Credit group anti-fraud method based on heterogeneous graph

Abstract

The invention relates to the technical field of anti-fraud, in particular to a credit group anti-fraud method based on a heterogeneous graph. The method comprises the steps of collecting multiple pieces of customer data, constructing a customer heterogeneous graph based on the customer data, performing graph analysis on the customer heterogeneous graph, extracting multiple features of the customer heterogeneous graph, then creating a customer heterogeneous model, and inputting the multiple features of the heterogeneous graph as a training set into the customer heterogeneous model to train the customer heterogeneous model. The method comprises the following steps: constructing and analyzing a heterogeneous graph, obtaining a trained client heterogeneous model, finally collecting client real-time data, inputting the client real-time data into the trained client heterogeneous model, and obtaining a client risk assessment report. According to the application, through constructing and analyzing the heterogeneous graph, comprehensively utilizing various data types of clients, and comprehensively capturing a complex relation network among the clients; therefore, the fraud detection accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the field of anti-fraud technology, and in particular to a credit gang anti-fraud method based on heterogeneous graphs. Background Art

[0002] Credit fraud occurs when criminals use false information or other means to obtain loans or credit lines, causing financial losses to financial institutions. With the development of internet finance, credit fraud methods have become increasingly diverse and complex, posing significant challenges to financial institutions.

[0003] Chinese patent publication number CN114579991A discloses an anti-fraud method, system, and computer device based on a heterogeneous graph neural network. The method constructs a user graph based on the heterogeneous graph, and cleans the user graph based on the number of activities to obtain a target user graph. The target user graph is used as input, and a convolution calculation is performed through a graph neural network model to output a user score, and the fraud risk is determined based on the user score. However, the existing technology lacks the ability to comprehensively analyze heterogeneous data, making it difficult to fully capture the complex relationship network between customers, thereby limiting its effectiveness and applicability in practical applications. Summary of the Invention

[0004] The purpose of the present invention is to address the problems existing in the background technology and propose an anti-fraud method for credit gangs based on heterogeneous graphs.

[0005] The technical solution of the present invention:

[0006] This application provides a heterogeneous graph-based credit gang anti-fraud method, including:

[0007] Collect multiple customer data and build a customer heterogeneous graph based on the customer data;

[0008] Perform graph analysis on the customer heterogeneous graph and extract multiple features of the customer heterogeneous graph;

[0009] Creating a customer heterogeneous model, inputting multiple features of the heterogeneous graph as a training set into the customer heterogeneous model to train the customer heterogeneous model, and obtaining a trained customer heterogeneous model;

[0010] Collect real-time customer data, input the real-time customer data into the trained customer heterogeneous model, and obtain customer risk assessment reports.

[0011] Preferably, collecting multiple customer data and constructing a customer heterogeneous graph based on the customer data includes:

[0012] Create a customer data table;

[0013] Collect multiple customer data and put all collected customer data into a customer data table; the customer data table includes customer information, transaction records and contact information.

[0014] Preferably, collecting multiple customer data and constructing a customer heterogeneous graph based on the customer data further includes:

[0015] Define the edges and nodes of the heterogeneous graph;

[0016] Customer data is imported into the nodes of the heterogeneous graph, and multiple nodes of the heterogeneous graph are connected based on the edges of the heterogeneous graph.

[0017] Preferably, the heterogeneous graph includes nodes and edges, the nodes of the heterogeneous graph include customer nodes, transaction nodes and contact information nodes, and the edges of the heterogeneous graph include the connection between customer information and transaction records and the connection between customer information and contact information.

[0018] Preferably, a graph analysis is performed on the customer heterogeneous graph to extract multiple features of the customer heterogeneous graph, including:

[0019] Randomly select a node or an edge of the heterogeneous graph from the heterogeneous graph; record the selected node or edge of the heterogeneous graph as a heterogeneous graph target;

[0020] Extract features of heterogeneous graph targets;

[0021] Returns a randomly selected node or edge of a heterogeneous graph until the features of all nodes or edges of the heterogeneous graph are obtained.

[0022] Preferably, a customer heterogeneous model is created, and multiple features of the heterogeneous graph are input into the customer heterogeneous model as a training set to train the customer heterogeneous model, thereby obtaining a trained customer heterogeneous model, including:

[0023] Create a customer heterogeneous model;

[0024] Normalize the feature vectors corresponding to the features of the heterogeneous graph, and divide the normalized feature vectors into training sets and test sets in random proportions;

[0025] Inputting the training set into the customer heterogeneous model to train the customer heterogeneous model and obtain a trained customer heterogeneous model;

[0026] The test set is input into the trained customer heterogeneous model to determine whether the trained customer heterogeneous model is complete.

