Loan prediction method, device, equipment, medium and program product

Abstract

The disclosure provides a loan prediction method, device, equipment, medium and program product, relates to the technical field of big data, and can be used in the field of financial technology, and the method comprises the following steps: obtaining authorization of a to-be-tested user to obtain user information; obtaining user information of the to-be-tested user after obtaining the authorization of the to-be-tested user to obtain the user information; predicting risk information of the to-be-tested user through a risk identification model and according to the user information; wherein the risk identification model is constructed based on loan transaction data of approved users and loan transaction data of rejected users after marking, the loan transaction data of the rejected users is marked by importance sampling and by using the loan transaction data of the approved users; and determining loan granting information of the to-be-tested user according to the risk information.

Description

Technical Field

[0001] This disclosure relates to the field of big data technology, and in particular to a loan prediction method, apparatus, device, medium, and program product. Background Technology

[0002] In financial risk control, financial institutions establish risk control models to screen applicants. Some applicants with higher scores can receive loans and are called "approved users," while applicants with lower scores are directly rejected for loans and are called "rejected users." The platform can only obtain the good or bad rating of the loan recipients, but it has no way to obtain the post-loan repayment status of the rejected users.

[0003] Since rejected users do not make subsequent loan repayments, no transaction data on their performance is generated. Over time, institutions can only obtain loan transaction data from approved users with "better scores," but not from rejected users with "lower scores" who have not made loan repayments. Therefore, they usually choose to predict the loan of the target user based on the loan transaction data of approved users.

[0004] However, the actual customer group that financial risk control faces is all potential future loan applicants. The loan transaction data of approved users cannot represent the characteristics of all customers, but only those with "better scores". Therefore, predicting user loans based solely on the loan transaction data of approved users may deviate from the actual situation, and may even approve many bad users who should have been rejected, leading to an increase in delinquency rates and causing significant economic losses to financial institutions. Summary of the Invention

[0005] In view of the above problems, a first aspect of this disclosure provides a loan prediction method, comprising: obtaining authorization from a user to be tested for obtaining user information; obtaining user information of the user to be tested after obtaining authorization from the user to be tested; predicting risk information of the user to be tested based on a risk identification model and the user information; wherein the risk identification model is constructed based on loan transaction data of approved users and loan transaction data of rejected users after being marked, and the loan transaction data of rejected users is marked by importance sampling and using the loan transaction data of approved users; and determining loan grant information of the user to be tested based on the risk information.

[0006] According to embodiments of this disclosure, the process of labeling loan transaction data of rejected users by importance sampling and utilizing loan transaction data of approved users includes: obtaining loan transaction data of approved users and loan transaction data of rejected users; predicting a first probability that an approved user or rejected user is a fulfilling customer using a first model and based on the loan transaction data of approved users and rejected users; predicting a second probability that an approved user or rejected user is a loan approval customer using a second model and based on the loan transaction data of approved users and rejected users; calculating a similarity between rejected users and approved users using a third model and based on the first and second probabilities; and labeling the loan transaction data of the corresponding rejected users using loan transaction data of approved users with a similarity greater than a preset value.

[0007] According to an embodiment of this disclosure, a logistic regression model is constructed based on the loan transaction data of approved users to obtain a first model.

[0008] According to an embodiment of this disclosure, a logistic regression model is constructed based on loan transaction data of approved users and loan transaction data of rejected users to obtain a second model.

[0009] According to embodiments of this disclosure, the third model includes a label propagation model.

[0010] According to embodiments of this disclosure, calculating the similarity between a rejecting user and an approving user using a third model and based on a first probability and a second probability includes: inputting loan transaction data of approving users and loan transaction data of rejecting users into the third model to construct a fully connected graph, wherein each node in the fully connected graph represents loan transaction data of an approving user or a rejecting user; determining a first node, the first node representing the loan transaction data of an approving user; determining a second node, the second node being the node closest to the first node and representing the loan transaction data of a rejecting user; and calculating the similarity between the rejecting user corresponding to the second node and the approving user corresponding to the first node based on the first probability and the second probability corresponding to the first node and the first probability corresponding to the second node.

[0011] According to an embodiment of this disclosure, calculating the similarity between a user who rejects a user at a second node and a user who approves a user at a first node based on the first probability and the second probability corresponding to the first node and the first probability corresponding to the second node includes: calculating the difference between the first probability corresponding to the first node and the first probability corresponding to the second node; and calculating the similarity between the user who rejects a user at a second node and the user who approves a user at a first node based on the difference and the reciprocal of the second probability corresponding to the first node.

[0012] According to embodiments of this disclosure, the loan prediction method further includes: preprocessing the loan transaction data of approved users and the loan transaction data of rejected users, including: deleting the loan transaction data of approved users and the loan transaction data of rejected users respectively to fill in the gaps; deleting highly correlated features from the loan transaction data of approved users and the loan transaction data of rejected users respectively; and normalizing and variance filtering the loan transaction data of approved users and the loan transaction data of rejected users.

