Method, device and storage medium for identifying potential gangs performing fraudulent activities

By classifying and denoising database samples to generate high-quality samples, a knowledge graph is constructed. Community detection algorithms are then used to identify potential gangs, solving the problems of inconsistent sample quality and poor saliency of node relationships in existing technologies, and improving the efficiency of identifying fraud gangs.

CN118626806BActive Publication Date: 2024-11-08BAO DAFANG TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411104230.2
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-13
Publication Date
2024-11-08
Estimated Expiration
2044-08-13

AI Technical Summary

Technical Problem

In existing technologies, the inconsistent quality of samples in the database and the poor saliency between nodes result in poor effectiveness of directly using community detection algorithms to identify potential groups engaging in fraudulent activities.

Method used

The database samples are classified and denoised using a pre-trained recognition model to generate high-quality second black samples and second white samples. A knowledge graph is constructed, and a community detection algorithm is used to identify potential gangs.

Benefits of technology

It enhances the saliency of each node in the knowledge graph and improves the efficiency of the community discovery algorithm in identifying potential groups engaging in fraudulent activities.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118626806B_ABST
    Figure CN118626806B_ABST
Patent Text Reader

Abstract

The present application discloses a method, device and storage medium for identifying potential gangs that perform fraudulent behavior, including: obtaining multiple samples from a database, and dividing the multiple samples into a first sample set consisting of multiple first black samples and a second sample set consisting of multiple first white samples; determining multiple second black samples in the first sample set based on a pre-trained first recognition model; determining multiple second white samples in the second sample set based on a pre-trained second recognition model; constructing a knowledge graph based on multiple second black samples, first relationship media between each second black sample, multiple second white samples and second relationship media between each second white sample; and using a community discovery algorithm to identify potential gangs that perform fraudulent behavior in the knowledge graph. The technical effect of improving the significance of each node in the knowledge graph and ensuring the efficiency of using a community discovery algorithm to identify potential gangs that perform fraudulent behavior.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a method, device and storage medium for identifying potential gangs that perform fraudulent activities. Background Art

[0002] With the advancement of technology, the means of fraud in the black and gray industries are constantly upgrading and improving. From the initial simple means to today's high-tech violations, the clues and information left by the black and gray industries are getting fewer and fewer. As a result, the effect of simple detection methods for risky users has been minimal.

[0003] In order to solve the above problems, existing graph mining algorithms have been proposed. For example, if users want to avoid complex heterogeneous networks, they not only need to replace equipment or purchase the latest simulator technology, but also need to purchase a new payment account. This increases the fraud cost of the black and gray industries, thereby reducing the probability of fraud risk.

[0004] In some business scenarios, fraud incidents have obvious clustering patterns. That is, a certain group structure is formed between nodes, rather than long-term random dispersion. Thus, this group structure with obvious clustering patterns forms a community structure. Among them, the community structure is defined as a group of nodes, and the relationship between each node in the community structure is closer than other parts of the community network.

[0005] Therefore, it can be imagined that there are multiple different community structures in the massive data. These community structures include both low-risk community structures and high-risk community structures. Among them, the group fraud risk of low-risk community structures is lower, and the group fraud risk of high-risk community structures is higher.

[0006] When using graph mining to identify high-risk users in massive data, in order to achieve the best mining effect, all users in the database are usually mined, and all users are divided into communities through various relationship media. After the group with higher fraud risk is found, the users in the group are removed. However, in most cases, due to the uneven quality of each sample in the database and the poor significance of the relationship between each node in the knowledge graph, the direct use of community discovery algorithms to identify potential groups that perform fraudulent behavior is ineffective.

[0007] The publication number is CN117787993A, and the name is a financial anti-fraud method and device based on layer correlation propagation. It is mainly divided into two main steps. The first is to capture and integrate the different local structural features of the graph through feature decoupling and aggregation to separate local features and provide a powerful feature representation for graph classification. Then, in order to reduce the interference of redundant features on classification performance, the generated low-correlation features are suppressed in the process of layer correlation propagation to reduce the impact of redundant features on model decisions.

[0008] The publication number is CN116843432A, and the name is an anti-fraud method and device based on address text information. It includes: extracting features from the address text information to obtain a text feature vector for each character of the address text information, dividing the address text information into at least one address element with independent semantics corresponding to the address text information, and correcting and completing each address element with independent semantics obtained by the division to obtain a pre-processed address element corresponding to the address element with independent semantics; and performing anti-fraud identification on the address text information according to the pre-processed address element and / or text feature vector.

[0009] With respect to the technical problem existing in the above-mentioned prior art that the quality of each sample in the database is uneven and the significance of the relationship between each node in constructing the knowledge graph is poor, directly using the community discovery algorithm to identify potential gangs that perform fraudulent activities is poor, no effective solution has been proposed so far. Summary of the invention

[0010] The embodiments of the present disclosure provide a method, device and storage medium for identifying potential groups that perform fraudulent activities, so as to at least solve the technical problem in the prior art that the quality of each node in the database is uneven and the significance of the relationship between each node is poor, so the effect of directly using the community discovery algorithm to identify potential groups that perform fraudulent activities is poor.

[0011] According to one aspect of an embodiment of the present disclosure, a method for identifying potential gangs that perform fraudulent behavior is provided, including: obtaining multiple samples from a database, and dividing the multiple samples into a first sample set consisting of multiple first black samples and a second sample set consisting of multiple first white samples, wherein the first black samples are used to indicate data samples that perform fraudulent behavior in a specified business scenario, and the first white samples are used to indicate data samples that perform safe operation behavior in a specified business scenario; determining multiple second black samples in the first sample set based on a pre-trained first recognition model, wherein the similarity between the behavior pattern of the second black sample and the behavior pattern of the first black sample is greater than a first preset threshold; determining multiple second white samples in the second sample set based on a pre-trained second recognition model, wherein the similarity between the behavior pattern of the second white sample and the behavior pattern of the first white sample is less than a second preset threshold; constructing a knowledge graph based on multiple second black samples, first relationship media between each second black sample, multiple second white samples, and second relationship media between each second white sample; and identifying potential gangs that perform fraudulent behavior in the knowledge graph using a community discovery algorithm.

[0012] According to another aspect of an embodiment of the present disclosure, a storage medium is further provided, the storage medium including a stored program, wherein when the program is running, a processor executes any one of the methods described above.

[0013] According to another aspect of an embodiment of the present disclosure, there is also provided a device for identifying a potential gang that performs fraudulent behavior, including: a first sample set construction module, used to obtain multiple samples from a database, and divide the multiple samples into a first sample set consisting of multiple first black samples and a second sample set consisting of multiple first white samples, wherein the first black samples are used to indicate data samples that perform fraudulent behavior in a specified business scenario, and the first white samples are used to indicate data samples that perform safe operation behavior in a specified business scenario; a second black sample determination module, used to determine multiple second black samples in the first sample set based on a pre-trained first recognition model, wherein the similarity between the behavior pattern of the second black sample and the behavior pattern of the first black sample is greater than a first preset threshold; a second white sample determination module, used to determine multiple second white samples in the second sample set based on a pre-trained second recognition model, wherein the similarity between the behavior pattern of the second white sample and the behavior pattern of the first white sample is less than a second preset threshold; a knowledge graph construction module, used to construct a knowledge graph based on multiple second black samples, first relationship media between each second black sample, multiple second white samples and second relationship media between each second white sample; and a potential gang identification module, used to identify potential gangs that perform fraudulent behavior in the knowledge graph using a community discovery algorithm.