[0027] Preferably, real-time customer data is collected and input into a trained customer heterogeneous model to obtain a customer risk assessment report, including:

[0028] Collect real-time customer data;

[0029] Build a customer heterogeneous graph based on real-time customer data;

[0030] Input the heterogeneous graph into the trained customer heterogeneous model and obtain the customer risk assessment report output by the trained heterogeneous graph.

[0031] Preferably, the heterogeneous graph is based on a graph database, which includes JanusGraph and Neo4j.

[0032] Preferably, the model structure of the customer heterogeneous model is set to an Encoder-Decoder architecture.

[0033] Preferably, the characteristics of the heterogeneous graph target include degree distribution of nodes, shortest paths of nodes, and clustering coefficients.

[0034] Compared with the prior art, the above technical solution of the present invention has the following beneficial technical effects:

[0035] By collecting multiple customer data, building a customer heterogeneous graph based on the customer data, and performing graph analysis on the customer heterogeneous graph, multiple features of the customer heterogeneous graph are extracted, and then a customer heterogeneous model is created. The multiple features of the heterogeneous graph are input into the customer heterogeneous model as a training set to train the customer heterogeneous model to obtain a trained customer heterogeneous model. Finally, real-time customer data is collected, and the real-time customer data is input into the trained customer heterogeneous model to obtain a customer risk assessment report. This application improves the accuracy of fraud detection by constructing and analyzing heterogeneous graphs, comprehensively utilizing multiple customer data types, and comprehensively capturing the complex relationship network between customers. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 This is a flow chart of a heterogeneous graph-based credit group anti-fraud method proposed by the present invention;

[0037] Figure 2 This is a structural diagram of the customer heterogeneous model proposed in this invention. DETAILED DESCRIPTION

[0038] like Figure 1 As shown, the present invention proposes a credit gang anti-fraud method based on heterogeneous graph, including:

[0039] S100, collecting multiple customer data and building a customer heterogeneous graph based on the customer data;

[0040] S200, performing graph analysis on the customer heterogeneous graph to extract multiple features of the customer heterogeneous graph;

[0041] S300, creating a customer heterogeneous model, inputting multiple features of the heterogeneous graph as a training set into the customer heterogeneous model to train the customer heterogeneous model, and obtaining a trained customer heterogeneous model;

[0042] S400: Collect real-time customer data, input the real-time customer data into the trained customer heterogeneous model, and obtain a customer risk assessment report.

[0043] In the present invention, by collecting multiple customer data, building a customer heterogeneous graph based on the customer data, and performing graph analysis on the customer heterogeneous graph, multiple features of the customer heterogeneous graph are extracted, and then a customer heterogeneous model is created. The multiple features of the heterogeneous graph are input into the customer heterogeneous model as a training set to train the customer heterogeneous model to obtain a trained customer heterogeneous model. Finally, real-time customer data is collected and input into the trained customer heterogeneous model to obtain a customer risk assessment report. This application improves the accuracy of fraud detection by constructing and analyzing heterogeneous graphs, comprehensively utilizing multiple data types of customers, and comprehensively capturing the complex relationship network between customers.

[0044] In an optional embodiment, the S100 includes:

[0045] S110, create a customer data table;

[0046] S120, collecting multiple customer data, and putting all collected customer data into a customer data table; the customer data table includes customer information, transaction records and contact information.

[0047] It should be noted that by collecting different customer data of multiple customers, more complete customer data can be obtained, so that the heterogeneous graph constructed thereby can fully capture the customer relationship network.

[0048] In an optional embodiment, the S100 further includes:

[0049] S130, defining edges and nodes of the heterogeneous graph;

[0050] S140 , importing customer data into a node of the heterogeneous graph, and connecting multiple nodes of the heterogeneous graph based on edges of the heterogeneous graph.

[0051] It should be noted that heterogeneous graphs can be created through graph databases. In this application, by making full use of customer data for comprehensive analysis, it is not limited to a single type of nodes and edges, thereby improving the detection effect of heterogeneous graphs.

[0052] In an optional embodiment, the heterogeneous graph includes nodes and edges, the nodes of the heterogeneous graph include customer nodes, transaction nodes and contact information nodes, and the edges of the heterogeneous graph include the connection between customer information and transaction records and the connection between customer information and contact information.

[0053] It should be noted that the types of nodes in the heterogeneous graph include customers, telephones, email addresses, and transaction records, while the types of edges in the heterogeneous graph include customer-telephone and customer-transaction. The edge of the heterogeneous graph actually refers to the association relationship between two nodes.

