate account. Then, the trained graph neural network model is used to perform forward propagation on at least one target seed account to obtain the deep feature representation of any first candidate account.

[0151] For example, after selecting the first candidate account A, target seed accounts B, C, and D that are connected to A are identified. Then, a graph neural network model is used to perform forward propagation on B, C, and D to obtain the deep feature representation of A.

[0152] When using a trained graph neural network model to perform forward propagation with at least one target seed account, the calculation can be performed using formula (1):

[0153] H (k+1) =f(H (k) ,A)=σ(AH (k) W (k) ) Formula (1)

[0154] Where H is the representation set of all target seed accounts, k is the current round of calculation (used only in the formula to represent the relationship between different rounds), A is the adjacency matrix, and W is the weight matrix. For any node in H and j that has a connection with i, (AH) i =A i H = ∑ j A ij H j A is only valid if there is a connecting line between i and j. ii Not zero.

[0155] After obtaining the deep feature representation of any first candidate account, a preset activation function can be used to predict the probability value of the deep feature representation to obtain the prediction result of any first candidate account. Based on the prediction result, it can be determined whether any first candidate account is a target-related account. The probability value is predicted using formula (2):

[0156] p = H K =sigmoid(H (K-1) ,A) Formula (2)

[0157] Where p is the probability value of the prediction result, sigmoid is the activation function, and K is the preset number of rounds. For example, if the prediction result of the first candidate account B is 0.87, then the first candidate account B is determined to be a target-related account.

[0158] Then, the next first candidate account is traversed from the account graph network, and the second detection described above is repeated until at least one first candidate account has been completely detected, thus obtaining the final target-related account.

[0159] exist Figure 7 The diagram showing fund flows illustrates the fund flows between different types of accounts. The fund flows of the seed account and the detected communication account mainly point to the first associated account (i.e., the receiving account that transacts with the seed account), and the fund flows of the first associated account mainly point to the detected receiving account.

[0160] In practical applications, this invention was used to perform correlation analysis on approximately 700 seed accounts, identifying over 200 receiving accounts and nearly 3,500 confirmed accounts. Without any feedback, this invention uncovered five times the number of confirmed accounts, involving transactions totaling nearly 2 million RMB.

[0161] In this embodiment of the invention, correlation analysis is performed on the identified seed accounts based on fund flow information to obtain at least one first associated account that has fund transactions with the seed accounts. Then, correlation analysis is performed on each of the at least one first associated account based on fund flow information to obtain at least one first candidate account that has fund transactions with the at least one first associated account. Finally, a trained graph neural network model is used to detect the at least one first candidate account to determine the target related accounts among the at least one first candidate account. In this way, deep mining can be carried out using the identified seed accounts. Even target related accounts without feedback information can be mined through the above method, thereby solving the problem of not being able to accurately discover accounts in the target cluster. At the same time, it also achieves the purpose of early detection and discovery of target related accounts.

[0162] Furthermore, the introduction of unstructured graph networks and graph neural network models during the correlation analysis process greatly improves the accuracy and coverage of the correlation analysis. At the same time, the correlation analysis process does not require manual rule formulation, which greatly increases the scalability and ease of use of the correlation analysis.

[0163] This invention provides a training method for a trained graph neural network model in an account detection method, such as... Figure 8 As shown, the method includes:

[0164] Step S801: Perform correlation analysis on the sample seed account to obtain at least one second related account that has financial transactions with the sample seed account.

[0165] Step S802: Based on the fund flow information, perform correlation analysis on at least one second associated account to obtain at least one second candidate account that has fund transactions with at least one second associated account;

[0166] Step S803: Use a preset original graph neural network model to detect any second candidate account among at least one second candidate account, and obtain the detection result of any second candidate account;

[0167] Steps S801 to S803 are essentially the same as steps S201 to S203, and will not be repeated here to avoid repetition.

[0168] Step S804: Based on the detection results and labels of any second candidate account, calculate the forward propagation error using the loss function in the original graph neural network model to obtain the error value;

[0169] Step S805: Backpropagation is performed using the error value to update the weight matrix of the original graph neural network model, resulting in the updated graph neural network model.

