Customer grading and classifying method, device and equipment and storage medium

Abstract

The invention discloses a client grading and classifying method and device, equipment and a storage medium, and relates to the technical field of finance, and the method comprises the steps: obtaining a transaction relation data set, and extracting the transaction structure features of all clients according to the transaction relation data set; obtaining the singleton feature of each customer, and splicing the transaction structure feature of each customer with the singleton feature corresponding to each customer to obtain the original feature of each customer; obtaining customer investment relation data; splicing the original features of the target customer and the investment relation features of the target customer to obtain final features of the target customer; and inputting the final features of the target customer into a preset extreme gradient lifting model to obtain a grading classification result of the target customer. According to the invention, the accuracy of grading and classifying public customers by banks can be improved.

Description

Technical Field

[0001] The present application relates to the field of financial technology, and in particular to a method, device, equipment and storage medium for grading and classifying customers. Background Art

[0002] When classifying corporate bank customers, traditional data mining and machine learning technologies are usually used to achieve refined management. These methods usually use static attribute data of customers, such as financial statements, operating status reports, and credit ratings, to classify and evaluate customers by building different algorithm models, such as decision trees, support vector machines, random forests, and logistic regression. However, existing corporate bank customer classification technologies usually ignore the importance of mutual influence between customers. For example, the risk and credit status of an enterprise depends not only on its own financial status, but also on the related enterprises in its investment relationship and transaction relationship graph. This mutual influence is often ignored in traditional data mining and machine learning methods, resulting in limited prediction accuracy and stability of the model. Therefore, there is a problem of low accuracy in the classification of corporate bank customers.

[0003] The above contents are only used to assist in understanding the technical solution of the present application and do not constitute an admission that the above contents are prior art. Summary of the invention

[0004] The main purpose of this application is to provide a customer grading and classification method, device, equipment and storage medium, aiming to solve the technical problem of low accuracy in grading and classification of corporate customers of banks.

[0005] To achieve the above objectives, the present application proposes a customer grading and classification method, which includes: Acquire a transaction relationship data set, and extract transaction structure features of each customer according to the transaction relationship data set; Obtain the singleton features of each customer, and respectively concatenate the transaction structure features of each customer with the singleton features corresponding to each customer to obtain the original features of each customer, wherein the singleton features are the customer's business information, business features, transaction flow information, and transaction flow features; Acquire customer investment relationship data, and determine the investment relationship characteristics of each customer based on the original characteristics of each customer and the customer investment relationship data; Extracting the original characteristics of the target customer and the investment relationship characteristics of the target customer from the original characteristics of each customer and the investment relationship characteristics of each customer, respectively, and concatenating the original characteristics of the target customer and the investment relationship characteristics of the target customer to obtain the final characteristics of the target customer; The final features of the target customers are input into a preset extreme gradient boosting model to obtain the hierarchical classification results of the target customers.

[0006] In one embodiment, the step of extracting the transaction structure features of each customer according to the transaction relationship data set includes: Acquire a plurality of preset network structures, wherein the preset network structures represent the position and role of the customer in the transaction network and the interaction mode between the customer and other customers; Traversing all the transaction relationship data in the transaction relationship data set, matching the transaction relationship data corresponding to each customer with each of the preset network structures, and counting the number of successful matches between the transaction relationship data corresponding to each customer and each of the preset network structures; Filtering a target network structure from each of the preset network structures according to the number of successful matches between the transaction relationship data corresponding to each customer and each of the preset network structures; The transaction structure characteristics of each customer are determined according to the transaction relationship data corresponding to each customer and the number of successful matches of each preset network structure and the target network structure.

[0007] In one embodiment, the step of selecting a target network structure from each of the preset network structures according to the number of successful matches between the transaction relationship data corresponding to each customer and each of the preset network structures comprises: Acquire a customer classification label data set, and calculate the information value of each of the preset network structures based on the number of successful matches between the transaction relationship data corresponding to each customer and each of the preset network structures and the preset customer classification label data set; According to the information value of each of the preset network structures, a target network structure whose information value is within a preset range is screened out from the preset network structures.

[0008] In one embodiment, the step of determining the investment relationship characteristics of each customer based on the original characteristics of each customer and the customer investment relationship data includes: Generate an investment relationship graph according to each of the customer investment relationship data; The original features of each customer are used as node features of a preset graph sampling aggregation model, the investment relationship graph is used as the graph structure of the preset graph sampling aggregation model, the embedding vector of each node is calculated based on the preset graph sampling aggregation model, and the embedding vector of each node is used as the investment relationship feature of the customer corresponding to each node.

[0009] In one embodiment, the step of generating an investment relationship graph according to the customer investment relationship data includes: Determine the shareholding relationship between enterprises and enterprises held by the same natural person based on the customer investment relationship data; Take each customer as a node, add reverse edges between the corresponding customers of the enterprises with the above-mentioned shareholding relationship, and add bidirectional edges between the corresponding customers of the enterprises held by the same natural person, and generate an investment relationship graph.

[0010] In one embodiment, the customer classification method further includes: Obtaining a customer original feature training set and a preset investment relationship graph corresponding to each customer in the customer original feature training set, wherein the customer original feature training set includes a plurality of customer original feature data, and the preset investment relationship graph represents the investment relationship between each customer in the customer original feature training set; Acquire a first customer classification label data set, wherein the first customer classification label data set includes a plurality of customer classification label data, and the customer classification label data corresponds one-to-one to each customer in the customer original feature training set; The original feature data of each customer is used as the node feature of the graph sampling aggregation model to be trained, the preset investment relationship graph is used as the graph structure of the graph sampling aggregation model to be trained, and the first customer classification label data set is used as the training label of the graph sampling aggregation model to be trained. The graph sampling aggregation model to be trained is trained to obtain the preset graph sampling aggregation model for predicting the embedding vector of each node.