[0013] According to embodiments of this disclosure, the loan prediction method further includes: correcting the risk identification model based on the transaction data corresponding to the user to be tested.

[0014] A second aspect of this disclosure provides a loan prediction apparatus, comprising: a first acquisition module for acquiring authorization from a user to be tested to acquire user information; a second acquisition module for acquiring user information of the user to be tested after obtaining authorization from the user to be tested; a prediction module for predicting risk information of the user to be tested based on a risk identification model and the user information; wherein the risk identification model is constructed based on loan transaction data of approved users and loan transaction data of rejected users after being marked, and the loan transaction data of rejected users is marked by importance sampling and using the loan transaction data of approved users; and a determination module for determining loan grant information of the user to be tested based on the risk information.

[0015] A third aspect of this disclosure provides an electronic device including one or more processors; and a storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors perform the loan prediction method described above.

[0016] A fourth aspect of this disclosure provides a computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, cause the processor to perform the loan prediction method described above.

[0017] A fifth aspect of this disclosure provides a computer program product, including a computer program that is executed by a processor using the loan prediction method described above.

[0018] According to the loan prediction method provided in this disclosure, the loan transaction data of the rejected user is marked by importance sampling and using the loan transaction data of the approved user. Then, a risk identification model is constructed based on the loan transaction data of the approved user and the loan transaction data of the rejected user to identify the risk information of the user to be tested. The loan grant information of the user to be tested is then determined based on the risk identification information, thereby determining whether to grant a loan to the user to be tested. This method can improve the accuracy of loan prediction and reduce financial risk. Attached Figure Description

[0019] The foregoing contents, as well as other objects, features, and advantages of this disclosure, will become clearer from the following description of embodiments with reference to the accompanying drawings, in which:

[0020] Figure 1 The system architecture 100 of the loan prediction method and system according to an embodiment of the present disclosure is illustrated schematically.

[0021] Figure 2 A flowchart illustrating a loan prediction method according to an embodiment of the present disclosure is shown schematically.

[0022] Figure 3 The flowchart illustrating a method for marking loan transaction data of rejected users according to an embodiment of the present disclosure is shown.

[0023] Figure 4 The flowchart illustrates a method for calculating the similarity between a rejecting user and an approving user based on a first probability and a second probability according to an embodiment of the present disclosure.

[0024] Figure 5 The flowchart illustrates a method for calculating the similarity between a rejecting user corresponding to a second node and an approving user corresponding to a first node, according to an embodiment of the present disclosure, based on a first probability and a second probability corresponding to a first node and a first probability corresponding to a second node.

[0025] Figure 6 A flowchart illustrating a loan prediction method according to another embodiment of this disclosure is shown schematically.

[0026] Figure 7 A flowchart illustrating a loan prediction method according to yet another embodiment of this disclosure is shown.

[0027] Figure 8 A block diagram of a loan prediction apparatus according to an embodiment of the present disclosure is shown schematically.

[0028] Figure 9 A block diagram of a loan prediction apparatus according to another embodiment of the present disclosure is shown schematically.

[0029] Figure 10 A block diagram of a marking module 850 according to an embodiment of the present disclosure is shown schematically.

[0030] Figure 11 A block diagram of a loan prediction apparatus according to yet another embodiment of the present disclosure is shown schematically.

[0031] Figure 12 A block diagram of a loan prediction apparatus according to yet another embodiment of the present disclosure is shown schematically.

[0032] Figure 13A block diagram of an electronic device suitable for implementing the methods described above, according to embodiments of the present disclosure, is illustrated schematically. Detailed Implementation

[0033] The embodiments of the present disclosure will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the disclosure. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of the present disclosure for ease of explanation. However, it will be apparent that one or more embodiments may be practiced without these specific details. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concepts of the present disclosure.

[0034] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit this disclosure. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0035] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0036] When using expressions such as "at least one of A, B, and C," the expression should generally be interpreted in accordance with the meaning commonly understood by a person skilled in the art (e.g., "a system having at least one of A, B, and C" should include, but is not limited to, systems having A alone, having B alone, having C alone, having A and B, having A and C, having B and C, and / or having A, B, and C, etc.). Similarly, when using expressions such as "at least one of A, B, or C," the expression should generally be interpreted in accordance with the meaning commonly understood by a person skilled in the art (e.g., "a system having at least one of A, B, or C" should include, but is not limited to, systems having A alone, having B alone, having C alone, having A and B, having A and C, having B and C, and / or having A, B, and C, etc.).

