Credit evaluation model generation method, credit evaluation method and device

Abstract

The invention belongs to the technical field of data analysis, and particularly relates to a credit evaluation model generation method and device, and a credit evaluation method and device, and the generation method comprises the steps: obtaining a customer social side table, a customer transaction side table, a customer node table, and customer information sampled in different business scenes; according to the client social side table, the client transaction side table and the client node table, generating a client association relationship graph; generating a credit assessment data set according to the customer information sampled in the different business scenes and the customer association relationship graph; and training a graph neural network model according to the credit evaluation data set to obtain a credit evaluation model. According to the method, the problems of low interpretability, insufficient universality and poor credit evaluation effect of a credit evaluation model are solved.

Description

Technical Field

[0001] The present invention relates to the field of data analysis technology, and in particular to a credit assessment model generation method, a credit assessment method and a device. Background Art

[0002] In the financial field, credit assessment can help financial institutions effectively identify and control credit risks; in social management, credit assessment can be used as a reference for the issuance of subsidies and qualification certification; in human resource management, credit assessment can help employers evaluate job candidates; therefore, credit assessment is of great significance in many fields.

[0003] Existing technologies usually use machine learning algorithms to establish credit assessment models, which are unable to capture the impact of the deep-level correlation between the target to be assessed and the surrounding nodes, thus affecting the assessment effect; or they establish credit assessment models through deep learning algorithms, which have poor interpretability of the assessment results; in addition, existing technologies construct different credit assessment models according to different business stage scenarios, which are only applicable to corresponding business scenarios and cannot have good predictive capabilities in different scenarios throughout the entire business life cycle. Summary of the invention

[0004] The present invention provides a credit assessment model generation method, a credit assessment method and a device, which are used to solve the problems of low interpretability, insufficient versatility and poor credit assessment effect of the credit assessment model.

[0005] In order to solve the above technical problems, a first aspect of this document provides a method for generating a credit assessment model, the method comprising:

[0006] Obtaining a customer social edge table, a customer transaction edge table, a customer node table, and sampled customer information in different business scenarios, wherein the customer social edge table includes the customer's social relationship data, the customer transaction edge table includes the customer's transaction data, and the customer node table includes the customer node information, the credit score of the customer by the pre-trained machine learning model, and statistical feature information;

[0007] Generate a customer association relationship graph based on the customer social edge table, the customer transaction edge table and the customer node table;

[0008] Generate a credit assessment data set based on the sampled customer information in the different business scenarios and the customer association graph;

[0009] A graph neural network model is trained according to the credit assessment data set to obtain a credit assessment model.

[0010] Furthermore, the process of generating the customer social edge table includes:

[0011] Constructing a social directed graph based on the social relationship data of the customer;

[0012] Reversing the source node and the target node in the social directed graph to obtain a social reverse directed graph;

[0013] The social directed graph and the social inverse directed graph are combined to generate a customer social edge table.

[0014] Furthermore, the process of generating the customer transaction side table includes:

[0015] Determine the transferor and the recipient of each transaction based on the transaction data of the customer;

[0016] Constructing a transaction directed graph according to preset relationship features and transaction data between the transferor and the payee; wherein the preset relationship features include transaction relationship features, relationship type features, and relationship strength;

[0017] Exchanging the transferor and the payee in each transaction to obtain an exchanged transferor and payee;

[0018] Constructing a transaction reversal directed graph according to the preset relationship characteristics, the transferor and the payee after the exchange;

[0019] The transaction directed graph and the transaction reversal directed graph are combined to generate a customer transaction edge table.

[0020] Furthermore, the preset relationship characteristics include preset relationship characteristics of transactional relationships and preset relationship characteristics of non-transactional relationships;

[0021] Determining the relationship strength of the transaction relationship according to the transaction relationship characteristics of the transaction relationship; the transaction relationship characteristics of the transaction relationship include the transaction amount;

[0022] The relationship strength of the non-transactional relationship is determined based on the relationship type characteristics of the non-transactional relationship; the relationship type characteristics of the non-transactional relationship include quantified intimacy characteristics.

[0023] Furthermore, a credit assessment data set is generated based on the sampled customer information under the different business scenarios and the customer association relationship map, including:

[0024] Remove duplicates from sampled customer information in different business scenarios to obtain a customer information set;

[0025] Comparing customer information sets of different time periods to determine overlapping customer information; the overlapping customer information refers to customer information commonly included in the customer information sets of different time periods;

[0026] Extracting characteristic data of customer information in customer information sets of different time periods from the customer association relationship map;

[0027] The feature data and credit labels corresponding to the customer information identical to the overlapping customer information in the customer information sets of different time periods are classified into the training set corresponding to the current time period; the feature data and credit labels corresponding to the customer information different from the overlapping customer information in the customer information sets of different time periods are partially classified into the training set corresponding to the current time period;

[0028] Based on the training sets of different time periods, multiple credit assessment data sets are generated in sequence; the multiple credit assessment data sets are used to train the graph neural network model in sequence.