[0011] In one embodiment, the customer classification method further includes: Obtaining a customer final feature training set, wherein the customer final feature training set includes multiple customer final sample data; Acquire a second customer classification label data set, wherein the second customer classification label data set includes a plurality of customer classification label data, and the customer classification label data corresponds one-to-one to each customer in the customer final feature training set; The final sample data of each customer is used as the input data of the extreme gradient boosting model to be trained, and the second customer classification label data set is used as the training label of the extreme gradient boosting model to be trained. The extreme gradient boosting model to be trained is trained to obtain a preset extreme gradient boosting model for predicting customer classification.

[0012] In addition, to achieve the above purpose, the present application also proposes a customer grading and classification device, the customer grading and classification device comprising: A transaction module, used to obtain a transaction relationship data set, and extract transaction structure features of each customer based on the transaction relationship data set; The first splicing module is used to obtain the singleton features of each customer, and splice the transaction structure features of each customer with the singleton features corresponding to each customer to obtain the original features of each customer, wherein the singleton features are the customer's business information, business characteristics, transaction flow information and transaction flow characteristics; An investment module, used to obtain customer investment relationship data, and determine the investment relationship characteristics of each customer based on the original characteristics of each customer and the customer investment relationship data; A second splicing module is used to extract the original characteristics of the target customer and the investment relationship characteristics of the target customer from the original characteristics of each customer and the investment relationship characteristics of each customer, and splice the original characteristics of the target customer and the investment relationship characteristics of the target customer to obtain the final characteristics of the target customer; The grading module is used to input the final features of the target customers into a preset extreme gradient boosting model to obtain the grading and classification results of the target customers.

[0013] In addition, to achieve the above-mentioned purpose, the present application also proposes a customer grading and classification device, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, and the computer program is configured to implement the steps of the customer grading and classification method as described above.

[0014] In addition, to achieve the above-mentioned purpose, the present application also proposes a storage medium, which is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the customer grading and classification method described above are implemented.

[0015] In addition, to achieve the above-mentioned purpose, the present application also provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, the steps of the customer grading and classification method described above are implemented.

[0016] This application obtains a transaction relationship data set, extracts the transaction structure characteristics of each customer based on the transaction relationship data set; obtains the singleton characteristics of each customer, and splices the transaction structure characteristics of each customer with the singleton characteristics corresponding to each customer to obtain the original characteristics of each customer, where the singleton characteristics are the customer's industrial and commercial information, industrial and commercial characteristics, transaction flow information and transaction flow characteristics; obtains customer investment relationship data, and determines the investment relationship characteristics of each customer based on the original characteristics of each customer and the customer investment relationship data; extracts the original characteristics of the target customer and the investment relationship characteristics of the target customer from the original characteristics of each customer and the investment relationship characteristics of each customer, and splices the original characteristics of the target customer and the investment relationship characteristics of the target customer to obtain the final characteristics of the target customer; inputs the final characteristics of the target customer into the preset extreme gradient boosting model to obtain the grading and classification results of the target customer. This application comprehensively evaluates the customer's situation by comprehensively utilizing the customer's transaction behavior, basic information and investment relationship, and more accurately captures the customer's risk status and value potential based on multi-dimensional data, thereby providing strong support for customer grading and classification, so as to improve the accuracy of the bank's grading and classification of corporate customers. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0018] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0019] Figure 1 A flowchart of the first embodiment of the method for grading and classifying customers in this application is provided; Figure 2 Schematic diagram of 33 network structures in this application; Figure 3 A brief flow chart of the customer grading and classification method provided for this application; Figure 4 This is a schematic diagram of the module structure of the customer grading and classification device according to an embodiment of the present application; Figure 5 Schematic diagram of the device structure of the hardware operating environment involved in the customer grading and classification method in the embodiment of the present application.

[0020] The purpose, features and advantages of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0021] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application and are not used to limit the present application.

[0022] In order to better understand the technical solution of the present application, a detailed description will be given below in conjunction with the accompanying drawings and specific implementation methods.

[0023] Based on this, the present application embodiment provides a method for grading and classifying customers. Figure 1 , Figure 1 This is a flow chart of the first embodiment of the customer grading and classification method of the present application.

[0024] In this embodiment, the customer classification method includes steps S10 to S50: Step S10, obtaining a transaction relationship data set, and extracting transaction structure features of each customer according to the transaction relationship data set; It should be noted that the transaction relationship dataset includes internal bank data, such as account transaction records, transfer records, loan records, etc., and can also include external data, such as data from third-party payment platforms, data from partners, etc. Then, a transaction network can be constructed, with each customer as a node and each transaction as an edge, connecting the nodes of both parties. Edges can be directed (indicating the flow of funds) or undirected (indicating only the transaction relationship), and can be accompanied by weights (such as transaction amounts). According to business needs and data characteristics, a series of network structures can be defined based on the constructed transaction network. For each customer, count the number of various network structures that appear in the network. For example, count the number of times a customer is a "fund distribution" node, or the number of times it participates in a "one-way closed loop of funds". Calculate the information value (IV value) of each network structure, select features with IV values ​​greater than or equal to 0.3 as target transaction structures, and then extract the transaction structure features of each customer based on the target transaction structure.