[0014] According to another aspect of an embodiment of the present disclosure, there is also provided a device for identifying potential gangs that perform fraudulent behavior, including: a processor; and a memory connected to the processor, for providing the processor with instructions for processing the following processing steps: obtaining multiple samples from a database, and dividing the multiple samples into a first sample set consisting of multiple first black samples and a second sample set consisting of multiple first white samples, wherein the first black samples are used to indicate data samples that perform fraudulent behavior in a specified business scenario, and the first white samples are used to indicate data samples that perform safe operation behavior in a specified business scenario; determining multiple second black samples in the first sample set based on a pre-trained first recognition model, wherein the similarity between the behavior pattern of the second black sample and the behavior pattern of the first black sample is greater than a first preset threshold; determining multiple second white samples in the second sample set based on a pre-trained second recognition model, wherein the similarity between the behavior pattern of the second white sample and the behavior pattern of the first white sample is less than a second preset threshold; constructing a knowledge graph based on multiple second black samples, first relationship media between each second black sample, multiple second white samples, and second relationship media between each second white sample; and identifying potential gangs that perform fraudulent behavior in the knowledge graph using a community discovery algorithm.

[0015] The present application discloses a method for identifying potential gangs that perform fraudulent activities. First, a processor obtains multiple samples from a database and divides the multiple samples into a first sample set consisting of multiple first black samples and a second sample set consisting of multiple first white samples. Then, the processor determines multiple second black samples in the first sample set based on a pre-trained first recognition model. At the same time, the processor determines multiple second white samples in the second sample set based on a pre-trained second recognition model. Further, the processor constructs a knowledge graph based on multiple second black samples, first relationship media between each second black sample, multiple second white samples, and second relationship media between each second white sample. Finally, the processor uses a community discovery algorithm to identify potential gangs that perform fraudulent activities in the knowledge graph.

[0016] With reference to the above contents, the technical solution of the present application pre-classifies the samples in the database (i.e., divides the multiple samples in the database into a first black sample and a first white sample), and determines the second black sample in the first sample set based on the pre-trained first recognition model (i.e., the second black sample is the black sample remaining after the first black sample in the first sample set is subjected to noise reduction processing), and determines the second white sample in the second sample set based on the pre-trained second recognition model (i.e., the second white sample is the white sample remaining after the first white sample in the second sample set is subjected to noise reduction processing).

[0017] Therefore, unlike the existing method of directly using graph mining algorithms to mine all samples in the database (i.e., the first black sample and the first white sample), the technical solution of the present application pre-noises the samples in the database (i.e., the first black sample and the first white sample), so that the remaining samples (i.e., the second black sample and the second white sample) are more representative. Then, the remaining samples (i.e., the second black sample and the second white sample) and the relationship media between the samples (i.e., the first relationship media and the second relationship media) are used to construct a knowledge graph. Finally, a community discovery algorithm is used to identify potential gangs that perform fraudulent activities in the knowledge graph.

[0018] This can achieve the technical effect of improving the significance of each node in the knowledge graph and ensuring the efficiency of using the community discovery algorithm to identify potential groups that perform fraudulent behavior. This solves the technical problem in the prior art that the quality of each sample in the database is uneven and the significance of the relationship between each node is poor, so the direct use of the community discovery algorithm to identify potential groups that perform fraudulent behavior is poor. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The drawings described herein are used to provide a further understanding of the present disclosure and constitute a part of the present application. The illustrative embodiments of the present disclosure and their descriptions are used to explain the present disclosure and do not constitute an improper limitation on the present disclosure. In the drawings:

[0020] Figure 1 is a hardware structure block diagram of a computing device for implementing the method according to Embodiment 1 of the present disclosure;

[0021] Figure 2 is a schematic diagram of a potential gang identification system for fraudulent behavior according to Embodiment 1 of the present disclosure;

[0022] Figure 3 It is a flowchart of a method for identifying potential gangs for fraudulent behavior according to Embodiment 1 of the present disclosure;

[0023] Figure 4 It is a schematic diagram of the overall method flow for determining a potential gang according to Embodiment 1 of the present disclosure;

[0024] Figure 5 is a flow chart of a method for reducing noise on a first black sample according to Embodiment 1 of the present disclosure;

[0025] Figure 6 is a flowchart of a method for reducing noise on a first white sample according to Embodiment 1 of the present disclosure;

[0026] Figure 7is a flow chart of a method for determining a first relationship medium between respective second black samples according to Embodiment 1 of the present disclosure;

[0027] Figure 8 is a schematic flow chart of a method for determining a second relationship medium between respective second white samples according to Embodiment 1 of the present disclosure;

[0028] Fig. 9 is a schematic diagram of a device for identifying a potential group that performs fraudulent behavior according to Embodiment 2 of the present disclosure; and

[0029] Fig.10 It is a schematic diagram of a device for identifying potential gangs that perform fraudulent activities according to Embodiment 3 of the present disclosure. DETAILED DESCRIPTION

[0030] In order to enable those skilled in the art to better understand the technical solutions of the present disclosure, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are only embodiments of a part of the present disclosure, not all of the embodiments. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in this field without making creative work should fall within the scope of protection of the present disclosure.

[0031] It should be noted that the terms "first", "second", etc. in the specification and claims of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products, or devices. Example 1

[0032] According to this embodiment, a method embodiment for identifying potential gangs that perform fraudulent activities is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0033] The method embodiment provided in this embodiment can be executed in a mobile terminal, a computer terminal, a server or a similar computing device. Figure 1FIG. 1 shows a hardware structure block diagram of a computing device for implementing a method for identifying a potential group that performs fraudulent behavior. Figure 1 As shown, the computing device may include one or more processors (the processor may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory for storing data, a transmission device for communication functions, and an input / output interface. The memory, the transmission device, and the input / output interface are connected to the processor via a bus. In addition, it may also include: a display, a keyboard, and a cursor control device connected to the input / output interface. A person skilled in the art can understand that Figure 1 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations are shown.

[0034] It should be noted that the one or more processors and / or other data processing circuits described above may generally be referred to herein as "data processing circuits". The data processing circuits may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. In addition, the data processing circuit may be a single independent processing module, or may be incorporated in whole or in part into any of the other components in the computing device. As involved in the embodiments of the present disclosure, the data processing circuit acts as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0035] The memory can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the method for identifying potential gangs that perform fraudulent behavior in the embodiment of the present disclosure. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, that is, the method for identifying potential gangs that perform fraudulent behavior of the above-mentioned application program is realized. The memory may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include a memory remotely arranged relative to the processor, and these remote memories may be connected to the computing device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0036] The transmission device is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the computing device. In one example, the transmission device includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0037] The display may be, for example, a touch screen liquid crystal display (LCD) that may enable a user to interact with a user interface of the computing device.

[0038] It should be noted that, in some optional embodiments, the above Figure 1 The computing device shown may include hardware elements (including circuits), software elements (including computer code stored on a computer-readable medium), or a combination of both hardware elements and software elements. Figure 1 This is merely one example of a particular embodiment and is intended to illustrate the types of components that may be present in the computing devices described above.

[0039] Figure 2 is a schematic diagram of a system for identifying potential fraud gangs according to this embodiment. Figure 2 As shown, the system includes: a terminal device 100 and a processor 200 that is communicatively connected to the terminal device 100. The terminal device 100 is used to send a potential group identification request for performing fraudulent behavior to the processor 200. The processor 200 receives the potential group identification request for performing fraudulent behavior sent by the terminal device 100, and divides multiple samples in the database respectively, thereby generating a first sample set consisting of a first black sample and a second sample set consisting of a first white sample. After that, the processor 200 performs noise reduction and enhancement processing on the first black sample in the first sample set and the first white sample in the second sample set, and generates a second black sample and a second white sample. Finally, the processor 200 constructs a knowledge graph based on the second black sample, the first relationship medium between each second black sample, the second white sample, and the second relationship medium between each second white sample, and uses a community discovery algorithm to identify the potential group performing fraudulent behavior in the knowledge graph.