[0029] Furthermore, the embedding features of each layer in the graph neural network model are expressed according to the following formula:

[0030]

[0031]

[0032]

[0033] Among them, i represents node i, j represents node j, k represents node k, and N(i) represents the set of neighbor nodes of node i in the Nth layer of the graph neural network model. Represents the embedded features of node i in the N+1th layer of the graph neural network model, represents the embedding feature of node i in the Nth layer of the graph neural network model, represents the embedding feature of node j in the Nth layer of the graph neural network model, σ represents the activation function, and α ij represents the attention weight of node j to node i, b represents the bias, W1, W2, W self , W neigh , W neigh2 , W edge Both represent parameter matrices, score ij represents the attention score of node j on node i, score ik represents the attention score of node k on node i, e ij represents the relationship feature between node i and node j, and |||| represents the concatenation operation.

[0034] Furthermore, the business scenarios include: pre-loan application scenario, loan use scenario and loan withdrawal scenario;

[0035] The graph neural network model is trained using the following total loss function:

[0036] Loss = α app L app +α loan L loan +α balance Lbalance ;

[0037] in,

[0038]

[0039]

[0040] L app represents the loss function in the pre-loan application scenario, L loan represents the loss function in the scenario of using credit in loan, L balance represents the loss function in the loan repayment scenario, N app represents the number of samples in the pre-loan application scenario, N loan The number of samples in the loan-using credit scenario, N balance The number of samples in the loan repayment scenario, L n Represents the loss value of the nth sample, W yn represents the calculation weight of the nth sample, Loss represents the total loss function of the graph neural network model, α app represents the calculation weight in the pre-loan application scenario, α loan represents the calculation weight in the scenario of using credit during loan, α balance Indicates the calculation weight in the loan repayment scenario.

[0041] A second aspect of the present invention provides a credit assessment method, the method comprising:

[0042] The characteristic data of the customer to be predicted is input into the credit assessment model generated by the credit assessment model generation method described in any of the above embodiments to obtain the credit assessment result of the customer.

[0043] A third aspect of the present invention provides a credit assessment model generation device, the device comprising:

[0044] An acquisition module, used to acquire a customer social edge table, a customer transaction edge table, a customer node table, and sampled customer information in different business scenarios, wherein the customer social edge table includes the customer's social relationship data, the customer transaction edge table includes the customer's transaction data, and the customer node table includes the customer node information, the credit score of the customer by the pre-trained machine learning model, and statistical feature information;

[0045] A graph generation module, used to generate a customer association graph based on the customer social edge table, the customer transaction edge table and the customer node table;

[0046] A data set generation module, used to generate a credit assessment data set based on the sampled customer information under the different business scenarios and the customer association relationship map;

[0047] The model generation module is used to train a graph neural network model according to the credit assessment data set to obtain a credit assessment model.

[0048] A fourth aspect of the present invention provides a credit assessment device, the device comprising:

[0049] The credit assessment module is used to input the characteristic data of the customer to be predicted into the credit assessment model generated by the credit assessment model generation method described in any of the above embodiments to obtain the credit assessment result of the customer.

[0050] The fifth aspect of this document provides a computer device, including a memory, a processor, and a computer program stored in the memory, and when the computer program is executed by the processor, it executes the instructions of the credit assessment model generation method and the credit assessment method described in any of the aforementioned embodiments.

[0051] A sixth aspect of the present invention provides a computer storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor of a computer device, the instructions of the credit assessment model generation method and the credit assessment method described in any of the aforementioned embodiments are executed.

[0052] The seventh aspect of the present invention provides a computer program product, which includes a computer program. When the computer program is executed by a processor of a computer device, it executes the instructions of the credit assessment model generation method and the credit assessment method described in any of the aforementioned embodiments.

[0053] The credit assessment model generation method, credit assessment method and device provided in this article capture the deep correlation between the target to be evaluated and the surrounding nodes by training the graph neural network model, which has better learning effect than the machine learning model; the credit score of the customer by the machine learning model is added as a feature to the training of the graph neural network model, which improves the credit assessment accuracy of the model, and machine learning has stronger interpretability than the graph neural network. The model score of machine learning is added to improve the interpretability of the graph neural network; the sampled data in different business scenarios is used as the training data of the graph neural network model, so that the trained credit assessment model has better versatility; the characteristic data of the customer to be predicted is input into the credit assessment model to obtain accurate credit assessment results.