[0025] It is understandable that transaction structure features that reflect the customer's behavior patterns and structural characteristics in the network can be extracted from the transaction relationship dataset. These features not only provide detailed information about the customer's transaction behavior, but also reveal the customer's position and role in the network. By calculating the IV value and screening out the most predictive features, the accuracy and stability of the customer classification model can be further improved.

[0026] Step S20, obtaining the singleton features of each customer, and respectively concatenating the transaction structure features of each customer with the singleton features corresponding to each customer, to obtain the original features of each customer, wherein the singleton features are the customer's business information, business features, transaction flow information, and transaction flow features; It should be noted that singleton features are independent attributes and features directly related to each customer. These features are usually static or periodic and do not involve interactive information with other customers. For example, internal bank data includes basic customer information, account information, historical transaction records, etc. It can also be external data, including business registration information, credit rating reports, data from third-party data providers, etc. The transaction structure features of each customer are spliced ​​with their corresponding singleton features to form a comprehensive feature vector, that is, the original features of the customer.

[0027] It is understandable that the original features of each customer are formed by splicing the customer's transaction structure features with the singleton features. The comprehensive features not only include the customer's own static attributes, but also reflect its behavior patterns and structural characteristics in the network, thus providing more comprehensive and rich information for subsequent customer classification. This step is the key to building a high-quality feature vector and is crucial to improving the accuracy and robustness of the model.

[0028] Step S30, acquiring customer investment relationship data, and determining the investment relationship characteristics of each customer based on the original characteristics of each customer and the customer investment relationship data; It should be noted that customer investment relationship data includes internal data, such as customer shareholding information recorded within banks or financial institutions, investment relationships between enterprises, etc., as well as external data, such as public business registration information, annual reports of enterprises, data from third-party data providers, etc. Customer investment relationship data can be enterprise shareholding relationships, including the proportion of shares held by enterprise A in enterprise B, and the proportion of shares held by enterprise B in enterprise A (if any), natural person shareholder information, including the proportion of shares held by natural person shareholders in multiple enterprises, and other relevant information, such as the cooperation history between enterprises, jointly participated projects, etc.

[0029] Convert the investment relationship data into a graph structure to facilitate feature extraction. Treat each enterprise as a node. If necessary, natural person shareholders can also be treated as nodes. For the shareholding relationship between enterprises, add directed edges. For example, if enterprise A holds 20% of the shares of enterprise B, add an edge from A to B with a weight of 20%. At the same time, in order to capture the two-way risk and credit transmission, add a reverse edge to each shareholding relationship edge. For example, in addition to the edge from A to B, add an edge from B to A, and the weights can be the same or adjusted according to the specific situation. For enterprises held by the same natural person, add two-way edges between them. For example, if natural person C holds shares in both enterprise A and enterprise B, add two-way edges from A to B and from B to A.

[0030] Graph algorithms (such as GraphSAGE) are used to extract the investment relationship characteristics of each customer from the investment relationship graph. The original features of each customer that have been spliced ​​together are used as the initial features of the node, the constructed investment relationship graph is used as the graph structure input, and the trained GraphSAGE model is used to calculate the embedding representation of each node. These embedding representations are the customer's investment relationship characteristics. Among them, GraphSAGE is a graph neural network algorithm based on sampling and aggregation, which can generate an embedding representation (embedding) for each node. By iteratively sampling neighbor nodes and aggregating their features, GraphSAGE can capture the local structural information and global position information of the node in the graph.

[0031] It is understandable that investment relationship features that can reflect the customer's position, role and behavior pattern in the network can be extracted from the customer's investment relationship data. These features not only provide detailed information about the customer's investment relationship, but also reveal the customer's potential risk and credit transmission path in the network. By combining the customer's original features and investment relationship features, a more comprehensive and rich feature representation can be formed, thereby improving the accuracy and stability of the customer classification model.

[0032] Step S40, extracting the original characteristics of the target customer and the investment relationship characteristics of the target customer from the original characteristics of each customer and the investment relationship characteristics of each customer, respectively, and concatenating the original characteristics of the target customer and the investment relationship characteristics of the target customer to obtain the final characteristics of the target customer; It should be noted that in the previous step, the original characteristics and investment relationship characteristics of each customer are obtained, and the original characteristics and investment relationship characteristics of the corresponding customer are spliced, that is, the original characteristics and investment relationship characteristics of the target customer are spliced ​​to form a comprehensive feature vector, that is, the final characteristics of the target customer.

[0033] It is understandable that the characteristics of specific target customers are extracted from the original characteristics and investment relationship characteristics of all customers and spliced ​​into a comprehensive feature vector. This comprehensive feature representation not only includes the static attributes and transaction behavior patterns of the customer itself, but also reflects its investment relationship and position characteristics in the network, thus providing more comprehensive and rich information for subsequent customer classification.

[0034] Step S50, inputting the final features of the target customers into a preset extreme gradient boosting model to obtain the hierarchical classification results of the target customers.

[0035] It should be noted that the preset extreme gradient boosting model is the XGBoost model.

[0036] In a feasible implementation manner, the customer classification method further includes steps C10 to C30: Step C10, obtaining a customer final feature training set, wherein the customer final feature training set includes a plurality of customer final sample data; Step C20, obtaining a second customer classification label data set, wherein the second customer classification label data set includes a plurality of customer classification label data, and the customer classification label data corresponds one-to-one to each customer in the customer final feature training set; Step C30, using the final sample data of each customer as the input data of the extreme gradient boosting model to be trained, using the second customer classification label data set as the training label of the extreme gradient boosting model to be trained, training the extreme gradient boosting model to be trained, and obtaining a preset extreme gradient boosting model for predicting customer classification.