[0170] Step S806: Use the updated graph neural network model as the current original graph neural network, and repeat steps S801 to S805 until the loss function converges to its minimum value, thus obtaining the trained graph neural network model.

[0171] Specifically, when training the original graph neural network model, the sample accounts include not only the seed sample accounts, but also the candidate sample accounts and non-target related accounts (i.e., ordinary accounts). Each sample account has a preset label. After step S803, the forward propagation error can be calculated using a loss function based on the detection result and label of any detected second candidate account to obtain the error value. Then, the error value is used for backpropagation to update the weight matrix W of the original graph neural network model, thereby obtaining the updated graph neural network model.

[0172] The forward propagation error can be calculated using formula (3):

[0173]

[0174] Where ce is the loss function, y i For the default tags of account i, p i Let be the predicted probability value for account i.

[0175] Then, the updated graph neural network model is used as the current original graph neural network, and the above detection process is repeated until the loss function converges to its minimum value, thus obtaining the trained graph neural network model.

[0176] For ease of understanding, the complete steps for training the original graph neural network model are explained below.

[0177] 1) Conduct correlation analysis on the sample seed accounts to identify at least one second related account with which the sample seed account has financial transactions;

[0178] 2) Based on the fund flow information, perform correlation analysis on at least one second related account to obtain at least one second candidate account that has fund transactions with at least one second related account;

[0179] 3) An account graph network is constructed using sample seed accounts and at least one second candidate account;

[0180] 4) When it is detected that the seed account and any of the at least one second candidate accounts have fund flow information with the same target account, a connection line is generated between the seed account and any of the second candidate accounts;

[0181] 5) Repeat step 4) with the next second candidate account encountered during the traversal as the current second candidate account, until at least one second candidate account has been completely traversed and detected, and at least one connection line is obtained;

[0182] 6) Generate an account graph network based on at least one connection line;

[0183] 7) For any second candidate account in the account graph network, identify at least one target sample seed account that has a connection line with any second candidate account;

[0184] 8) Use the original graph neural network model to perform forward propagation on at least one target sample seed account to obtain the deep feature representation of any second candidate account;

[0185] 9) The activation function in the original graph neural network is used to predict the probability value of the deep feature representation to obtain the prediction result of any second candidate account;

[0186] 10) Based on the prediction results and labels of any second candidate account, calculate the forward propagation error using the loss function in the original graph neural network model to obtain the error value;

[0187] 11) Backpropagation is performed using the error values ​​to update the weight matrix of the original graph neural network model, resulting in the updated graph neural network model;

[0188] 12) Use the updated graph neural network model as the current original graph neural network, and repeat steps 1) to 11) until the loss function converges to the minimum value, thus obtaining the trained graph neural network model.

[0189] In this embodiment of the invention, unstructured graph networks and graph neural network models are introduced into the correlation analysis process, which greatly improves the accuracy and coverage of the correlation analysis. At the same time, no manual rule formulation is required in the correlation analysis process, which greatly increases the scalability and ease of use of the correlation analysis.

[0190] Figure 9 This is a schematic diagram of the structure of an account detection device provided in an embodiment of this application, as shown below. Figure 9 As shown, the apparatus of this embodiment may include:

[0191] The first mining module 901 is used to perform correlation analysis on the identified seed accounts based on fund flow information to obtain at least one first associated account that has fund transactions with the seed account.

[0192] The second mining module 902 is used to perform correlation analysis on at least one first associated account based on fund flow information, and obtain at least one first candidate account that has fund transactions with at least one first associated account.

[0193] The detection module 903 is used to detect at least one first candidate account based on the trained graph neural network model, and to identify the target related account among at least one first candidate account.

[0194] In this embodiment of the invention, the detection module includes:

[0195] A submodule is constructed to build an account graph network using a seed account and at least one first candidate account.

[0196] The detection submodule is used to detect account graph networks using a graph neural network model.

[0197] In this embodiment of the invention, the construction of the sub-module includes:

[0198] The connection line generation unit is used to generate a connection line between the seed account and any of the current first candidate accounts when it is detected that the seed account and any of the current first candidate accounts have fund flow information with the same target account.