[0054] In order to make the above and other purposes, features and advantages of this article more obvious and easy to understand, the following specifically cites preferred embodiments and describes them in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the technical solutions in the embodiments of this article or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of this article. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0056] Figure 1 A flow chart of a method for generating a credit assessment model according to an embodiment of this invention is shown;

[0057] Figure 2 A flow chart of the generation process of the customer social edge table in the embodiment of this article is shown;

[0058] Figure 3 The flowchart of the generation process of the customer transaction side table of the embodiment of this article is shown;

[0059] Figure 4 A flow chart of determining relationship strength according to an embodiment of this invention is shown;

[0060] Figure 5 The flowchart of generating a credit assessment data set according to the embodiment of this article is shown;

[0061] Figure 6 A structural diagram of a credit assessment model generating device according to an embodiment of this invention is shown;

[0062] Figure 7 A structural diagram of a computer device according to an embodiment of this invention is shown.

[0063] Description of the accompanying symbols:

[0064] 610. Acquisition module;

[0065] 620. Graph generation module;

[0066] 630, data set generation module;

[0067] 640, model generation module;

[0068] 702. Computer equipment;

[0069] 704, processor;

[0070] 706. Memory;

[0071] 708, driving mechanism;

[0072] 710, input / output module;

[0073] 712. Input devices;

[0074] 714. Output devices;

[0075] 716. Presentation equipment;

[0076] 718. Graphical user interface;

[0077] 720, network interface;

[0078] 722. Communication link;

[0079] 724. Communication bus. DETAILED DESCRIPTION

[0080] The following will be combined with the drawings in the embodiments of this article to clearly and completely describe the technical solutions in the embodiments of this article. Obviously, the described embodiments are only part of the embodiments of this article, not all of the embodiments. Based on the embodiments of this article, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this article.

[0081] It should be noted that the terms "first", "second", etc. in the specification and claims of this article and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of this article described here can be implemented in an order other than those illustrated or described here. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, device, product or equipment that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or equipment.

[0082] This specification provides method operation steps as described in the embodiments or flow charts, but more or fewer operation steps may be included based on routine or non-creative work. The order of steps listed in the embodiments is only one way of executing the order of many steps and does not represent the only execution order. When the system or device product is executed in practice, it can be executed in the order of the method shown in the embodiments or the drawings or in parallel.

[0083] It should be noted that the acquisition, storage, use, and processing of data in the technical solutions of the embodiments of this specification comply with the relevant provisions of national laws and regulations.

[0084] It should be noted that in the embodiments of the present application, certain software, components, models and other existing solutions in the industry may be mentioned, and they should be regarded as exemplary. Their purpose is only to illustrate the feasibility of implementing the technical solution of the present application, but it does not mean that the applicant has or will necessarily use the solution.

[0085] In one embodiment of the present invention, a credit assessment model generation method is provided to solve the problems of low interpretability, insufficient versatility and poor credit assessment effect of the credit assessment model.

[0086] Specifically, Figure 1 As shown, the credit assessment model generation method includes:

[0087] Step 110, obtaining a customer social edge table, a customer transaction edge table, a customer node table, and sampled customer information in different business scenarios;

[0088] The customer social edge table includes the customer's social relationship data, the customer transaction edge table includes the customer's transaction data, and the customer node table includes the customer node information, the credit score of the customer by the pre-trained machine learning model, and statistical feature information;

[0089] Step 120, generating a customer association relationship graph according to the customer social edge table, the customer transaction edge table and the customer node table;

[0090] Step 130, generating a credit assessment data set according to the sampled customer information in the different business scenarios and the customer association relationship map;

[0091] Step 140: training a graph neural network model according to the credit assessment data set to obtain a credit assessment model.

[0092] This embodiment captures the deep correlation between the target to be evaluated and the surrounding nodes by training the graph neural network model, which has a better learning effect than the machine learning model; the credit score of the customer by the machine learning model is added as a feature to the training of the graph neural network model, which improves the credit evaluation accuracy of the model. Machine learning has stronger interpretability than graph neural networks, and the model score of machine learning is added to improve the interpretability of the graph neural network; the sampled data in different business scenarios is used as the training data of the graph neural network model, so that the trained credit assessment model has better versatility.

[0093] In some embodiments of this article, the pre-trained machine learning model can be logistic regression, decision tree, xgboost, lightgbm, etc., and the graph neural network model can be GCN, GraphSAGE, GAT, etc., which is not limited in this article.

[0094] In one embodiment of this invention, Figure 2 As shown, the process of generating the customer social edge table includes:

[0095] Step 210, constructing a social directed graph according to the social relationship data of the customer;

[0096] Step 220, inverting the source node and the target node in the social directed graph to obtain a social inverted directed graph;

[0097] Step 230: Combine the social directed graph and the social inverse directed graph to generate a customer social edge table.

[0098] This embodiment constructs a social directed graph through social relationship data, and obtains a customer social edge table by combining the social inverse directed graph and the social directed graph, providing a basis for generating a customer association relationship map.