[0037] This embodiment obtains a transaction relationship data set, extracts the transaction structure characteristics of each customer based on the transaction relationship data set; obtains the singleton characteristics of each customer, and respectively splices the transaction structure characteristics of each customer with the singleton characteristics corresponding to each customer to obtain the original characteristics of each customer, wherein the singleton characteristics are the customer's industrial and commercial information, industrial and commercial characteristics, transaction flow information and transaction flow characteristics; obtains customer investment relationship data, and determines the investment relationship characteristics of each customer based on the original characteristics of each customer and the customer investment relationship data; extracts the original characteristics of the target customer and the investment relationship characteristics of the target customer from the original characteristics of each customer and the investment relationship characteristics of each customer, and splices the original characteristics of the target customer and the investment relationship characteristics of the target customer to obtain the final characteristics of the target customer; inputs the final characteristics of the target customer into the preset extreme gradient boosting model to obtain the grading and classification results of the target customer. This application comprehensively evaluates the customer's situation by comprehensively utilizing the customer's transaction behavior, basic information and investment relationship, and more accurately captures the customer's risk status and value potential based on multi-dimensional data, thereby providing strong support for customer grading and classification, so as to improve the accuracy of the bank's grading and classification of corporate customers.

[0038] Based on the first embodiment of the present application, in the second embodiment of the present application, the same or similar contents as those in the above-mentioned embodiment 1 can be referred to the above introduction, and will not be repeated in the following. On this basis, step S10 includes steps A10~A40: Step A10, obtaining a plurality of preset network structures, wherein the preset network structures represent the position and role of the customer in the transaction network and the interaction mode between the customer and other customers; It should be noted that the preset network structures are representative network structure patterns that can reflect the different behavioral characteristics of customers in the network. Each preset network structure can accurately reflect the customer's position, role, and interaction with other customers in the network. The preset network structure can describe the relative position of the customer in the network, such as the central node, edge node, intermediary node, etc., and describe the customer's function or role in the network, such as the fund initiator, fund recipient, intermediary node, etc., and describe the transaction mode between customers, such as one-way transaction, two-way transaction, closed-loop transaction, etc.

[0039] Preset network structure reference Figure 2 There are 33 in total, which are represented by the serial number: 1. First-order fund chain payer, 2. First-order fund chain payee, 3. Second-order fund chain payee, 4. Second-order fund chain intermediary, 5. Second-order fund chain payer, 6. Fund distribution chain core node, 7. Fund distribution chain edge node, 8. Fund collection chain core node, 9. Fund collection chain edge node, 10. One-way closed-loop fund chain participant, 11. First and second-order mixed fund chain payer, 12. First and second-order mixed fund chain payee, 13. First and second-order mixed fund chain intermediary, 14. Same-account transfer participant, 15. Second-order fund chain payee with return to intermediary, 16. Second-order fund chain intermediary with return to intermediary, 17. Second-order fund chain payer with return to intermediary, 18. Second-order fund chain payer with return to payer, 19. Second-order fund chain intermediary with return to payer, 20. .The second-order fund chain payee who has paid back the payee, 21. The core node of the distribution and collection mixed fund chain, 22. The edge node of the distribution and collection mixed fund chain, 23. The edge nodes have the core nodes of the fund collection chain that trade with each other, 24. The edge nodes have the edge nodes of the fund collection chain that trade with each other, 25. The edge nodes have the core nodes of the fund distribution chain that trade with each other, 26. The edge nodes have the edge nodes of the fund distribution chain that trade with each other, 27. The one-way closed-loop fund chain that has paid back once and is paid back, 28. The one-way closed-loop fund chain that has paid back once and is not involved in the payback, 29. The payback party of the one-way closed-loop fund chain that has paid back once, 30. The one-way closed-loop fund chain that has paid back twice and is the payback party once, 31. The one-way closed-loop fund chain that has paid back twice and is the payee twice, 32. The one-way closed-loop fund chain that has paid back twice and is the payee once, 33. The participants of the two-way closed-loop fund chain.

[0040] It is understandable that multiple preset network structures are defined and obtained, which can characterize the customer's position, role and interaction with other customers in the transaction network. These preset network structures not only provide a detailed description of the customer's transaction behavior, but also provide a basis for subsequent feature extraction and customer classification. By carefully designing and selecting the preset network structure, it can be ensured that the model can capture the key behavioral characteristics of customers in the network, thereby improving the accuracy and effectiveness of the model.

[0041] Step A20, traversing all transaction relationship data in the transaction relationship data set, matching the transaction relationship data corresponding to each customer with each preset network structure, and counting the number of successful matches between the transaction relationship data corresponding to each customer and each preset network structure; It should be noted that all transaction records are read from the transaction relationship data set, and the transaction relationship data of each customer is matched with the preset network structure to determine whether it conforms to a certain specific transaction mode. According to the characteristics of the preset network structure, the corresponding matching algorithm is designed. Common graph matching algorithms include subgraph isomorphism matching (SubgraphIsomorphism), frequent subgraph mining (Frequent Subgraph Mining), etc. For each preset network structure, we need to implement the corresponding matching logic. For example, for the "fund distribution" mode, you can check whether a customer has multiple different recipients.