[0040] It should be noted that the terminal device 100 and the processor 200 in the system can both be applicable to the hardware structure described above.

[0041] In the above operating environment, according to the first aspect of this embodiment, a method for identifying a potential gang that performs fraudulent behavior is provided. The method comprises: Figure 2The processor 200 shown in FIG. Figure 3 A schematic diagram showing the process of the method is shown in FIG. Figure 3 As shown, the method includes:

[0042] S302: Acquire multiple samples from a database, and divide the multiple samples into a first sample set consisting of multiple first black samples and a second sample set consisting of multiple first white samples, wherein the first black samples are used to indicate data samples that perform fraudulent behaviors in a specified business scenario, and the first white samples are used to indicate data samples that perform security operation behaviors in the specified business scenario;

[0043] S304: determining a plurality of second black samples in the first sample set based on the pre-trained first recognition model, wherein a similarity between a behavior pattern of the second black sample and a behavior pattern of the first black sample is greater than a first preset threshold;

[0044] S306: determining a plurality of second white samples in the second sample set based on the pre-trained second recognition model, wherein the similarity between the behavior pattern of the second white sample and the behavior pattern of the first white sample is less than a second preset threshold;

[0045] S308: constructing a knowledge graph based on the plurality of second black samples, the first relationship medium between the second black samples, the plurality of second white samples, and the second relationship medium between the second white samples; and

[0046] S310: Use community discovery algorithms to identify potential groups that perform fraudulent activities in knowledge graphs.

[0047] Specifically, Figure 4 is a flow chart of the overall method for determining potential gangs according to the embodiment of the present application. Figure 4 As shown, first, the processor 200 obtains multiple samples from the database, and divides the multiple samples into a first sample set consisting of multiple first black samples and a second sample set consisting of multiple first white samples (S302). Since the database stores, for example, the full amount of samples of the latest quarter of the insurance industry, the processor 200 can divide the full amount of samples into multiple first black samples and multiple first white samples, and generate a first sample set based on the multiple first black samples, and generate a second sample set based on the multiple first white samples. Among them, the first black sample is used to indicate a data sample that performs a fraudulent behavior in a specified business scenario, and the first white sample is used to indicate a data sample that performs a safe operation behavior in a specified business scenario. And in this embodiment, the specified business scenario can be, for example, a business scenario included in the insurance industry, which will not be repeated here.

[0048] Then, the processor 200 determines a plurality of second black samples in the first sample set based on the pre-trained first recognition model (S304). Specifically, first, the processor 200 extracts a plurality of third black samples from a plurality of first black samples in the first sample set. Then, the pre-constructed first recognition model is trained using the plurality of third black samples. Finally, the first black samples in the first sample set are identified using the trained first recognition model, thereby determining a plurality of second black samples. Among them, the similarity between the behavior pattern of the second black sample and the behavior pattern of the first black sample is greater than the first preset threshold. That is, the second black sample can be marked as a seed user with a higher risk of fraud.

[0049] Therefore, it can be seen from the above that the present application first extracts a significant third black sample from multiple first black samples, and uses the third black sample to train the first recognition model. Among them, the third black sample can be, for example, a useful feature extracted from multiple first black samples, and these features should be able to represent the key behavior patterns of the black samples. Therefore, the first recognition model generated after the final training can better recognize the second black sample whose behavior pattern is highly similar to the first black sample. The above content will be described in detail later, so it will not be repeated here.

[0050] At the same time, the processor 200 determines the second white sample in the second sample set based on the pre-trained second recognition model (S306). Specifically, first, the processor 200 removes the abnormal white samples in the second sample set and constructs a fourth sample set. The abnormal white samples may be, for example, white samples that do not belong to certain business attributes in a specified business scenario but are uniformly filled as white samples under the business attribute category. Then, the processor 200 removes the hot white samples in the fourth sample set and generates a fifth sample set containing multiple third white samples. The hot white samples may be, for example, white samples with a large number of relationship media. Further, the processor 200 constructs a second recognition model and trains the second recognition model using multiple third white samples in the fifth sample set. Finally, the processor 200 uses the trained second recognition model to identify each first white sample in the second sample set and determines the second white sample. The similarity between the behavior pattern of the second white sample and the behavior pattern of the first white sample is less than the second preset threshold.

[0051] Therefore, it can be seen from the above that the present application first removes invalid popular white samples and abnormal white samples from multiple first white samples, and uses the generated third white samples to train the second recognition model. Thus, the second recognition model generated by the training is finally able to identify valid second white samples from multiple first white samples.

[0052] Afterwards, the processor 200 constructs a knowledge graph based on the plurality of second black samples, the first relationship medium between each second black sample, the plurality of second white samples, and the second relationship medium between each second white sample (S308). The first relationship medium is used to indicate the association information between each second black sample, and the second relationship medium is used to indicate the association information between each second white sample. Specifically, first, the processor 200 determines the first relationship medium between each second black sample and the second relationship medium between each second white sample, respectively. Then, the processor 200 constructs a knowledge graph based on the plurality of second black samples, the first relationship medium between each second black sample, the plurality of second white samples, and the second relationship medium between each second white sample.

[0053] Finally, the processor 200 uses a community discovery algorithm to identify potential groups that perform fraudulent behavior in the knowledge graph (S310). Specifically, the processor 200 first uses a community discovery algorithm to determine the community structure in the knowledge graph. In this embodiment, the community discovery algorithm can be, for example, a greedy algorithm. Then, the processor 200 uses a reinforcement learning algorithm to analyze the community structure and determine the potential groups in the community structure. In addition, the processor 200 can also verify the analysis results and explain the decomposition results, thereby providing an execution strategy for mining potential groups. The above content will be described in detail later, so it will not be repeated here.

[0054] As described in the background technology, when using graph mining to identify high-risk users in massive data, in order to achieve the best mining effect, all users in the database are generally mined, and all users are divided into communities through various relationship media. After the group with higher fraud risk is found, the users in the group are removed. However, in most cases, due to the uneven quality of each sample in the database and the poor significance of the relationship between each node in the knowledge graph, the direct use of community discovery algorithms to identify potential groups that perform fraudulent behavior is ineffective.

[0055] In view of this, the present application provides a method for identifying potential gangs that perform fraudulent activities. And referring to the above-mentioned contents, the technical solution of the present application pre-classifies the samples in the database (i.e., multiple samples in the database are divided into first black samples and first white samples), and determines the second black sample in the first sample set based on the pre-trained first recognition model (i.e., the second black sample is the black sample remaining after the first black sample in the first sample set is subjected to noise reduction processing), and determines the second white sample in the second sample set based on the pre-trained second recognition model (i.e., the second white sample is the white sample remaining after the first white sample in the second sample set is subjected to noise reduction processing).

[0056] Therefore, unlike the existing method of directly using graph mining algorithms to mine all samples in the database (i.e., the first black sample and the first white sample), the technical solution of the present application pre-noises the samples in the database (i.e., the first black sample and the first white sample), so that the remaining samples (i.e., the second black sample and the second white sample) are more representative. Then, the remaining samples (i.e., the second black sample and the second white sample) and the relationship media between the samples (i.e., the first relationship media and the second relationship media) are used to construct a knowledge graph. Finally, a community discovery algorithm is used to identify potential gangs that perform fraudulent activities in the knowledge graph.