[0099] In some embodiments of the present invention, the social relationship data includes social relationships between customers, such as social relationships with spouses, parents, children, other relatives, colleagues, friends, other contacts, and social relationships between customers with the same mobile phone and the same email address.

[0100] In some embodiments of the present invention, for a social directed graph including parent-child social relationships, the directional relationship between parents and children is reversed, that is, the social relationship originally directed from parents to children is reversed to directed from children to parents, and the same is true for other social relationships, to obtain a social reversed directed graph, and then the social reversed directed graph and the social directed graph are combined to obtain a social directed graph with bidirectional directional relationships, namely, the customer social edge table.

[0101] In one embodiment of this invention, Figure 3 As shown, the process of generating the customer transaction side table includes:

[0102] Step 310, determining the transferor and the payee in each transaction according to the transaction data of the customer;

[0103] Step 320, constructing a directed transaction graph according to the preset relationship characteristics and the transaction data between the transferor and the payee;

[0104] The preset relationship characteristics include transaction relationship characteristics, relationship type characteristics, and relationship strength;

[0105] Step 330, exchanging the transferor and the payee in each transaction to obtain an exchanged transferor and payee;

[0106] Step 340, constructing a transaction reversal directed graph according to the preset relationship characteristics, the exchanged transferor and the recipient;

[0107] Step 350: Generate a customer transaction edge table by combining the transaction directed graph and the transaction reversal directed graph.

[0108] This embodiment constructs a transaction directed graph based on preset relationship features and transaction data, and in the preset relationship features, fully considers the influence of transaction relationship features, relationship type and relationship strength, and combines the transaction reversal directed graph and the transaction directed graph to obtain the customer transaction edge table, providing a basis for generating a customer association relationship map.

[0109] In some embodiments of this invention, the transaction data of a customer includes deposits and withdrawals, consumption, investment, transfers and remittances, etc. The flow direction of transaction funds is obtained based on the transaction data, the party with outflow of funds is determined as the transferor, and the party with inflow of funds is determined as the recipient. The transaction relationship characteristics include the transaction amount, and the relationship type characteristics include colleagues, customers, couples, parents, children, etc. The relationship strength is represented by a numerical value, which can be selected from 0-10, where 0 represents the weakest relationship strength and 10 represents the strongest relationship strength. There may also be other representation methods, but the strength should be reflected, which is not limited in this invention.

[0110] In some embodiments of the present invention, in the constructed transaction directed graph, A is the transferor, B is the payee, and the transaction relationship characteristic of A pointing to B is the total amount of funds transferred from A to B. In the transaction reversal directed graph, the relationship between A and B is swapped, that is, B is the transferor, A is the payee, and the transaction relationship characteristic of B pointing to A is the total amount of funds transferred from B to A.

[0111] In some embodiments of the present invention, a customer node table is also constructed based on customer node information, a pre-trained machine learning model for the customer's credit score and statistical feature information. The customer node information includes basic information of the customer, such as node ID, node type, etc., and the node type includes personal type and enterprise type; the credit score refers to the score obtained by credit scoring the customer through a pre-trained credit scoring machine learning model, and the credit scoring machine learning model is trained based on the customer's historical transaction data and historical credit status; the statistical feature information refers to the calculation of the graph features of the customer's social edge table, including the out-degree, in-degree, centrality, etc. of each node in the customer's social edge table.

[0112] In one embodiment of the present invention, the preset relationship features include preset relationship features of transaction relationships and preset relationship features of non-transaction relationships, such as Figure 4 As shown, the process of determining the relationship strength of transactional relationships and non-transactional relationships includes:

[0113] Step 410, determining the relationship strength of the transaction relationship according to the transaction relationship characteristics of the transaction relationship;

[0114] Step 420: Determine the relationship strength of the non-transactional relationship according to the relationship type characteristics of the non-transactional relationship.

[0115] Among them, the transaction relationship characteristics of the transaction relationship include the transaction amount, and the relationship type characteristics of the non-transaction relationship include the quantitative intimacy characteristics; the transaction relationship refers to the financial transaction relationship such as deposit and withdrawal, consumption, investment, transfer and remittance, and the non-transaction relationship refers to the social relationship between parents, children, individuals and work units. The quantitative intimacy characteristics are the quantitative indicators of the social relationship between parents, children, individuals and work units. According to research, the higher the quantitative indicator of the social relationship, the greater the probability of default.

[0116] This embodiment distinguishes between preset relationship features of transactional relationships and preset relationship features of non-transactional relationships, and can determine a reference basis for relationship strength according to actual conditions, thereby obtaining more reasonable relationship strength features.