[0042] Then, the number of times the transaction relationship data corresponding to each customer successfully matches each preset network structure is counted, a counter is initialized for each customer and each preset network structure, the transaction records are traversed one by one, the matching algorithm is applied, the counter is updated, and the matching results of each customer are stored for subsequent processing.

[0043] It is understandable that the transaction relationship data set is systematically traversed, and the transaction relationship data of each customer is matched with the preset network structure, and the number of successful matches between the transaction relationship data corresponding to each customer and each preset network structure is counted. This process not only provides a detailed description of the customer's transaction behavior, but also provides a basis for subsequent feature extraction and customer classification. By carefully designing and implementing matching logic, it can be ensured that the model can capture the key behavioral characteristics of customers in the network, thereby improving the accuracy and effectiveness of the model.

[0044] In a feasible implementation manner, step A20 includes steps A21-A22: Step A21, obtaining a customer classification label data set, and calculating the information value of each preset network structure based on the number of successful matches between the transaction relationship data corresponding to each customer and each preset network structure and the preset customer classification label data set; It should be noted that the customer's hierarchical label data is collected from the existing business system or database as the customer hierarchical label data set. These labels are usually determined based on historical data, expert evaluation or other criteria. Based on the number of successful matches between the transaction relationship data corresponding to each customer and each preset network structure, and the preset customer hierarchical label data set, the information value (IV value) of each preset network structure is calculated.

[0045] It is understandable that the information value (IV value) of each preset network structure is systematically calculated to provide a basis for the subsequent screening of the target network structure. This step not only provides a detailed description of the customer's transaction behavior, but also provides a basis for subsequent feature extraction and customer classification. By carefully designing and implementing the IV value calculation logic, it can be ensured that the model can capture the key behavioral characteristics of customers in the network, thereby improving the accuracy and effectiveness of the model.

[0046] Step A22, selecting a target network structure whose information value is within a preset range from each preset network structure according to the information value of each preset network structure.

[0047] It should be noted that, based on business needs and experience, a lower limit and an upper limit of the IV value are set. For example, the IV value can be set between 0.3 and 0.5. According to the preset IV value range, the target network structure that meets the conditions is screened out, and the preset network structure with the IV value within the preset range is selected as the target network structure.

[0048] It can be understood that according to the information value (IV value) of each preset network structure, the target network structure whose information value is within the preset range is screened out. This process not only provides a detailed description of the customer's transaction behavior, but also provides a basis for subsequent feature extraction and customer classification. By carefully designing and implementing IV value screening logic, it can be ensured that the model can capture the key behavioral characteristics of customers in the network, thereby improving the accuracy and effectiveness of the model.

[0049] This implementation method calculates the information value (IV value) of each preset network structure based on the number of successful matches between the transaction relationship data corresponding to each customer and the preset network structure, combined with the customer classification label data set, and screens out the target network structure whose information value is within the preset range. This process not only provides a detailed description of the customer's transaction behavior, but also provides a basis for subsequent feature extraction and customer classification. By carefully designing and implementing the IV value calculation logic, it can be ensured that the model can capture the key behavioral characteristics of customers in the network, thereby improving the accuracy and effectiveness of the model.

[0050] Step A30, selecting a target network structure from each preset network structure according to the number of successful matches between the transaction relationship data corresponding to each customer and each preset network structure; It should be noted that the number of successful matches between each customer and each preset network structure counted in step A20 is summarized to obtain a customer classification label data set, that is, the actual classification label of each customer (such as high credit, medium credit, low credit, etc.). The information value (IV value) of each preset network structure is calculated to evaluate its predictive ability for customer classification, and the target network structure with higher information value is screened out according to the calculated IV value.

[0051] It can be understood that according to the number of successful matches between the transaction relationship data corresponding to each customer and the preset network structure, the information value (IV value) of each preset network structure is calculated, and the target network structure with high information value is screened out. This process not only provides a detailed description of the customer's transaction behavior, but also provides a basis for subsequent feature extraction and customer classification. By carefully designing and implementing the IV value calculation logic, it can be ensured that the model can capture the key behavioral characteristics of customers in the network, thereby improving the accuracy and effectiveness of the model.

[0052] Step A40, determining the transaction structure characteristics of each customer according to the number of successful matches between the transaction relationship data corresponding to each customer and each preset network structure and the target network structure.

[0053] It should be noted that for each customer, a feature vector containing the number of successful matches between the customer and the target network structure is created. If necessary, the feature vector is standardized or normalized to ensure that features of different scales do not affect model training.

[0054] It can be understood that, according to the number of successful matches between each customer and the target network structure, the transaction structure feature vector of each customer is constructed and integrated with other features to form the final feature vector. This process not only provides a detailed description of the customer's transaction behavior, but also provides a basis for subsequent customer classification. By carefully designing and implementing the feature construction logic, it can be ensured that the model can capture the key behavioral characteristics of customers in the network, thereby improving the accuracy and effectiveness of the model.

[0055] This embodiment extracts transaction structure features that reflect the customer's position, role, and behavior pattern in the network from the transaction relationship data set, which not only provides detailed information about the customer's transaction behavior, but also reveals the customer's potential risk and importance in the network. By screening out the most representative target network structure, the accuracy and stability of the customer classification model can be further improved.

[0056] Based on the first embodiment of the present application, in the third embodiment of the present application, the same or similar contents as those in the first embodiment can be referred to the above description, and will not be described in detail later. On this basis, step S30 includes steps S31~S32: Step S31, generating an investment relationship diagram according to the investment relationship data of each customer; It should be noted that customer investment relationship data may include the identity information of investors (investors) and investees (investees), investment amount, investment time, shareholding ratio and other specific investment details or other relevant information, such as industry classification, geographical location, etc.