[0199] The next first candidate account encountered during the traversal is used as the current first candidate account. The connection line generation unit is called repeatedly until at least one first candidate account has been completely traversed and detected, and at least one connection line is obtained.

[0200] The account graph network generation unit is used to generate an account graph network based on at least one connection line.

[0201] In this embodiment of the invention, the first mining module is specifically used for:

[0202] Identify at least one account with fund flow information that has a connection with the seed account, and obtain at least one first associated account that has fund transactions with the seed account;

[0203] The second mining module is specifically used for:

[0204] Identify at least one account that has fund flow information with at least one first associated account, and obtain at least one first candidate account that has fund transactions with at least one first associated account.

[0205] In this embodiment of the invention, the detection submodule includes:

[0206] The first processing unit is used to determine, for any first candidate account in the account graph network, at least one target seed account that has a connection line with any first candidate account.

[0207] The second processing unit is used to perform forward propagation on at least one target seed account using a graph neural network model to obtain a deep feature representation of any first candidate account.

[0208] The third processing unit is used to predict the probability value of the deep feature representation by using the activation function in the graph neural network model, to obtain the prediction result of any first candidate account, and to determine whether any first candidate account is a target-related account based on the prediction result.

[0209] Repeatedly call the first processing unit, the second processing unit, and the third processing unit until at least one first candidate account has been completely detected.

[0210] In this embodiment of the invention, it further includes:

[0211] The first mining module is also used to perform correlation analysis on the sample seed accounts to obtain at least one second related account that has financial transactions with the sample seed account.

[0212] The second mining module is also used to perform correlation analysis on at least one first associated account based on fund flow information to obtain at least one second candidate account that has fund transactions with at least one second associated account.

[0213] The detection module is also used to detect any one of the at least one second candidate accounts using a preset original graph neural network model, and to obtain the detection result of any one of the second candidate accounts.

[0214] The calculation module is used to calculate the forward propagation error based on the detection result and label of any second candidate account, using the loss function in the original graph neural network model, and obtain the error value.

[0215] The update module is used to backpropagate the error values ​​and update the weight matrix of the original graph neural network model to obtain the updated graph neural network model.

[0216] The updated graph neural network model is used as the current original graph neural network. The first mining module, the second mining module, the detection module, the calculation module, and the update module are repeatedly called until the minimum value of the loss function converges, thus obtaining the trained graph neural network model.

[0217] The account detection device in this embodiment can execute the account detection method shown in the foregoing embodiments of this application. The implementation principle is similar and will not be described again here.

[0218] In this embodiment of the invention, correlation analysis is performed on the identified seed accounts based on fund flow information to obtain at least one first associated account that has fund transactions with the seed accounts. Then, correlation analysis is performed on each of the at least one first associated account based on fund flow information to obtain at least one first candidate account that has fund transactions with the at least one first associated account. Finally, a trained graph neural network model is used to detect the at least one first candidate account to determine the target related accounts among the at least one first candidate account. In this way, deep mining can be carried out using the identified seed accounts. Even target related accounts without feedback information can be mined through the above method, thereby solving the problem of not being able to accurately discover accounts in the target cluster. At the same time, it also achieves the purpose of early detection and discovery of target related accounts.

[0219] Furthermore, the introduction of unstructured graph networks and graph neural network models during the correlation analysis process greatly improves the accuracy and coverage of the correlation analysis. At the same time, the correlation analysis process does not require manual rule formulation, which greatly increases the scalability and ease of use of the correlation analysis.

[0220] This application provides an electronic device comprising: a memory and a processor; at least one program stored in the memory, which, when executed by the processor, can, compared with the prior art, perform: a correlation analysis on a determined seed account based on fund flow information to obtain at least one first associated account with fund transactions with the seed account; then, perform correlation analysis on each of the at least one first associated account based on fund flow information to obtain at least one first candidate account with fund transactions with the at least one first associated account; and finally, detect the at least one first candidate account based on a trained graph neural network model to determine the target related account among the at least one first candidate account. In this way, deep mining can be performed using the determined seed account, and even target related accounts without feedback information can be mined using the above method, thereby solving the problem of not being able to accurately discover accounts in the target cluster, and also achieving the purpose of early detection and discovery of target related accounts.