[0117] In one embodiment of this invention, Figure 5 As shown, a credit assessment data set is generated based on the sampled customer information in different business scenarios and the customer association relationship map, including:

[0118] Step 510, removing duplicates from sampled customer information in different business scenarios to obtain a customer information set;

[0119] Step 520, comparing customer information sets from different time periods to determine overlapping customer information;

[0120] Step 530, extracting characteristic data of customer information in customer information sets of different time periods from the customer association relationship map;

[0121] Step 540, the feature data and credit labels corresponding to the customer information identical to the overlapping customer information in the customer information sets of different time periods are classified into the training set corresponding to the current time period; the feature data and credit labels corresponding to the customer information different from the overlapping customer information in the customer information sets of different time periods are partially classified into the training set corresponding to the current time period;

[0122] Step 550, sequentially generate multiple credit assessment data sets based on training sets of different time periods.

[0123] In the above steps, overlapping customer information refers to customer information commonly contained in the customer information sets of different time periods, and the multiple credit assessment data sets are used to train the graph neural network model in sequence.

[0124] In the above steps, the feature data includes node features, preset relationship features, and node connection relationships. The node features include transaction data and statistical feature information. The preset relationship features include transaction relationship features, relationship type features encoded by One-Hot, and relationship strength. The node connection relationship refers to the adjacency matrix between nodes in the customer association relationship graph.

[0125] In the above steps, the credit label is obtained based on the historical credit status of the customer, including two identifications: with credit risk or without credit risk. The customer's characteristic data is used as the learning data of the model, and the credit label is used as the classification standard for the model output results.

[0126] This embodiment samples customer information from different business scenarios and can learn data from different business scenarios when training the graph neural network model, thereby enhancing the versatility of the generated credit assessment model. The graph neural network model is trained by using multiple credit assessment data sets generated in different time periods, further enhancing the generalization ability of the generated credit assessment model.

[0127] In some embodiments of this document, there is overlap in customer information sampled from different business scenarios, that is, customers existing in business scenario A also exist in business scenario B. At this time, these duplicate customers need to be deduplicated to ensure that duplicate features are not learned during model training, thereby reducing training pressure.

[0128] In some embodiments of the present invention, multiple credit assessment data sets are generated according to different business scenarios in different time periods, and training sets, validation sets and test sets are divided in different time periods. For example, in time period A and time period B, in order to prevent customer information in the training set of time period A from appearing in the validation set or test set of time period B, resulting in credit label leakage, thereby affecting the evaluation result of the model, the same customer information in time period A and time period B is first counted to obtain overlapping customer information. In time period A, as long as the customer information is hit by the overlapping customer information, it is divided into the training set of time period A. In time period B, as long as the customer information is hit by the overlapping customer information, it is divided into the training set of time period B, ensuring that the customer information in the training set of time period A will not appear in the validation set or test set of time period B, thereby preventing credit label leakage; in addition, since the proportion of customer information divided in the training set is relatively large, the remaining customer information that is not hit by the overlapping customer information continues to be divided into the training set, or divided into the validation set and the test set according to the preset ratio. The preset ratio can be training set: validation set: test set = 0.6:0.2:0.2, or other ratios, which are not limited in this article.

[0129] In some embodiments of the present invention, for different time periods, such as time period A and time period B, the credit assessment dataset generated in time period A is first used to train the graph neural network model, and then the credit assessment dataset generated in time period B is used to train the graph neural network model, so that the graph neural network model generated based on time period A is fine-tuned in time period B to enhance the generalization ability for time period B. At the same time, first training the graph neural network model with the credit assessment dataset generated in time period A, and then training the graph neural network model with the credit assessment dataset generated in time period A, can reduce the learning pressure of the graph neural network model and improve the learning effect.

[0130] In some embodiments of this invention, the embedding features of each layer in the graph neural network model can be expressed by the following formula:

[0131]

[0132]

[0133]

[0134] Among them, i represents node i, j represents node j, k represents node k, and N(i) represents the set of neighbor nodes of node i in the Nth layer of the graph neural network model. Represents the embedding feature of node i in the N+1th layer of the graph neural network model, represents the embedding feature of node i in the Nth layer of the graph neural network model, represents the embedding feature of node j in the Nth layer of the graph neural network model, σ represents the activation function, and α ij represents the attention weight of node j to node i, b represents the bias, W1, W2, W self , W neigh , W neigh2 , W edge Both represent parameter matrices, score ij represents the attention score of node j on node i, score ik represents the attention score of node k on node i, e ij represents the relationship feature between node i and node j, and |||| represents the concatenation operation.

[0135] In the above formula, the relative importance of the embedded feature nodes is taken into account, and attention weights are assigned to different nodes, so that more attention is paid to nodes with greater contributions during the training process; at the same time, relationship features are also added to the calculation process, including transaction relationship features, relationship type features, relationship strength and other multi-dimensional features, which further improves the learning effect of the graph neural network model.