[0057] In the investment relationship graph, each customer will be defined as a node, and the edge is used to represent the investment relationship between customers. For each investment relationship record, we will draw an edge between the corresponding two nodes. If there is directional information in the data (i.e. who invested in whom), then the edge is directed; if there is no clear direction, the edge can be undirected. In some cases, we can also assign weights to the edges, depending on business needs. For example, the weight can be based on the investment amount, shareholding ratio, or number of investments. Based on the above content, the actual investment relationship graph is constructed.

[0058] In a feasible implementation, step S31 includes steps S311-S312: Step S311, determining the shareholding relationship between enterprises and enterprises held by the same natural person according to the customer investment relationship data; It should be noted that the shareholding relationship between enterprises includes direct shareholding and indirect shareholding. Direct shareholding includes identifying the situation where one enterprise directly holds shares of another enterprise. For example, if Company A owns 20% of the shares of Company B, then this direct shareholding relationship is recorded. Indirect shareholding includes indirect holding through other enterprises. For example, Company A holds 50% of the shares of Company C, and Company C holds 30% of the shares of Company B, then Company A indirectly holds a part of the shares of Company B. These shareholding relationships can be represented in the form of nodes and edges using graph theory methods, where each enterprise is a node and the shareholding relationship is represented by directed edges, with the direction being from the shareholder to the invested enterprise.

[0059] Step S312, taking each customer as a node, adding reverse edges between the corresponding customers of the enterprises with a shareholding relationship, and adding bidirectional edges between the corresponding customers of the enterprises held by the same natural person, to generate an investment relationship graph.

[0060] It should be noted that each customer (whether an enterprise or a natural person) is regarded as a node in the graph. Each node has a unique identifier, such as a customer ID or enterprise name. In addition to the identifier, the node can also contain other attribute information, such as the industry, size, and geographic location of the enterprise.

[0061] If enterprise A holds shares in enterprise B, a directed edge is added between A and B, with the direction from A to B. This is because the shareholding relationship is usually from the controlling party to the controlled party. The reverse edge refers to the edge from the controlled party to the controlling party. In other words, if A holds shares in B, there will not only be an edge from A to B in the graph, but also an edge from B to A. This can more comprehensively reflect the relationship between the two. If multiple enterprises are held or controlled by the same natural person, then bidirectional edges should be added between these enterprises. This means that there are two edges between each pair of related enterprises, one from A to B and the other from B to A. This bidirectional edge reflects that these enterprises are under the control of the same natural person and have a relationship of mutual connection and influence.

[0062] The investment relationship graph finally generated is a directed graph, in which each node represents a customer and the edge represents a shareholding relationship or a joint control relationship.

[0063] It is understandable that the customer investment relationship data is converted into a complex investment relationship graph, where nodes represent customers and edges represent shareholding relationships or joint control relationships. This graph structure can not only intuitively display the relationship between customers, but also provide a basis for further analysis. In this way, the structure and potential risks between enterprises can be better understood, supporting more accurate data analysis and decision-making.

[0064] The investment relationship graph constructed in this embodiment can be used for further data analysis tasks, such as assessing risks, discovering potential conflicts of interest, performing market segmentation, or providing support for financial supervision. In addition, the graph can also be used as input for more advanced algorithms such as neural networks to mine more complex patterns and relationships.

[0065] Step S32, using the original features of each customer as the node features of the preset graph sampling aggregation model, using the investment relationship graph as the graph structure of the preset graph sampling aggregation model, calculating the embedding vector of each node based on the preset graph sampling aggregation model, and using the embedding vector of each node as the investment relationship feature of the customer corresponding to each node.

[0066] It should be noted that the preset graph sampling aggregation model is the GraphSAGE model, which can update the representation of the current node through the information of neighboring nodes. The embedding vector is a low-dimensional vector that captures the position and characteristics of the node (in this case, the customer) in the network. For each node, the embedding of the node is updated according to the current embedding vector of its neighboring nodes and its own original features. After multiple iterations, the final result is a node representation that combines local structural information and global graph information. The final node embedding vector is used as the investment relationship feature of the corresponding customer.

[0067] In a feasible implementation manner, the customer classification method further includes steps B10 to B30: Step B10, obtaining a customer final feature training set, wherein the customer final feature training set includes a plurality of customer final sample data; Step B20, obtaining a second customer classification label data set, wherein the second customer classification label data set includes a plurality of customer classification label data, and the customer classification label data corresponds one-to-one to each customer in the customer final feature training set; Step B30, using the final sample data of each customer as the input data of the extreme gradient boosting model to be trained, using the second customer classification label data set as the training label of the extreme gradient boosting model to be trained, training the extreme gradient boosting model to be trained, and obtaining a preset extreme gradient boosting model for predicting customer classification.

[0068] This embodiment generates an investment relationship graph based on the original features of each customer and the customer investment relationship data, and uses a graph sampling aggregation model (such as GraphSAGE) to calculate the embedding vector of each node, thereby obtaining the investment relationship features of each customer. This process not only provides a detailed description of the customer investment relationship, but also provides a basis for subsequent customer grading and classification. By carefully designing and implementing the graph sampling aggregation model, it can be ensured that the model can capture the complex relationship features of customers in the network, thereby improving the accuracy and effectiveness of the model.