[0057] This can achieve the technical effect of improving the significance of each node in the knowledge graph and ensuring the efficiency of using the community discovery algorithm to identify potential groups that perform fraudulent behavior. This solves the technical problem in the prior art that the quality of each sample in the database is uneven and the significance of the relationship between each node is poor, so the direct use of the community discovery algorithm to identify potential groups that perform fraudulent behavior is poor.

[0058] Optionally, the operation of determining multiple second black samples in the first sample set based on the pre-trained first recognition model includes: extracting multiple third black samples from the multiple first black samples in the first sample set, and constructing a third sample set, wherein the third black samples are used to represent the key behavior patterns of the first black samples; constructing a first recognition model; training the first recognition model using the multiple third black samples in the third sample set; and using the trained first recognition model to identify each first black sample in the first sample set, and determine multiple second black samples.

[0059] Specifically, Figure 5 is a flow chart of a method for reducing noise on a first black sample according to an embodiment of the present application. Figure 5 As shown, in order to reduce the size of the network to be mined and improve the mining efficiency, before identifying the multiple first black samples in the database, it is also necessary to reduce the noise of the multiple first black samples. Thus, firstly, the processor 200 determines the first sample set. The first black samples in the first sample set may include, for example, user behavior records, transaction patterns, and social interaction information.

[0060] Then, the processor 200 extracts useful features from the first black samples in the first sample set, and generates corresponding feature vectors (i.e., third black samples) based on the extracted features. The extracted features can represent the key behavior patterns of the first black samples, such as frequent small transactions, abnormal capital flows, abnormal connections in social networks, etc. Further, the processor 200 constructs a first recognition model using a structured classification method. The first recognition model can be, for example, a SCIRL model.

[0061] Then, the processor 200 inputs the third black sample in the third sample set into the first recognition model, trains the first recognition model using the third black sample, and evaluates the accuracy and generalization ability of the first recognition model through cross-validation and other methods.

[0062] Finally, the processor 200 uses the trained first recognition model to evaluate the first black sample in the first sample set, determines a second black sample whose behavior pattern similarity with the first black sample is greater than a first preset threshold, and marks the second black sample as a seed user. The first preset threshold is a threshold determined by the operator based on actual conditions.

[0063] Therefore, since the technical solution of the present invention adopts the SCIRL algorithm to reduce the noise of the first black sample and determines the significant second black sample in the first black sample, it is possible to mine the potential gangs that perform fraudulent behavior based on the second black sample, thereby improving the effect of identifying the potential gangs that perform fraudulent behavior.

[0064] Optionally, the operation of determining multiple second white samples in the second sample set based on the pre-trained second recognition model includes: eliminating abnormal white samples in the second sample set and constructing a fourth sample set; eliminating popular white samples in the fourth sample set and generating a fifth sample set containing multiple third white samples; constructing a second recognition model and training the second recognition model using the multiple third white samples in the fifth sample set; and using the trained second recognition model to identify each first white sample in the second sample set and determine multiple second white samples.

[0065] Specifically, since there are a large number of first white samples in the database, directly performing graph mining without denoising the first white samples may cause a waste of hardware resources, or the mined network may be unusable. Therefore, in order to improve the mining accuracy and speed, business attributes can be added for judgment, and some abnormal nodes and hot nodes can be eliminated.

[0066] Figure 6 is a flow chart of a method for reducing noise on a first white sample according to an embodiment of the present application. Figure 6 As shown, first, the processor 200 counts the features of the first white sample and removes the abnormal white samples in the second sample set, thereby constructing a fourth sample set. Among them, the abnormal white sample can be, for example, a white sample that does not belong to certain business attributes in a specified business scenario but is uniformly filled as a white sample under the business attribute category. Then, the processor 200 excludes some white samples and popular product samples on the business whitelist according to the specific business.

[0067] Further, after removing abnormal white samples, white samples on the business white list, and popular commodity samples, the processor 200 calculates the number of sample edges connected to the remaining first white samples, and removes the first white samples (i.e., popular white samples) whose sample edge number is greater than the fifth preset threshold. The fifth preset threshold can be determined by the operator based on actual conditions. Thus, after removing abnormal white samples, white samples on the business white list, popular commodity samples, and popular white samples from the plurality of first white samples, the processor 200 generates a fifth sample set including the third white sample.

[0068] Then, the processor 200 constructs a second recognition model using a structured classification method. The second recognition model may be, for example, a SCIRL model. The processor 200 inputs the third white sample in the fifth sample set into the second recognition model, and trains the second recognition model using the plurality of third white samples.

[0069] Finally, the processor 200 uses the trained second recognition model to identify each first white sample in the second sample set, marks the first white samples with extremely low violation rates, and removes the first white samples with extremely low violation rates, thereby determining a plurality of second white samples. The violation rate is used to indicate the probability of whether the first white sample complies with the business logic.

[0070] Therefore, since the technical solution of the present invention adopts the SCIRL algorithm to reduce the noise of the first white sample, eliminates invalid white samples in the first white sample, and determines the second white sample that can represent the key behavior pattern of the first white sample, it is possible to identify potential groups that perform fraudulent activities based on the second white sample, thereby improving the effect of identifying potential groups that perform fraudulent activities.

[0071] Optionally, the operation of constructing a knowledge graph based on multiple second black samples, the first relationship medium between each second black sample, multiple second white samples and the second relationship medium between each second white sample includes: determining the first relationship medium between each second black sample, wherein the operation of determining the first relationship medium between each second black sample includes: determining the violation rate of the third relationship medium of each second black sample, and eliminating the third relationship medium whose violation rate is less than a third preset threshold, thereby determining the fourth relationship medium between each second black sample; determining the first similarity relationship and the first medium association between each second black sample, and determining the fifth relationship medium between each second black sample based on the first similarity relationship and the first medium association, wherein the first similarity relationship is used to indicate the business attribute similarity between each second black sample; and fusing the fourth relationship medium and the fifth relationship medium, thereby determining the first relationship medium between each second black sample.

[0072] Specifically, Figure 7is a flow chart of a method for determining a first relationship medium between each second black sample according to an embodiment of the present application. Figure 7 As shown, first, when the processor 200 determines the second black sample in the first sample set, it is necessary to denoise the third relationship medium between each second black sample (i.e., remove invalid third relationship medium), so as to determine the fourth relationship medium between each second black sample, thereby reducing the scale of the network that needs to be mined.

[0073] For example, the processor 200 determines the violation rate of the third relationship medium corresponding to each second black sample, and removes the third relationship medium whose violation rate is less than the third preset threshold, thereby determining the fourth relationship medium between each second black sample. The third preset threshold may be, for example, a violation rate threshold preset by an operator based on actual conditions.

[0074] Then, the processor 200 enhances the third relationship medium between each second black sample. For example, the processor 200 determines the first similarity relationship and the first medium association between each second black sample, and then the processor 200 merges the first similarity relationship and the first medium association to generate a fifth relationship medium (i.e., a newly added relationship medium). The first medium association is used to indicate that any two black samples between multiple second black samples use the same device. For example, a POS machine, etc.

[0075] Finally, the processor 200 merges the fourth relational medium (ie, the relational medium retained after noise reduction) and the fifth relational medium (ie, the relational medium increased after enhancement), thereby determining the first relational medium between the second black samples.

[0076] Optionally, the operation of constructing a knowledge graph based on multiple second black samples, the first relationship medium between each second black sample, multiple second white samples and the second relationship medium between each second white sample includes: determining the second relationship medium between each second white sample, wherein the operation of determining the second relationship medium between each second white sample includes: determining the violation rate of the sixth relationship medium of each second white sample, and eliminating the sixth relationship medium whose violation rate is less than a fourth preset threshold, thereby determining the seventh relationship medium between each second white sample; determining the second similarity relationship and the second medium association between each second white sample, and determining the eighth relationship medium between each second white sample based on the second similarity relationship and the second medium association, wherein the second similarity relationship is used to indicate the business attribute similarity between each second white sample; and fusing the seventh relationship medium and the eighth relationship medium, thereby determining the second relationship medium between each second white sample.