[0136] In some embodiments of the present invention, different business scenarios include pre-loan application scenarios, mid-loan credit use scenarios and mid-loan repayment scenarios. The pre-loan application scenario refers to a scenario in which a customer group is newly granted credit in a certain period of time. The mid-loan credit use scenario refers to a scenario in which a customer group uses the credit limit to generate a new loan note in a certain period of time. The mid-loan repayment scenario refers to a scenario in which a customer group still has a loan balance in a certain period of time.

[0137] In some embodiments of this document, the graph neural network model is trained using the following total loss function:

[0138] Loss = α app L app +α loan Lloan +α balance L balance ;

[0139] in,

[0140]

[0141]

[0142] L app represents the loss function in the pre-loan application scenario, L loan represents the loss function in the scenario of using credit in loan, L balance represents the loss function in the loan repayment scenario, N app represents the number of samples in the pre-loan application scenario, N loan The number of samples in the loan-using credit scenario, N balance The number of samples in the loan repayment scenario, L n Represents the loss value of the nth sample, W yn represents the calculation weight of the nth sample, Loss represents the total loss function of the graph neural network model, α app represents the calculation weight in the pre-loan application scenario, α loan represents the calculation weight in the scenario of using credit during loan, α balance Indicates the calculation weight in the loan repayment scenario.

[0143] In the above formula, the loss impact of different business scenarios is taken into account when calculating the total loss function to ensure that the credit assessment of each business scenario is taken into account when training the graph neural network model.

[0144] In one embodiment of the present invention, a credit assessment method is also provided to solve the problems of poor credit assessment effect and insufficient versatility. The credit assessment result of the customer to be predicted is obtained by inputting the characteristic data of the customer to be predicted into the credit assessment model. Since the credit assessment method is based on the credit assessment model generation method, the effect is similar to that of the credit assessment model generation method. The effect of the credit assessment method can refer to the credit assessment model generation method, and the repeated parts will not be repeated.

[0145] In some embodiments of the present invention, characteristic data of the customer to be predicted is extracted from the customer association relationship graph, including node characteristics, preset relationship characteristics, and node connection relationships of the customer to be predicted. The node characteristics include transaction data and statistical feature information. The preset relationship characteristics include transaction relationship characteristics, relationship type characteristics encoded by One-Hot, and relationship strength. The node connection relationship refers to the adjacency matrix between nodes in the customer association relationship graph. Afterwards, the characteristic data of the customer to be predicted is input into the credit assessment model to obtain the credit assessment result of the customer.

[0146] Based on the same inventive concept, this article also provides a credit assessment model generation device, as described in the following embodiments. Since the principle of solving the problem by the credit assessment model generation device is similar to that of the credit assessment model generation method, the implementation of the credit assessment model generation device can refer to the credit assessment model generation method, and the repeated parts will not be repeated.

[0147] Specifically, Figure 6 As shown, the credit assessment model generating device includes:

[0148] An acquisition module 610 is used to acquire a customer social edge table, a customer transaction edge table, a customer node table, and sampled customer information in different business scenarios;

[0149] The customer social edge table includes the customer's social relationship data, the customer transaction edge table includes the customer's transaction data, and the customer node table includes the customer node information, the credit score of the customer by the pre-trained machine learning model, and statistical feature information;

[0150] A graph generation module 620, configured to generate a customer association graph based on the customer social edge table, the customer transaction edge table, and the customer node table;

[0151] A data set generation module 630, configured to generate a credit assessment data set based on the sampled customer information under the different business scenarios and the customer association relationship graph;

[0152] The model generation module 640 is used to train a graph neural network model according to the credit assessment data set to obtain a credit assessment model.

[0153] This embodiment captures the deep correlation between the target to be evaluated and the surrounding nodes by training the graph neural network model, which has a better learning effect than the machine learning model; the credit score of the customer by the machine learning model is added as a feature to the training of the graph neural network model, which improves the credit evaluation accuracy of the model. Machine learning has stronger interpretability than graph neural networks, and the model score of machine learning is added to improve the interpretability of the graph neural network; the sampled data in different business scenarios is used as the training data of the graph neural network model, so that the trained credit assessment model has better versatility.

[0154] Based on the same inventive concept, this article also provides a credit evaluation device, as described in the following embodiments. Since the principle of solving the problem by the credit evaluation device is similar to that of the credit evaluation method, the implementation of the credit evaluation device can refer to the credit evaluation method, and the repeated parts will not be repeated.

[0155] Specifically, the credit assessment device includes:

[0156] The credit assessment module is used to input the characteristic data of the customer to be predicted into the credit assessment model generated by the credit assessment model generation method described in any of the above embodiments to obtain the credit assessment result of the customer.