[0069] For example, to help understand the implementation process of the customer classification method obtained by combining this embodiment with the above-mentioned embodiment 1, please refer to Figure 3 , Figure 3 A brief flow chart of a customer grading and classification method is provided.

[0070] It should be noted that the above examples are only used to understand the present application and do not constitute a limitation on the customer grading and classification method of the present application. More simple transformations based on this technical concept are all within the scope of protection of the present application.

[0071] This application also provides a customer grading and classification device, please refer to Figure 4 , the customer grading and classification device comprises: A transaction module 10 is used to obtain a transaction relationship data set and extract transaction structure features of each customer based on the transaction relationship data set; The first splicing module 20 is used to obtain the single-case features of each customer, and splice the transaction structure features of each customer with the single-case features corresponding to each customer to obtain the original features of each customer, wherein the single-case features are the customer's business information, business features, transaction flow information and transaction flow features; An investment module 30, for acquiring customer investment relationship data, and determining the investment relationship characteristics of each customer according to the original characteristics of each customer and the customer investment relationship data; A second splicing module 40 is used to extract the original characteristics of the target customer and the investment relationship characteristics of the target customer from the original characteristics of each customer and the investment relationship characteristics of each customer, and splice the original characteristics of the target customer and the investment relationship characteristics of the target customer to obtain the final characteristics of the target customer; The grading module 50 is used to input the final features of the target customers into a preset extreme gradient boosting model to obtain the grading and classification results of the target customers.

[0072] The customer grading and classification device provided by the present application adopts the customer grading and classification method in the above embodiment, which can solve the technical problem of low accuracy of grading and classification of corporate customers of banks. Compared with the prior art, the beneficial effects of the customer grading and classification device provided by the present application are the same as the beneficial effects of the customer grading and classification method provided by the above embodiment, and other technical features in the customer grading and classification device are the same as the features disclosed in the above embodiment method, which will not be repeated here.

[0073] The present application provides a customer grading and classification device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the customer grading and classification method in the above-mentioned embodiment one.

[0074] Reference below Figure 5 , which shows a schematic diagram of the structure of a customer grading and classification device suitable for implementing the embodiment of the present application. The customer grading and classification device in the embodiment of the present application may include but is not limited to mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 5 The customer grading and classification device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0075] like Figure 5As shown, the customer grading and classification device may include a processing device 1001 (e.g., a central processing unit, a graphics processor, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM: Read Only Memory) 1002 or a program loaded from a storage device 1003 to a random access memory (RAM: Random Access Memory) 1004. In RAM1004, various programs and data required for the operation of the customer grading and classification device are also stored. The processing device 1001, ROM1002, and RAM1004 are connected to each other through a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Generally, the following systems can be connected to the I / O interface 1006: an input device 1007 including, for example, a touch screen, a touch pad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; an output device 1008 including, for example, a liquid crystal display (LCD: Liquid Crystal Display), a speaker, a vibrator, etc.; a storage device 1003 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 1009. The communication device 1009 can allow the customer grading and classification device to communicate with other devices wirelessly or by wire to exchange data. Although the figure shows a customer grading and classification device with various systems, it should be understood that it is not required to implement or have all the systems shown. More or fewer systems can be implemented or have alternatively.

[0076] In particular, according to the embodiments disclosed in the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device, or installed from a storage device 1003, or installed from a ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present application are executed.

[0077] The customer grading and classification device provided by the present application adopts the customer grading and classification method in the above embodiment, which can solve the technical problem of low accuracy of grading and classification of corporate customers of banks. Compared with the prior art, the beneficial effects of the customer grading and classification device provided by the present application are the same as the beneficial effects of the customer grading and classification method provided by the above embodiment, and other technical features in the customer grading and classification device are the same as the features disclosed in the method of the previous embodiment, which will not be repeated here.

[0078] It should be understood that the various parts disclosed in this application can be implemented by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0079] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0080] The present application provides a computer-readable storage medium having computer-readable program instructions (ie, computer programs) stored thereon, the computer-readable program instructions being used to execute the customer grading and classification method in the above-mentioned embodiment.

[0081] The readable storage medium provided by the present application is a computer-readable storage medium, which stores computer-readable program instructions (i.e., computer programs) for executing the above-mentioned customer grading and classification method, and can solve the technical problem of low accuracy in grading and classification of corporate customers of banks. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided by the present application are the same as the beneficial effects of the customer grading and classification method provided by the above-mentioned embodiment, and will not be elaborated here.

[0082] The present application also provides a computer program product, including a computer program, which implements the steps of the above-mentioned customer grading and classification method when executed by a processor.

[0083] The computer program product provided by this application can solve the technical problem of low accuracy in grading and classifying corporate customers of banks. Compared with the prior art, the beneficial effects of the computer program product provided by this application are the same as those of the customer grading and classification method provided by the above embodiment, which will not be elaborated here.

[0084] The above descriptions are only some embodiments of the present application, and are not intended to limit the patent scope of the present application. All equivalent structural changes made using the contents of the present application specification and drawings under the technical concept of the present application, or direct / indirect applications in other related technical fields are included in the patent protection scope of the present application.