[0077] Specifically, Figure 8is a schematic diagram of determining a second relationship medium between each second white sample according to an embodiment of the present application. Figure 8 As shown, first, when the processor 200 determines the second white sample in the second sample set, it is necessary to denoise the sixth relationship medium between each second white sample (i.e., remove the invalid sixth relationship medium), so as to determine the seventh relationship medium between each second white sample, thereby reducing the scale of the network that needs to be mined.

[0078] For example, the processor 200 determines the violation rate of the sixth relationship medium corresponding to each second white sample, and removes the sixth relationship medium whose violation rate is less than the fourth preset threshold, thereby determining the seventh relationship medium between each second white sample. The fourth preset threshold may be, for example, a violation rate threshold preset by an operator based on actual conditions.

[0079] Then, the processor 200 enhances the eighth relationship medium between each second white sample. For example, the processor 200 determines the second similarity relationship and the second medium association between each second white sample, and then the processor 200 merges the second similarity relationship and the second medium association to generate the eighth relationship medium (i.e., the newly added relationship medium). The second medium association is used to indicate the same device used between any two second white samples among the plurality of second white samples.

[0080] Finally, the processor 200 merges the seventh relational medium (ie, the relational medium retained after noise reduction) and the eighth relational medium (ie, the relational medium increased after enhancement), thereby determining the second relational medium between the second white samples.

[0081] In addition, denoising can remove noise and outliers in the sample, which helps to reduce misleading information and improve the accuracy of the community discovery algorithm. The denoised sample can improve the performance of the model, making the results obtained using the community discovery algorithm more reliable and closer to the real-world gang structure. The denoised sample can reduce the complexity of the processor 200 calculation by reducing redundant and unnecessary information, thereby speeding up the execution speed of the community discovery algorithm. Through denoising, the community discovery algorithm can reveal clearer and more definite community boundaries, making the results easier to explain and understand. Noise reduction helps to eliminate normal user behaviors that may be misinterpreted as gang members, thereby reducing the false alarm rate, and important features and patterns are more prominent, which helps the community discovery algorithm to more accurately identify gang members.

[0082] Furthermore, accurate community discovery results can help law enforcement agencies and companies allocate resources more effectively and focus on high-risk gangs. In some cases, noise reduction can also remove or mask sensitive information to reduce the risk of data leakage. The noise-reduced data enables the community discovery algorithm to better adapt to data changes, such as the evolution of gang structure.

[0083] And high-quality denoising samples and accurate community discovery results can be extended to other fields, such as marketing, community network analysis, etc. Noise reduction is a key step in data analysis, which provides a more solid and reliable foundation for group mining using community discovery algorithms.

[0084] Optionally, the operation of using a community discovery algorithm to identify potential gangs that perform fraudulent activities in a knowledge graph includes: using a community discovery algorithm to determine the community structure in the knowledge graph; and using a reinforcement learning algorithm to analyze the community structure and determine potential gangs in the community structure.

[0085] Specifically, when the processor 200 constructs a knowledge graph based on a plurality of second black samples, a first relationship medium between each second black sample, a plurality of second white samples, and a second relationship medium between each second white sample, the processor 200 uses a community discovery algorithm to determine the community structure in the knowledge graph. Among them, the processor 200 can, for example, determine a suitable community discovery algorithm based on the characteristics of each second black sample and each second white sample. In this embodiment, for example, a greedy algorithm can be used to determine the community structure in the knowledge graph. In addition, the processor 200 can also adjust parameters based on the characteristics of the selected community discovery algorithm to obtain the best results.

[0086] Then, the processor 200 uses a reinforcement learning algorithm to analyze the community structure in the identified knowledge graph and determine whether there are potential gangs in the community structure. In addition, the processor 200 can also identify key nodes in the community structure based on the reinforcement learning algorithm, thereby assisting the operator to trace back from these key nodes to analyze whether the community structure is a gang that performs fraudulent behavior, or the distribution of these gangs that perform fraudulent behavior.

[0087] Finally, the processor 200 verifies the results obtained based on the community discovery algorithm, interprets the obtained results, and provides mining and insights into potential groups performing fraudulent activities.

[0088] Thus, by utilizing a community discovery algorithm to determine the community structure in a knowledge graph, and utilizing a reinforcement learning algorithm to analyze the community structure and determine the operations of potential gangs in the community structure, it is possible to achieve the technical effect of accurately determining potential gangs that may perform fraudulent activities.

[0089] Ideally, the knowledge graph obtained after denoising has a completely cohesive community structure, with zero connections between community structures, so that the community discovery algorithm can get the result directly after just one round of running. Of course, such a good situation is not possible in actual scenarios. The quality of the data source and the richness of expert experience will affect the effect of denoising. Generally speaking, as long as denoising can remove more than 90% of the noise, the community discovery algorithm can basically complete the overall community discovery process through several rounds of iterations. Therefore, the denoised samples can greatly increase the speed of mining.

[0090] Optionally, it also includes: determining the key nodes in the potential gang, and determining the operation mode and fraud strategy of the potential gang based on the determined key nodes. Further optionally, the operation of determining the key nodes in the potential gang includes: extracting the features of each node in the knowledge graph and generating a feature vector corresponding to each node; determining the behavior mode of each node in the knowledge graph; determining the training sample based on the feature vector of each node and the behavior mode of each node; using the training sample to train the third recognition model, and using the trained third recognition model to evaluate the value of each node in the knowledge graph; and determining the key node based on the value of each node in the knowledge graph and the fourth preset threshold.

[0091] Specifically, after determining the potential group that performs fraudulent behavior, the processor 200 can further determine the key nodes in the potential group, thereby identifying the role that the key node may play in the potential group, thereby helping to analyze the behavior pattern of the potential group and helping to understand the operation mode and fraud strategy of the potential group.

[0092] Thus, first, the processor 200 collects data of all nodes and edges in the knowledge graph. Wherein, the nodes may be, for example, each second black sample or each second white sample, and the edges may be, for example, the first relationship medium between each second black sample and the second relationship medium between each second white sample. And wherein, the data may include, for example, node features and edge attributes. Node features may include, for example, degree centrality and node clustering coefficient, etc. Edge attributes may include, for example, weight and type, etc.

[0093] Then, the processor 200 extracts features based on the attributes of the nodes and the network structure corresponding to the nodes, and generates feature vectors corresponding to each node. The features extracted by the processor 200 can reflect the role and influence of the nodes in the corresponding network structure.

[0094] Furthermore, the processor 200 determines the behavior pattern of each node in the knowledge graph, wherein the behavior pattern of each node may be, for example, the transaction behavior, information dissemination behavior or other interactive behavior of the node.

[0095] Then, the processor 200 combines the feature vectors corresponding to each node with the behavior patterns of each node to form state-behavior pairs, and uses the state-behavior pairs corresponding to each node as training samples for the third recognition model, wherein the third recognition model may be, for example, a SCIRL model.

[0096] Then, the processor 200 trains a third recognition model based on the state-behavior pairs (i.e., training samples) corresponding to each node, wherein the mathematical principle of the third recognition model is to learn a reward function, and the reward function will evaluate the value of the corresponding node based on the state-behavior pairs of each node.