[0054] By constructing and analyzing heterogeneous graphs, we can identify group fraud hidden within complex relationship networks, improving the anti-fraud defense capabilities of financial institutions. Group fraud is highly covert and organized, making it difficult for traditional anti-fraud technologies to address. However, this invention, by comprehensively analyzing multiple customer data types, can more effectively identify and prevent group fraud.

[0055] In an optional embodiment, the step S200 includes:

[0056] S210, randomly selecting a node or an edge of the heterogeneous graph from the heterogeneous graph; recording the selected node or edge of the heterogeneous graph as a heterogeneous graph target;

[0057] S220, extracting features of the heterogeneous graph target;

[0058] S230 , returning to randomly selecting a node of the heterogeneous graph or an edge of the heterogeneous graph from the heterogeneous graph, until the features of all nodes of the heterogeneous graph or the features of the edges of the heterogeneous graph are obtained.

[0059] It should be noted that, when executing step S220 , the features of the nodes of the heterogeneous graph or the features of the edges of the heterogeneous graph may be extracted through a graph analysis tool such as NetworkX.

[0060] The reason for extracting the features of heterogeneous graph targets is that these features can reflect the degree of correlation between different nodes or edges. For example, the higher the degree of a node, the more important the node is in the heterogeneous graph. Therefore, the features of each heterogeneous graph target are used as training samples in the training set, so that the customer heterogeneous model can discover potential risks of customers based on their real-time data.

[0061] In an optional embodiment, the S300 includes:

[0062] S310, creating a customer heterogeneous model;

[0063] S320, normalizing the feature vectors corresponding to the features of the heterogeneous graph, and dividing the normalized feature vectors into a training set and a test set in a random ratio;

[0064] S330, inputting the training set into the customer heterogeneous model to train the customer heterogeneous model, thereby obtaining a trained customer heterogeneous model;

[0065] S340, inputting the test set into the trained customer heterogeneous model to determine whether the trained customer heterogeneous model is complete;

[0066] Specifically, when executing step S340, whether the training of the trained customer heterogeneous model is completed is determined by judging whether the accuracy of the prediction results output by the trained customer heterogeneous model meets the requirements. When the accuracy of the prediction results output by the trained customer heterogeneous model is greater than or equal to the qualified threshold, the training of the trained customer heterogeneous model can be considered to be completed.

[0067] It should be noted that by inputting the training set into the customer heterogeneous model, the customer heterogeneous model trains the feature vectors of the heterogeneous graph constructed by the customer data, thereby continuously learning the relationship between the customer data and the fraud gang.

[0068] Compared to traditional rule-based engines and machine learning models, this invention can more accurately identify fraud hidden within complex networks. In testing, the invention's technical solution improved fraud detection accuracy by 20%. Specifically, by comprehensively analyzing customer information, transaction records, contact information, and other data, the invention can identify more potential fraudulent activities, particularly organized fraud.

[0069] In an optional embodiment, the step S400 includes:

[0070] S410, collecting real-time customer data;

[0071] S420, building a customer heterogeneous graph based on real-time customer data;

[0072] S430: Input the trained customer heterogeneous model into the heterogeneous graph to obtain a customer risk assessment report output by the trained heterogeneous graph.

[0073] It should be noted that after obtaining the trained customer heterogeneous model, the customer's real-time data can be input into the trained customer heterogeneous model to obtain the customer risk assessment report output by the trained customer heterogeneous model.

[0074] In an optional embodiment, the heterogeneous graph is based on a graph database, which includes JanusGraph and Neo4j.

[0075] It should be noted that JanusGraph is an open source distributed graph database that supports the storage and query of large-scale graph data. It is particularly suitable for processing graph data containing tens of billions of vertices and edges and can be distributed in multi-machine clusters.

[0076] Neo4j is a high-performance NoSQL graph database that stores structured data on a network rather than in tables. It is an embedded, disk-based, fully transactional Java persistence engine, but it stores structured data on a network (mathematically called a graph) rather than in tables. Neo4j can also be considered a high-performance graph engine with all the features of a mature database.

[0077] like Figure 2 As shown, in an optional embodiment, the model structure of the customer heterogeneous model is set to an Encoder-Decoder architecture.

[0078] It should be noted that the purpose of the Encoder part of the Encoder-Decoder architecture is to learn node features from a heterogeneous graph. It performs heterogeneous convolution by inputting target and source node features, target and source node types, and edge types. The output vector after the heterogeneous convolution is scored and normalized using Attention. After passing the activation function, it enters the next layer. The original features and the processed features are spliced using Resnet and sent to the hidden layer to calculate the risk score and label.