Claims

1. A customer classification method, characterized in that: The customer classification method comprises: Acquire a transaction relationship data set, and extract transaction structure features of each customer according to the transaction relationship data set; Obtain the singleton features of each customer, and respectively concatenate the transaction structure features of each customer with the singleton features corresponding to each customer to obtain the original features of each customer, wherein the singleton features are the customer's business information, business features, transaction flow information, and transaction flow features; Acquire customer investment relationship data, and determine the investment relationship characteristics of each customer based on the original characteristics of each customer and the customer investment relationship data; Extracting the original characteristics of the target customer and the investment relationship characteristics of the target customer from the original characteristics of each customer and the investment relationship characteristics of each customer, respectively, and concatenating the original characteristics of the target customer and the investment relationship characteristics of the target customer to obtain the final characteristics of the target customer; The final features of the target customers are input into a preset extreme gradient boosting model to obtain the hierarchical classification results of the target customers.

2. The customer classification method according to claim 1, characterized in that: The step of extracting the transaction structure features of each customer according to the transaction relationship data set comprises: Acquire a plurality of preset network structures, wherein the preset network structures represent the position and role of the customer in the transaction network and the interaction mode between the customer and other customers; Traversing all the transaction relationship data in the transaction relationship data set, matching the transaction relationship data corresponding to each customer with each of the preset network structures, and counting the number of successful matches between the transaction relationship data corresponding to each customer and each of the preset network structures; Filtering a target network structure from each of the preset network structures according to the number of successful matches between the transaction relationship data corresponding to each customer and each of the preset network structures; The transaction structure characteristics of each customer are determined according to the transaction relationship data corresponding to each customer and the number of successful matches of each preset network structure and the target network structure.

3. The customer classification method according to claim 2, characterized in that: The step of selecting a target network structure from each of the preset network structures according to the number of successful matches between the transaction relationship data corresponding to each customer and each of the preset network structures comprises: Acquire a customer classification label data set, and calculate the information value of each of the preset network structures based on the number of successful matches between the transaction relationship data corresponding to each customer and each of the preset network structures and the preset customer classification label data set; According to the information value of each of the preset network structures, a target network structure whose information value is within a preset range is screened out from the preset network structures.

4. The customer classification method according to claim 1, characterized in that: The step of determining the investment relationship characteristics of each customer based on the original characteristics of each customer and the customer investment relationship data comprises: Generate an investment relationship graph according to each of the customer investment relationship data; The original features of each customer are used as node features of a preset graph sampling aggregation model, the investment relationship graph is used as the graph structure of the preset graph sampling aggregation model, the embedding vector of each node is calculated based on the preset graph sampling aggregation model, and the embedding vector of each node is used as the investment relationship feature of the customer corresponding to each node.

5. The customer classification method according to claim 4, characterized in that: The step of generating an investment relationship diagram according to the customer investment relationship data comprises: Determine the shareholding relationship between enterprises and enterprises held by the same natural person based on the customer investment relationship data; Take each customer as a node, add reverse edges between the corresponding customers of the enterprises with the above-mentioned shareholding relationship, and add bidirectional edges between the corresponding customers of the enterprises held by the same natural person, and generate an investment relationship graph.

6. The customer grading and classification method according to any one of claims 1 to 5, characterized in that: The customer classification method further includes: Obtaining a customer original feature training set and a preset investment relationship graph corresponding to each customer in the customer original feature training set, wherein the customer original feature training set includes a plurality of customer original feature data, and the preset investment relationship graph represents the investment relationship between each customer in the customer original feature training set; Acquire a first customer classification label data set, wherein the first customer classification label data set includes a plurality of customer classification label data, and the customer classification label data corresponds one-to-one to each customer in the customer original feature training set; The original feature data of each customer is used as the node feature of the graph sampling aggregation model to be trained, the preset investment relationship graph is used as the graph structure of the graph sampling aggregation model to be trained, and the first customer classification label data set is used as the training label of the graph sampling aggregation model to be trained. The graph sampling aggregation model to be trained is trained to obtain the preset graph sampling aggregation model for predicting the embedding vector of each node.

7. The customer grading and classification method according to any one of claims 1 to 5, characterized in that: The customer classification method further includes: Obtaining a customer final feature training set, wherein the customer final feature training set includes multiple customer final sample data; Acquire a second customer classification label data set, wherein the second customer classification label data set includes a plurality of customer classification label data, and the customer classification label data corresponds one-to-one to each customer in the customer final feature training set; The final sample data of each customer is used as the input data of the extreme gradient boosting model to be trained, and the second customer classification label data set is used as the training label of the extreme gradient boosting model to be trained. The extreme gradient boosting model to be trained is trained to obtain a preset extreme gradient boosting model for predicting customer classification.

8. A customer grading and classification device, characterized in that: The customer grading and classification device comprises: A transaction module, used to obtain a transaction relationship data set, and extract transaction structure features of each customer based on the transaction relationship data set; The first splicing module is used to obtain the singleton features of each customer, and splice the transaction structure features of each customer with the singleton features corresponding to each customer to obtain the original features of each customer, wherein the singleton features are the customer's business information, business characteristics, transaction flow information and transaction flow characteristics; An investment module, used to obtain customer investment relationship data, and determine the investment relationship characteristics of each customer based on the original characteristics of each customer and the customer investment relationship data; A second splicing module is used to extract the original characteristics of the target customer and the investment relationship characteristics of the target customer from the original characteristics of each customer and the investment relationship characteristics of each customer, and splice the original characteristics of the target customer and the investment relationship characteristics of the target customer to obtain the final characteristics of the target customer; The grading module is used to input the final features of the target customers into a preset extreme gradient boosting model to obtain the grading and classification results of the target customers.

9. A customer grading and classification device, characterized in that: The device comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the customer grading and classification method according to any one of claims 1 to 7.

10. A storage medium, characterized in that: The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the customer grading and classification method according to any one of claims 1 to 7 are implemented.

Images