[0097] Finally, the processor 200 uses the third recognition model to evaluate each point in the knowledge graph, thereby obtaining the "importance score" of each node, and based on the "importance score" of each node and the preset fourth preset threshold, the nodes are divided into key nodes and secondary nodes. Wherein, when the "importance score" of the node is greater than the fourth preset threshold, the node is divided into a key node; when the "importance score" of the node is less than or equal to the fourth preset threshold, the node is divided into a secondary node.

[0098] In this embodiment, the key node can be, for example, a user who plays a key role in a potential group that performs fraudulent behavior, such as a core member of a potential group or a node that connects multiple potential groups. Therefore, in the technical solution of the invention, by combining the potential group that performs fraudulent behavior identified based on the community discovery algorithm with the key nodes in the potential group, the operation mode and fraud strategy of the potential group can be understood.

[0099] The key points of the present invention are:

[0100] 1. The present invention uses the SCIRL algorithm to cluster the first black samples in the database, and finds the second black samples with high similarity to the first black samples as seed users, so that after determining the second black samples, fraudulent behaviors similar to the second black samples can be found from the second black samples.

[0101] 2. The present invention can effectively avoid the occurrence of abnormal nodes and hot nodes in network mining by adding business attributes to the first white sample, and then train the second recognition model by using the second white sample obtained after the denoising, so that the second recognition model generated after the training can be used to eliminate some invalid first white samples and reduce the amount of data to be mined.

[0102] 3. The present invention not only performs noise reduction on the third relationship medium between each second black sample and the sixth relationship medium between each second white sample, but also adds the fifth relationship medium between each second black sample and the eighth relationship medium between each second white sample, so that the first relationship medium between each second black sample can be determined based on the third relationship medium determined after noise reduction and the fifth relationship medium determined after enhancement; the second relationship medium between each second white sample can be determined based on the seventh relationship medium determined after noise reduction and the eighth relationship medium determined after enhancement, and finally the first relationship medium and the second relationship medium can be used to more directly mine the business relationship.

[0103] 4. The present invention uses the SCIRL algorithm to learn the reward function, and evaluates the importance or influence of nodes based on node characteristics and network structure, and then classifies the nodes into key nodes and secondary nodes according to the "importance score" of the nodes, so that the key nodes can be combined with the community discovery algorithm, and the key roles and secondary key roles in the potential gangs can be determined, thereby assisting business personnel in determining whether it is gang fraud.

[0104] Therefore, according to the first aspect of this embodiment, the technical effect of improving the significance of each node in the knowledge graph and ensuring the efficiency of identifying potential gangs that perform fraudulent activities using community discovery algorithms can be achieved.

[0105] In addition, reference Figure 1 As shown, according to the second aspect of this embodiment, a storage medium is provided, wherein the storage medium includes a stored program, wherein when the program is run, a processor executes any one of the above methods.

[0106] Therefore, according to this embodiment, the technical effect of improving the significance of each node in the knowledge graph and ensuring the efficiency of identifying potential gangs that perform fraudulent activities using community discovery algorithms can be achieved.

[0107] It should be noted that, for the above-mentioned method embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should know that the present invention is not limited by the described action sequence, because according to the present invention, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present invention.

[0108] Through the description of the above implementation methods, those skilled in the art can clearly understand that the method according to the above embodiment can be implemented by means of software plus a necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for enabling a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods described in each embodiment of the present invention. Example 2

[0109] Fig. 9 FIG. 9 shows a device 900 for identifying a potential group that performs fraudulent behavior according to this embodiment, and the device 900 corresponds to the method according to embodiment 1. Fig. 9 As shown, the device 900 includes: a first sample set construction module 910, which is used to obtain multiple samples from a database and divide the multiple samples into a first sample set consisting of multiple first black samples and a second sample set consisting of multiple first white samples, wherein the first black samples are used to indicate data samples that perform fraudulent behavior in a specified business scenario, and the first white samples are used to indicate data samples that perform safe operation behavior in a specified business scenario; a second black sample determination module 920, which is used to determine multiple second black samples in the first sample set based on a pre-trained first recognition model, wherein the behavior pattern of the second black samples is different from the behavior pattern of the first black samples. The similarity between the two is greater than a first preset threshold; a second white sample determination module 930, used to determine the second white sample in the second sample set based on the pre-trained second recognition model, wherein the similarity between the behavior pattern of the second white sample and the behavior pattern of the first white sample is less than a second preset threshold; a knowledge graph construction module 940, used to construct a knowledge graph based on multiple second black samples, first relationship media between each second black sample, multiple second white samples and second relationship media between each second white sample; and a potential gang identification module 950, used to use a community discovery algorithm to identify potential gangs that perform fraudulent behavior in the knowledge graph.

[0110] Optionally, the second black sample determination module 920 includes: a second sample set construction module, used to extract multiple third black samples from multiple first black samples in the first sample set, and construct a third sample set, wherein the third black samples are used to represent the key behavior patterns of the first black samples; a first recognition model construction module, used to construct a first recognition model; a first recognition model training module, used to train the first recognition model using multiple third black samples in the third sample set; and a second black sample determination submodule, used to use the trained first recognition model to identify each first black sample, and determine multiple second black samples.

[0111] Optionally, the second white sample determination module 930 includes: a third sample set construction module, used to eliminate abnormal white samples in the second sample set and construct a fourth sample set; a fourth sample set construction module, used to eliminate popular white samples in the fourth sample set and generate a fifth sample set containing multiple third white samples; a second recognition model training module, used to construct a second recognition model, and use the multiple third white samples in the fifth sample set to train the second recognition model; and a second white sample determination submodule, used to use the trained second recognition model to identify each first white sample, and determine multiple second white samples.

[0112] Optionally, the knowledge graph construction module 940 includes: a first relationship medium determination module, which is used to determine the first relationship medium between each second black sample, wherein the first relationship medium determination module includes: a first relationship medium determination submodule, which is used to determine the violation rate of the third relationship medium of each second black sample, and eliminate the third relationship medium with a violation rate less than a third preset threshold, so as to determine the fourth relationship medium between each second black sample; a second relationship medium determination submodule, which is used to determine the first similarity relationship and the first medium association between each second black sample, and determine the fifth relationship medium between each second black sample based on the first similarity relationship and the first medium association, wherein the first similarity relationship is used to indicate the business attribute similarity between each second black sample; and a third relationship medium determination submodule, which is used to merge the fourth relationship medium and the fifth relationship medium, so as to determine the first relationship medium between each second black sample.

[0113] Optionally, the knowledge graph construction module 940 includes: a second relationship medium determination module, which is used to determine the second relationship medium between each second white sample, wherein the second relationship medium determination module includes: a fourth relationship medium determination submodule, which is used to determine the violation rate of the sixth relationship medium of each second white sample, and eliminate the sixth relationship medium whose violation rate is less than the fourth preset threshold, so as to determine the seventh relationship medium between each second white sample; a fifth relationship medium determination submodule, which is used to determine the second similarity relationship and the second medium association between each second white sample, and determine the eighth relationship medium between each second white sample based on the second similarity relationship and the second medium association, wherein the second similarity relationship is used to indicate the business attribute similarity between each second white sample; and a sixth relationship medium determination submodule, which is used to merge the seventh relationship medium and the eighth relationship medium, so as to determine the second relationship medium between each second white sample.

[0114] Optionally, the potential group identification module 950 includes: a community structure determination module for determining the community structure in the knowledge graph using a community discovery algorithm; and a potential group determination module for analyzing the community structure and determining potential groups in the community structure using a reinforcement learning algorithm.

[0115] Optionally, the device 900 further includes: a key node determination module, which is used to determine the key nodes in the potential gang, and based on the determined key nodes, determine the operation mode and fraud strategy of the potential gang.