[0157] The credit assessment model generation method, credit assessment method and device provided in this article capture the deep correlation between the target to be evaluated and the surrounding nodes by training the graph neural network model, which has better learning effect than the machine learning model; the credit score of the customer by the machine learning model is added as a feature to the training of the graph neural network model, which improves the credit assessment accuracy of the model, and machine learning has stronger interpretability than the graph neural network. The model score of machine learning is added to improve the interpretability of the graph neural network; the sampled data in different business scenarios is used as the training data of the graph neural network model, so that the trained credit assessment model has better versatility; the characteristic data of the customer to be predicted is input into the credit assessment model to obtain accurate credit assessment results.

[0158] In one embodiment of the present invention, a computer device is also provided for implementing the method described in any of the above embodiments, such as Figure 7 The structure diagram of the computer device of the embodiment of this article is shown, and the computer device 702 may include one or more processors 704, such as one or more central processing units (CPUs), and each processing unit may implement one or more hardware threads. The computer device 702 may also include any memory 706, which is used to store any kind of information such as code, settings, data, etc. Non-limiting, for example, the memory 706 may include any one or more combinations of the following: any type of RAM, any type of ROM, flash memory device, hard disk, optical disk, etc. More generally, any memory can use any technology to store information. Further, any memory can provide volatile or non-volatile retention of information. Further, any memory can represent a fixed or removable component of the computer device 702. In one case, when the processor 704 executes an associated instruction stored in any memory or a combination of memories, the computer device 702 can perform any operation of the associated instruction. The computer device 702 also includes one or more drive mechanisms 708 for interacting with any memory, such as a hard disk drive mechanism, an optical disk drive mechanism, etc.

[0159] The computer device 702 may also include an input / output module 710 (I / O) for receiving various inputs (via input devices 712) and for providing various outputs (via output devices 714). A specific output mechanism may include a presentation device 716 and an associated graphical user interface (GUI) 718. In other embodiments, the input / output module 710 (I / O), the input device 710, and the output device 714 may not be included, and the computer device 702 may be used as a computer device in a network. The computer device 702 may also include one or more network interfaces 720 for exchanging data with other devices via one or more communication links 722. One or more communication buses 724 couple the components described above together.

[0160] The communication link 722 may be implemented in any manner, for example, through a local area network, a wide area network (e.g., the Internet), a point-to-point connection, etc., or any combination thereof. The communication link 722 may include any combination of hardwired links, wireless links, routers, gateway functions, name servers, etc. governed by any protocol or combination of protocols.

[0161] Corresponds to Figure 1-Figure 5 The method in this article also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the above method are executed.

[0162] The embodiment of the present invention also provides a computer readable instruction, wherein when the processor executes the instruction, the program therein causes the processor to execute the following Figures 1 to 5 The method shown.

[0163] It should be understood that in the various embodiments of this document, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this document.

[0164] It should also be understood that in the embodiments of this document, the term "and / or" is only a description of the association relationship of the associated objects, indicating that three relationships may exist. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this document generally indicates that the associated objects before and after are in an "or" relationship.

[0165] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this article.

[0166] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0167] In the several embodiments provided herein, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of the units is only a logical function division, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, or can be electrical, mechanical or other forms of connection.

[0168] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the embodiments of this article.

[0169] In addition, each functional unit in each embodiment of this invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above integrated unit may be implemented in the form of hardware or in the form of software functional unit.

[0170] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this article is essentially or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of this article. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0171] Specific embodiments are used in this article to illustrate the principles and implementation methods of this article. The description of the above embodiments is only used to help understand the methods and core ideas of this article. At the same time, for general technicians in this field, according to the ideas of this article, there will be changes in the specific implementation methods and application scopes. In summary, the content of this specification should not be understood as a limitation on this article.

Claims

1. A method for generating a credit assessment model, characterized in that: The method comprises: Obtaining a customer social edge table, a customer transaction edge table, a customer node table, and sampled customer information in different business scenarios, wherein the customer social edge table includes the customer's social relationship data, the customer transaction edge table includes the customer's transaction data, and the customer node table includes the customer node information, the credit score of the customer by the pre-trained machine learning model, and statistical feature information; Generate a customer association relationship graph based on the customer social edge table, the customer transaction edge table and the customer node table; Generate a credit assessment data set based on the sampled customer information in the different business scenarios and the customer association graph; A graph neural network model is trained according to the credit assessment data set to obtain a credit assessment model.

2. The method according to claim 1, characterized in that The process of generating the customer social side table includes: Constructing a social directed graph based on the social relationship data of the customer; Reversing the source node and the target node in the social directed graph to obtain a social reverse directed graph; The social directed graph and the social inverse directed graph are combined to generate a customer social edge table.