[0221] In one alternative embodiment, an electronic device is provided, such as Figure 10 As shown, Figure 10The illustrated electronic device 10000 includes a processor 10001 and a memory 10003. The processor 10001 and the memory 10003 are connected, for example, via a bus 10002. Optionally, the electronic device 10000 may further include a transceiver 10004, which can be used for data interaction between the electronic device and other electronic devices, such as sending and / or receiving data. It should be noted that in practical applications, the transceiver 10004 is not limited to one type, and the structure of the electronic device 10000 does not constitute a limitation on the embodiments of this application.

[0222] Processor 10001 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. Processor 10001 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0223] Bus 10002 may include a pathway for transmitting information between the aforementioned components. Bus 10002 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. Bus 10002 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 10 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0224] The memory 10003 may be ROM (Read Only Memory) or other types of static storage devices capable of storing static information and instructions, RAM (Random Access Memory) or other types of dynamic storage devices capable of storing information and instructions, or EEPROM (Electrically Erasable Programmable Read Only Memory), CD-ROM (Compact Disc Read Only Memory) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto.

[0225] The memory 10003 is used to store computer program code (computer program) for executing the scheme of this application, and its execution is controlled by the processor 10001. The processor 10001 is used to execute the computer program code stored in the memory 10003 to implement the content shown in the foregoing method embodiments.

[0226] Among them, electronic devices include, but are not limited to: mobile terminals such as mobile phones, laptops, digital radio receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), and in-vehicle terminals (such as in-vehicle navigation terminals), as well as fixed terminals such as digital TVs and desktop computers.

[0227] This application provides a computer-readable storage medium storing a computer program that, when run on a computer, enables the computer to execute the corresponding content in the aforementioned method embodiments.

[0228] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0229] The above description is only a partial embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. An account detection method, characterized in that, include: Based on the fund flow information, a correlation analysis is performed on the identified seed accounts to obtain at least one first associated account that has fund transactions with the seed accounts; wherein, the seed account is a communication account, the first associated account is a receiving account, and the seed account is a pre-determined known target account; Based on the fund flow information, a correlation analysis is performed on the at least one first associated account to obtain at least one first candidate account that has fund transactions with the at least one first associated account, wherein each first candidate account is a communication account or a receiving account. For each first candidate account and each seed account, if both the first candidate account and the seed account have fund flow information with the same target account, then a connection line is generated between the first candidate account and the seed account to construct an account graph network. The trained graph neural network model is invoked, and the graph neural network model detects each first candidate account in the account graph network based on the account graph network to determine the target related account among the at least one first candidate account.

2. The account detection method according to claim 1, characterized in that, The correlation analysis based on fund flow information of the identified seed accounts yields at least one first associated account that has fund transactions with the seed accounts, including: Identify at least one account that has fund flow information with the seed account, and obtain at least one first associated account that has fund transactions with the seed account; The correlation analysis is performed on the at least one first associated account based on the fund flow information to obtain at least one first candidate account that has fund transactions with the at least one first associated account, including: Identify at least one account that has fund flow information with at least one first associated account, and obtain at least one first candidate account that has fund transactions with the at least one first associated account.

3. The account detection method according to claim 1, characterized in that, The step of detecting each first candidate account in the account graph network by the graph neural network model based on the account graph network includes: Perform the following second detection step: For any first candidate account in the account graph network, identify at least one target seed account that has a connection line with any first candidate account; The graph neural network model is used to perform forward propagation on the at least one target seed account to obtain the deep feature representation of any first candidate account; The activation function in the graph neural network model is used to predict the probability value of the deep feature representation to obtain the prediction result of any first candidate account, and based on the prediction result, it is determined whether any first candidate account is a target-related account; Repeat the second detection step described above until all at least one first candidate account has been detected.