[0116] Optionally, the key node determination module includes: a feature vector generation module, which is used to extract the features of each node in the knowledge graph and generate a feature vector corresponding to each node; a behavior pattern determination module, which is used to determine the behavior pattern of each node in the knowledge graph; a training sample determination module, which is used to determine the training sample based on the feature vector of each node and the behavior pattern of each node; a value determination module, which is used to train the third recognition model using the training sample, and use the trained third recognition model to evaluate the value of each node in the knowledge graph; and a key node determination sub-module, which is used to determine the key node based on the value of each node in the knowledge graph and a fourth preset threshold.

[0117] Therefore, according to this embodiment, the technical effect of improving the significance of each node in the knowledge graph and ensuring the efficiency of identifying potential gangs that perform fraudulent activities using community discovery algorithms can be achieved. Example 3

[0118] Fig.10 FIG. 1 shows a device 1000 for identifying a potential group that performs fraudulent behavior according to this embodiment, and the device 1000 corresponds to the method according to Embodiment 1. Fig.10As shown, the device 1000 includes: a processor 1010; and a memory 1020, which is connected to the processor 1010 and is used to provide the processor 1010 with instructions for processing the following processing steps: obtaining multiple samples from a database, and dividing the multiple samples into a first sample set consisting of multiple first black samples and a second sample set consisting of multiple first white samples, wherein the first black samples are used to indicate data samples that perform fraudulent behavior in a specified business scenario, and the first white samples are used to indicate data samples that perform safe operation behavior in a specified business scenario; determining multiple second black samples in the first sample set based on a pre-trained first recognition model, wherein the similarity between the behavior pattern of the second black sample and the behavior pattern of the first black sample is greater than a first preset threshold; determining second white samples in the second sample set based on a pre-trained second recognition model, wherein the similarity between the behavior pattern of the second white sample and the behavior pattern of the first white sample is less than a second preset threshold; constructing a knowledge graph based on multiple second black samples, first relationship media between each second black sample, multiple second white samples, and second relationship media between each second white sample; and using a community discovery algorithm to identify potential gangs that perform fraudulent behavior in the knowledge graph.

[0119] Optionally, the operation of determining multiple second black samples in the first sample set based on the pre-trained first recognition model includes: extracting multiple third black samples from the multiple first black samples in the first sample set, and constructing a third sample set, wherein the third black samples are used to represent the key behavior patterns of the first black samples; constructing a first recognition model; training the first recognition model using the multiple third black samples in the third sample set; and using the trained first recognition model to identify each first black sample and determine multiple second black samples.

[0120] Optionally, the operation of determining multiple second white samples in the second sample set based on the pre-trained second recognition model includes: eliminating abnormal white samples in the second sample set and constructing a fourth sample set; eliminating popular white samples in the fourth sample set and generating a fifth sample set containing multiple third white samples; constructing a second recognition model and training the second recognition model using the multiple third white samples in the fifth sample set; and using the trained second recognition model to identify each first white sample and determine multiple second white samples.

[0121] Optionally, the operation of constructing a knowledge graph based on multiple second black samples, the first relationship medium between each second black sample, multiple second white samples and the second relationship medium between each second white sample includes: determining the first relationship medium between each second black sample, wherein the operation of determining the first relationship medium between each second black sample includes: determining the violation rate of the third relationship medium of each second black sample, and eliminating the third relationship medium whose violation rate is less than a third preset threshold, thereby determining the fourth relationship medium between each second black sample; determining the first similarity relationship and the first medium association between each second black sample, and determining the fifth relationship medium between each second black sample based on the first similarity relationship and the first medium association, wherein the first similarity relationship is used to indicate the business attribute similarity between each second black sample; and fusing the fourth relationship medium and the fifth relationship medium, thereby determining the first relationship medium between each second black sample.

[0122] Optionally, the operation of constructing a knowledge graph based on multiple second black samples, the first relationship medium between each second black sample, multiple second white samples and the second relationship medium between each second white sample includes: determining the second relationship medium between each second white sample, wherein the operation of determining the second relationship medium between each second white sample includes: determining the violation rate of the sixth relationship medium of each second white sample, and eliminating the sixth relationship medium whose violation rate is less than a fourth preset threshold, thereby determining the seventh relationship medium between each second white sample; determining the second similarity relationship and the second medium association between each second white sample, and determining the eighth relationship medium between each second white sample based on the second similarity relationship and the second medium association, wherein the second similarity relationship is used to indicate the business attribute similarity between each second white sample; and fusing the seventh relationship medium and the eighth relationship medium, thereby determining the second relationship medium between each second white sample.

[0123] Optionally, the operation of using a community discovery algorithm to identify potential gangs that perform fraudulent activities in a knowledge graph includes: using a community discovery algorithm to determine the community structure in the knowledge graph; and using a reinforcement learning algorithm to analyze the community structure and determine potential gangs in the community structure.

[0124] Optionally, it also includes: determining key nodes in the potential gang, and based on the determined key nodes, determining the operation mode and fraud strategy of the potential gang.

[0125] Optionally, the operation of determining the key nodes in a potential gang includes: extracting features of each node in the knowledge graph and generating feature vectors corresponding to each node; determining the behavior pattern of each node in the knowledge graph; determining training samples based on the feature vectors of each node and the behavior pattern of each node; using the training samples to train a third recognition model, and using the trained third recognition model to evaluate the value of each node in the knowledge graph; and determining the key nodes based on the value of each node in the knowledge graph and a fourth preset threshold.

[0126] Therefore, according to this embodiment, the technical effect of improving the significance of each node in the knowledge graph and ensuring the efficiency of identifying potential gangs that perform fraudulent activities using community discovery algorithms can be achieved.

[0127] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0128] In the above embodiments of the present invention, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0129] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0130] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0131] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0132] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or optical disk, etc. Various media that can store program codes.

[0133] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for identifying potential gangs that perform fraudulent activities, characterized in that: include: Acquire multiple samples from a database, and divide the multiple samples into a first sample set consisting of multiple first black samples and a second sample set consisting of multiple first white samples, wherein the first black samples are used to indicate data samples that perform fraudulent behaviors in a specified business scenario, and the first white samples are used to indicate data samples that perform safe operation behaviors in a specified business scenario; Determine a plurality of second black samples in the first sample set based on a pre-trained first recognition model, wherein a similarity between a behavior pattern of the second black samples and a behavior pattern of the first black samples is greater than a first preset threshold, and wherein the second black samples are used to indicate data samples that are not determined to have performed fraudulent behavior in a specified business scenario but have a high risk of fraud; Determine a plurality of second white samples in the second sample set based on a pre-trained second recognition model, wherein a similarity between a behavior pattern of the second white samples and a behavior pattern of the first white samples is less than a second preset threshold, and wherein the second white samples are used to indicate data samples that are not determined to perform a security operation behavior in a specified business scenario but have a low fraud risk; Based on the plurality of second black samples, the first relationship medium between each second black sample, the plurality of second white samples and the second relationship medium between each second white sample, constructing a knowledge graph, wherein the first relationship medium is used to indicate association information between each second black sample, and the second relationship medium is used to indicate association information between each second white sample; as well as Using a community discovery algorithm, potential groups that perform fraudulent activities in the knowledge graph are identified, where The method further includes: determining key nodes in the potential gang, and determining the operation mode and fraud strategy of the potential gang based on the determined key nodes, wherein Determine the operations of key nodes in the potential gang, including: Extracting features of each node in the knowledge graph and generating feature vectors corresponding to each node; Determine the behavior pattern of each node in the knowledge graph; Determining training samples based on the feature vectors of the nodes and the behavior patterns of the nodes; Using the training samples to train a third recognition model, and using the trained third recognition model to evaluate the value of each node in the knowledge graph; and The key node is determined based on the value of each node in the knowledge graph and a fourth preset threshold.