3. The method according to claim 1, characterized in that The process of generating the customer transaction side table includes: Determine the transferor and the recipient of each transaction based on the transaction data of the customer; Constructing a transaction directed graph according to preset relationship features and transaction data between the transferor and the payee; wherein the preset relationship features include transaction relationship features, relationship type features, and relationship strength; Exchanging the transferor and the payee in each transaction to obtain an exchanged transferor and payee; Constructing a transaction reversal directed graph according to the preset relationship characteristics, the transferor and the payee after the exchange; The transaction directed graph and the transaction reversal directed graph are combined to generate a customer transaction edge table.

4. The method according to claim 3, characterized in that The preset relationship characteristics include the preset relationship characteristics of transactional relationships and the preset relationship characteristics of non-transactional relationships; Determining the relationship strength of the transaction relationship according to the transaction relationship characteristics of the transaction relationship; the transaction relationship characteristics of the transaction relationship include the transaction amount; The relationship strength of the non-transactional relationship is determined based on the relationship type characteristics of the non-transactional relationship; the relationship type characteristics of the non-transactional relationship include quantified intimacy characteristics.

5. The method according to claim 1, characterized in that Generate a credit assessment data set based on the sampled customer information in the different business scenarios and the customer association relationship map, including: Remove duplicates from sampled customer information in different business scenarios to obtain a customer information set; Comparing customer information sets of different time periods to determine overlapping customer information; the overlapping customer information refers to customer information commonly included in the customer information sets of different time periods; Extracting characteristic data of customer information in customer information sets of different time periods from the customer association relationship map; The feature data and credit labels corresponding to the customer information identical to the overlapping customer information in the customer information sets of different time periods are classified into the training set corresponding to the current time period; the feature data and credit labels corresponding to the customer information different from the overlapping customer information in the customer information sets of different time periods are partially classified into the training set corresponding to the current time period; Based on the training sets of different time periods, multiple credit assessment data sets are generated in sequence; the multiple credit assessment data sets are used to train the graph neural network model in sequence.

6. The method according to claim 1, characterized in that The embedding features of each layer in the graph neural network model are expressed according to the following formula: Among them, i represents node i, j represents node j, k represents node k, and N(i) represents the set of neighbor nodes of node i in the Nth layer of the graph neural network model. Represents the embedded features of node i in the N+1th layer of the graph neural network model, represents the embedding feature of node i in the Nth layer of the graph neural network model, represents the embedding feature of node j in the Nth layer of the graph neural network model, σ represents the activation function, and α ij represents the attention weight of node j to node i, b represents the bias, W1, W2, W self , W neigh , W neigh2 , W edge Both represent parameter matrices, score ij represents the attention score of node j on node i, score ik represents the attention score of node k on node i, e ij represents the relationship feature between node i and node j, and |||| represents the concatenation operation.

7. The method according to claim 1, characterized in that The business scenarios include: pre-loan application scenario, loan use scenario and loan withdrawal scenario; The graph neural network model is trained using the following total loss function: Loss=a app L app +a loan L loan +a balance L balance ; in, L app represents the loss function in the pre-loan application scenario, L loan represents the loss function in the scenario of using credit in loan, L balance represents the loss function in the loan repayment scenario, N app represents the number of samples in the pre-loan application scenario, N loan The number of samples in the loan-using credit scenario, N balance The number of samples in the loan repayment scenario, L n Represents the loss value of the nth sample, W yn represents the calculation weight of the nth sample, Loss represents the total loss function of the graph neural network model, α app represents the calculation weight in the pre-loan application scenario, α loan represents the calculation weight in the scenario of using credit during loan, α balance Indicates the calculation weight in the loan repayment scenario.

8. A credit assessment method, characterized in that: The method comprises: The characteristic data of the customer to be predicted is input into the credit assessment model generated by any one of the methods of claims 1 to 7 to obtain the credit assessment result of the customer.

9. A credit assessment model generating device, characterized in that: The device comprises: An acquisition module, used to acquire a customer social edge table, a customer transaction edge table, a customer node table, and sampled customer information in different business scenarios, wherein the customer social edge table includes the customer's social relationship data, the customer transaction edge table includes the customer's transaction data, and the customer node table includes the customer node information, the credit score of the customer by the pre-trained machine learning model, and statistical feature information; A graph generation module, used to generate a customer association graph based on the customer social edge table, the customer transaction edge table and the customer node table; A data set generation module, used to generate a credit assessment data set based on the sampled customer information under the different business scenarios and the customer association relationship map; The model generation module is used to train a graph neural network model according to the credit assessment data set to obtain a credit assessment model.

10. A credit assessment device, characterized in that: The device comprises: A credit assessment module is used to input the characteristic data of the customer to be predicted into the credit assessment model generated by any method of claims 1 to 7 to obtain the credit assessment result of the customer.

11. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 8 is implemented.

12. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor of a computer device, the method according to any one of claims 1 to 8 is implemented.

13. A computer program product, comprising a computer program, characterized in that: When the computer program is executed by a processor of a computer device, the method according to any one of claims 1 to 8 is implemented.

Images