4. The account detection method according to claim 1, characterized in that, The trained graph neural network model is obtained through the following method: Perform the following training steps: conduct correlation analysis on the sample seed account to obtain at least one second associated account that has financial transactions with the sample seed account; Based on the fund flow information, a correlation analysis is performed on the at least one second associated account to obtain at least one second candidate account that has fund transactions with the at least one second associated account; A preset original graph neural network model is used to detect any one of the at least one second candidate accounts, and the detection result of any one of the second candidate accounts is obtained. Based on the detection results and tags of any of the second candidate accounts, the forward propagation error is calculated using the loss function in the original graph neural network model to obtain the error value; The error value is used for backpropagation to update the weight matrix of the original graph neural network model, resulting in an updated graph neural network model. The updated graph neural network model is used as the current original graph neural network, and the above training steps are repeated until the loss function converges to obtain the trained graph neural network model.

5. An account detection device, characterized in that, include: The first mining module is used to perform correlation analysis on the identified seed accounts based on fund flow information to obtain at least one first associated account that has fund transactions with the seed account; wherein, the seed account is a communication account, the first associated account is a receiving account, and the seed account is a pre-determined known target account. The second mining module is used to perform correlation analysis on the at least one first associated account based on the fund flow information, and obtain at least one first candidate account that has fund transactions with the at least one first associated account, wherein each first candidate account is a communication account or a receiving account. A detection module is used to construct an account graph network and call a trained graph neural network model. The graph neural network model detects each first candidate account in the account graph network based on the account graph network to determine the target related account among the at least one first candidate account. The account graph network is constructed in the following manner: For each first candidate account and each seed account, if both the first candidate account and the seed account have fund flow information with the same target account, then a connection line is generated between the first candidate account and the seed account.

6. The apparatus according to claim 5, characterized in that, The first mining module is specifically used to: identify at least one account that has fund flow information with the seed account, and obtain at least one first associated account that has fund transactions with the seed account; The second mining module is specifically used to: identify at least one account that has fund flow information with at least one first associated account, and obtain at least one first candidate account that has fund transactions with the at least one first associated account.

7. The apparatus according to claim 5, characterized in that, The detection module, when detecting each first candidate account in the account graph network by calling the graph neural network model, is specifically used for: Perform the following second detection step: For any first candidate account in the account graph network, identify at least one target seed account that has a connection line with any first candidate account; The graph neural network model is used to perform forward propagation on the at least one target seed account to obtain the deep feature representation of any first candidate account; The activation function in the graph neural network model is used to predict the probability value of the deep feature representation to obtain the prediction result of any first candidate account, and based on the prediction result, it is determined whether any first candidate account is a target-related account; Repeat the second detection step described above until all at least one first candidate account has been detected.

8. The apparatus according to claim 5, characterized in that, The trained graph neural network model is obtained through the following method: Perform the following training steps: conduct correlation analysis on the sample seed account to obtain at least one second associated account that has financial transactions with the sample seed account; Based on the fund flow information, a correlation analysis is performed on the at least one second associated account to obtain at least one second candidate account that has fund transactions with the at least one second associated account; A preset original graph neural network model is used to detect any one of the at least one second candidate accounts, and the detection result of any one of the second candidate accounts is obtained. Based on the detection results and tags of any of the second candidate accounts, the forward propagation error is calculated using the loss function in the original graph neural network model to obtain the error value; The error value is used for backpropagation to update the weight matrix of the original graph neural network model, resulting in an updated graph neural network model. The updated graph neural network model is used as the current original graph neural network, and the above training steps are repeated until the loss function converges to obtain the trained graph neural network model.

9. An electronic device, characterized in that, The electronic device includes: One or more processors; Memory; One or more computer programs, wherein the one or more computer programs are stored in the memory and configured to be executed by the one or more processors, the one or more computer programs being configured to: perform the account detection method according to any one of claims 1 to 4.

10. A computer-readable storage medium, characterized in that, The storage medium stores at least one instruction, at least one program, code set, or instruction set, wherein the at least one instruction, the at least one program, the code set, or instruction set is loaded and executed by a processor to implement the account detection method as described in any one of claims 1 to 4.

Images