[0079] Traditional rule engines and machine learning models are prone to false positives due to data and feature limitations, resulting in unnecessary business losses for financial institutions. The technical solution of the present invention reduces the false positive rate by 15%. Through more comprehensive and accurate fraud detection, the present invention can reduce false positives and improve the efficiency and accuracy of fraud detection.

[0080] In an optional embodiment, the characteristics of the heterogeneous graph target include degree distribution of nodes, shortest paths of nodes, and clustering coefficients.

[0081] It should be noted that the degree of a node refers to the number of edges connected to a node, so the degree distribution of a node can reflect the importance of a node in the entire heterogeneous graph, and the shortest path of a node can reflect the distance and correlation between nodes. The shorter the path between two nodes, the stronger the closeness between the two nodes. The clustering coefficient can reflect the degree of aggregation of nodes in the local network. The higher the clustering coefficient of a node, the stronger the group attribute of the node.

[0082] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. A credit gang anti-fraud method based on heterogeneous graph, characterized in that: include: Collect multiple customer data and build a customer heterogeneous graph based on the customer data; Perform graph analysis on the customer heterogeneous graph and extract multiple features of the customer heterogeneous graph; Creating a customer heterogeneous model, inputting multiple features of the heterogeneous graph as a training set into the customer heterogeneous model to train the customer heterogeneous model, and obtaining a trained customer heterogeneous model; Collect real-time customer data, input the real-time customer data into the trained customer heterogeneous model, and obtain customer risk assessment reports.

2. The anti-fraud method for credit gangs based on heterogeneous graphs according to claim 1, characterized in that: Collect multiple customer data and build a customer heterogeneous graph based on the customer data, including: Create a customer data table; Collect multiple customer data and put all collected customer data into a customer data table; the customer data table includes customer information, transaction records and contact information.

3. The anti-fraud method for credit gangs based on heterogeneous graphs according to claim 2, characterized in that: Collect multiple customer data and build a customer heterogeneous graph based on the customer data, which also includes: Define the edges and nodes of the heterogeneous graph; Customer data is imported into the nodes of the heterogeneous graph, and multiple nodes of the heterogeneous graph are connected based on the edges of the heterogeneous graph.

4. The anti-fraud method for credit gangs based on heterogeneous graphs according to claim 3 is characterized in that: The heterogeneous graph includes nodes and edges. The nodes of the heterogeneous graph include customer nodes, transaction nodes and contact information nodes. The edges of the heterogeneous graph include the connection between customer information and transaction records and the connection between customer information and contact information.

5. The anti-fraud method for credit gangs based on heterogeneous graphs according to claim 4, characterized in that: Perform graph analysis on the customer heterogeneous graph to extract multiple features of the customer heterogeneous graph, including: Randomly select a node or an edge of the heterogeneous graph from the heterogeneous graph; record the selected node or edge of the heterogeneous graph as a heterogeneous graph target; Extract features of heterogeneous graph targets; Returns a randomly selected node or edge of a heterogeneous graph until the features of all nodes or edges of the heterogeneous graph are obtained.

6. The anti-fraud method for credit gangs based on heterogeneous graphs according to claim 5, characterized in that: Create a customer heterogeneous model and input multiple features of the heterogeneous graph as a training set into the customer heterogeneous model to train the customer heterogeneous model. The trained customer heterogeneous model includes: Create a customer heterogeneous model; Normalize the feature vectors corresponding to the features of the heterogeneous graph, and divide the normalized feature vectors into training sets and test sets in random proportions; Inputting the training set into the customer heterogeneous model to train the customer heterogeneous model and obtain a trained customer heterogeneous model; The test set is input into the trained customer heterogeneous model to determine whether the trained customer heterogeneous model is complete.

7. The anti-fraud method for credit gangs based on heterogeneous graphs according to claim 6, characterized in that: Collect real-time customer data, input it into the trained customer heterogeneous model, and obtain a customer risk assessment report, including: Collect real-time customer data; Build a customer heterogeneous graph based on real-time customer data; Input the heterogeneous graph into the trained customer heterogeneous model and obtain the customer risk assessment report output by the trained heterogeneous graph.

8. The anti-fraud method for credit gangs based on heterogeneous graphs according to claim 7, characterized in that: The heterogeneous graph is based on a graph database, which includes JanusGraph and Neo4j.

9. The anti-fraud method for credit gangs based on heterogeneous graphs according to claim 8, characterized in that: The model structure of the customer heterogeneous model is set to an Encoder-Decoder architecture.

10. The anti-fraud method for credit gangs based on heterogeneous graphs according to claim 9, characterized in that: The characteristics of the heterogeneous graph target include degree distribution of nodes, shortest paths of nodes, and clustering coefficients.

Images