2. The method according to claim 1, characterized in that The operation of determining a plurality of second black samples in the first sample set based on the pre-trained first recognition model includes: Extracting a plurality of third black samples from a plurality of first black samples in the first sample set, and constructing a third sample set, wherein the third black samples are used to represent key behavior patterns of the first black samples; constructing the first recognition model; Using a plurality of third black samples in the third sample set to train the first recognition model; and The trained first recognition model is used to recognize each first black sample, and the plurality of second black samples are determined.

3. The method according to claim 1, characterized in that The operation of determining a plurality of second white samples in the second sample set based on the pre-trained second recognition model includes: Eliminate abnormal white samples in the second sample set and construct a fifth sample set; Eliminate the popular white samples in the fifth sample set, and generate a fourth sample set including a plurality of third white samples; constructing the second recognition model, and using a plurality of third white samples in the fourth sample set to train the second recognition model; and The trained second recognition model is used to recognize each first white sample, and the plurality of second white samples are determined.

4. The method according to claim 1, characterized in that: The operation of constructing a knowledge graph based on the plurality of second black samples, the first relationship medium between each second black sample, the plurality of second white samples, and the second relationship medium between each second white sample includes: determining the first relationship medium between each second black sample, wherein the operation of determining the first relationship medium between each second black sample includes: Determine the violation rate of the third relationship medium of each of the second black samples, and eliminate the third relationship medium whose violation rate is less than a third preset threshold, so as to determine the fourth relationship medium between each of the second black samples; Determining a first similarity relationship and a first medium association between the second black samples, and determining a fifth medium relationship between the second black samples based on the first similarity relationship and the first medium association, wherein the first similarity relationship is used to indicate a business attribute similarity between the second black samples; and The fourth relational medium and the fifth relational medium are merged to determine the first relational medium between the second black samples.

5. The method according to claim 1, characterized in that The operation of constructing a knowledge graph based on the plurality of second black samples, the first relationship medium between each second black sample, the plurality of second white samples, and the second relationship medium between each second white sample includes: determining the second relationship medium between each second white sample, wherein the operation of determining the second relationship medium between each second white sample includes: Determine the violation rate of the sixth relationship medium of each of the second white samples, and eliminate the sixth relationship medium whose violation rate is less than a fourth preset threshold, so as to determine the seventh relationship medium between each of the second white samples; determining a second similarity relationship and a second medium association between the second white samples, and determining an eighth medium relationship between the second white samples based on the second similarity relationship and the second medium association, wherein the second similarity relationship is used to indicate a business attribute similarity between the second white samples; and The seventh relational medium and the eighth relational medium are merged to determine a second relational medium between the second white samples.

6. The method according to claim 1, characterized in that Using a community discovery algorithm, identify the operations of potential groups performing fraudulent activities in the knowledge graph, including: Determining the community structure in the knowledge graph using the community discovery algorithm; and The community structure is analyzed by using a reinforcement learning algorithm, and the potential gangs in the community structure are determined.

7. A storage medium, characterized in that: The storage medium includes a stored program, wherein when the program is executed, the processor executes the method according to any one of claims 1 to 6.

8. A device for identifying potential gangs that perform fraudulent activities, characterized in that: include: A first sample set construction module is used to obtain multiple samples from a database and divide the multiple samples into a first sample set consisting of multiple first black samples and a second sample set consisting of multiple first white samples, wherein the first black samples are used to indicate data samples that perform fraudulent behaviors in a specified business scenario, and the first white samples are used to indicate data samples that perform security operation behaviors in a specified business scenario; a second black sample determination module, configured to determine a plurality of second black samples in the first sample set based on a pre-trained first recognition model, wherein a similarity between a behavior pattern of the second black samples and a behavior pattern of the first black samples is greater than a first preset threshold, and wherein the second black samples are used to indicate data samples that are not determined to have performed fraudulent behavior in a specified business scenario but have a high risk of fraud; A second white sample determination module, configured to determine a second white sample in the second sample set based on a pre-trained second recognition model, wherein a similarity between a behavior pattern of the second white sample and a behavior pattern of the first white sample is less than a second preset threshold, and wherein the second white sample is used to indicate a data sample that is not determined to perform a security operation behavior in a specified business scenario but has a low fraud risk; a knowledge graph construction module, configured to construct a knowledge graph based on the plurality of second black samples, the first relationship medium between the second black samples, the plurality of second white samples, and the second relationship medium between the second white samples, wherein the first relationship medium is used to indicate association information between the second black samples, and the second relationship medium is used to indicate association information between the second white samples; as well as A potential gang identification module is used to identify potential gangs that perform fraudulent activities in the knowledge graph using a community discovery algorithm, wherein The device also includes: a key node determination module, which is used to determine the key nodes in the potential gang, and based on the determined key nodes, determine the operation mode and fraud strategy of the potential gang, and wherein The key node determination module includes: a feature vector generation module, which is used to extract the features of each node in the knowledge graph and generate a feature vector corresponding to each node; A behavior pattern determination module, used to determine the behavior pattern of each node in the knowledge graph; A training sample determination module, used for determining a training sample based on the feature vector of each node and the behavior pattern of each node; a value determination module, configured to train a third recognition model using the training samples, and evaluate the value of each node in the knowledge graph using the trained third recognition model; and The key node determination submodule is used to determine the key node based on the value of each node in the knowledge graph and a fourth preset threshold.

9. A device for identifying potential gangs that perform fraudulent activities, characterized in that: include: processor; as well as A memory, connected to the processor, configured to provide the processor with instructions for processing the following processing steps: Acquire multiple samples from a database, and divide the multiple samples into a first sample set consisting of multiple first black samples and a second sample set consisting of multiple first white samples, wherein the first black samples are used to indicate data samples that perform fraudulent behaviors in a specified business scenario, and the first white samples are used to indicate data samples that perform safe operation behaviors in a specified business scenario; Determine a plurality of second black samples in the first sample set based on a pre-trained first recognition model, wherein a similarity between a behavior pattern of the second black samples and a behavior pattern of the first black samples is greater than a first preset threshold, and wherein the second black samples are used to indicate data samples that are not determined to have performed fraudulent behavior in a specified business scenario but have a high risk of fraud; Determine a second white sample in the second sample set based on a pre-trained second recognition model, wherein a similarity between a behavior pattern of the second white sample and a behavior pattern of the first white sample is less than a second preset threshold, and wherein the second white sample is used to indicate a data sample that is not determined to perform a security operation behavior in a specified business scenario but has a low fraud risk; Based on the plurality of second black samples, the first relationship medium between each second black sample, the plurality of second white samples and the second relationship medium between each second white sample, constructing a knowledge graph, wherein the first relationship medium is used to indicate association information between each second black sample, and the second relationship medium is used to indicate association information between each second white sample; as well as Using a community discovery algorithm, potential groups that perform fraudulent activities in the knowledge graph are identified, where The device also includes: determining key nodes in the potential gang, and based on the determined key nodes, determining the operation mode and fraud strategy of the potential gang, wherein Determine the operations of key nodes in the potential gang, including: Extracting features of each node in the knowledge graph and generating feature vectors corresponding to each node; Determine the behavior pattern of each node in the knowledge graph; Determining training samples based on the feature vectors of the nodes and the behavior patterns of the nodes; Using the training samples to train a third recognition model, and using the trained third recognition model to evaluate the value of each node in the knowledge graph; and The key node is determined based on the value of each node in the knowledge graph and a fourth preset threshold.