[0037] The accompanying drawings illustrate several block diagrams and / or flowcharts. It should be understood that some blocks, or combinations thereof, in the block diagrams and / or flowcharts can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that, when executed by the processor, these instructions can create means for implementing the functions / operations described in these block diagrams and / or flowcharts. The technology of this disclosure can be implemented in hardware and / or software (including firmware, microcode, etc.). Alternatively, the technology of this disclosure can take the form of a computer program product stored on a computer-readable storage medium, which is available for use by or in conjunction with an instruction execution system.

[0038] In the technical solution disclosed herein, the collection, storage, use, processing, transmission, provision, disclosure, and application of user personal information comply with the provisions of relevant laws and regulations, necessary confidentiality measures have been taken, and there is no violation of public order and good morals.

[0039] In the technical solution disclosed herein, the user's authorization or consent is obtained before acquiring or collecting the user's personal information.

[0040] To address the problem in related technologies of ensuring excellent loan prediction capabilities for all approved and rejected users when only the tags of high-quality users who have received loans are available, this disclosure provides a loan prediction method, including: obtaining authorization from the user to be tested for obtaining user information; obtaining the user information of the user to be tested after obtaining authorization; predicting the risk information of the user to be tested based on the user information using a risk identification model; wherein, the risk identification model is constructed based on the loan transaction data of approved users and the loan transaction data of rejected users after being tagged, and the loan transaction data of rejected users is tagged using the loan transaction data of approved users through importance sampling; and determining the loan granting information of the user to be tested based on the risk information.

[0041] Figure 1 The system architecture 100 of the loan prediction method and system according to an embodiment of this disclosure is illustrated schematically. It should be noted that... Figure 1 The examples shown are merely examples of system architectures that can be applied to the embodiments of this disclosure, in order to help those skilled in the art understand the technical content of this disclosure, but do not mean that the embodiments of this disclosure cannot be used in other devices, systems, environments or scenarios.

[0042] like Figure 1 As shown, the system architecture 100 according to this embodiment may include a client 101, a database 102, a network 103, and a server 104. The network 103 provides a communication link between the client 101, the database 102, and the server 104.

[0043] Client 101 may include, but is not limited to, smartphones, tablets, desktop PCs, laptops, netbooks, workstations, servers, etc. Client 101 is used by users to fill in relevant user information when applying for a loan. Database 102 can be used to store historical loan transaction data of approved users and transaction data of rejected users. It should be understood that this loan transaction data of approved users and rejected users is confidential and can only be accessed with the user's authorization. Network 103 may include various connection types, such as wired, wireless communication links, or fiber optic cables, etc. Wired connections may use cables and any of the following interfaces: Fibre Channel, infrared interface, Type-D data interface, serial interface, USB interface, USB Type-C interface, or Dock interface. Wireless connections may use wireless communication methods, such as Bluetooth, Wi-Fi, Infrared, ZigBee, etc. The user information (loan application) filled in by the test user through client 101 is used by server 104 to obtain the loan transaction data of approved users and rejected users in database 102 through network 103. The server 104 uses importance sampling and the loan transaction data of approved users to mark the loan transaction data of rejected users. A risk identification model is constructed based on the loan transaction data of approved users and the marked loan transaction data of rejected users. The loan grant information of the test user is determined through the risk identification model and according to the user information filled in by the test user.

[0044] It should be noted that the loan prediction method provided in this embodiment can be executed by server 104. Accordingly, the loan prediction device provided in this embodiment can be located in server 104. Alternatively, the loan prediction method provided in this embodiment can also be executed by a server or server cluster different from server 104 and capable of communicating with client 101, database 102, and / or server 104. Correspondingly, the loan prediction device provided in this embodiment can also be located in a server or server cluster different from server 104 and capable of communicating with client 101, database 102, and / or server 104. Alternatively, the loan prediction method provided in this embodiment can also be partially executed by server 104 and partially executed by client 101 and database 102. Accordingly, the loan prediction device provided in this embodiment can also be partially located in server 104 and partially located in client 101 and database 102.

[0045] It should be understood that Figure 1The number of clients, databases, networks, and servers shown is merely illustrative. Depending on implementation needs, there can be any number of clients, databases, networks, and servers.

[0046] The loan forecasting method provided in this disclosure can be applied to the fintech field. For example, to meet the needs of a wide range of users and businesses, banks provide credit services. However, existing loan forecasting methods have low accuracy, which may lead to losses for banks. To avoid losses to banks due to customer or business defaults, the loan forecasting method provided in this disclosure can at least partially solve the technical problems existing in the prior art.

[0047] It should be understood that the loan prediction method provided in this disclosure is not limited to the field of financial technology. The above description is merely exemplary. The loan prediction method of this disclosure can be applied to other technical fields involving loan prediction.

[0048] Figure 2 A flowchart illustrating a loan prediction method according to an embodiment of the present disclosure is shown schematically.

[0049] like Figure 2 As shown, this loan prediction method is based on transfer learning and semi-supervised learning. It adds the idea of ​​transfer learning to the graph label propagation algorithm, corrects the loan transaction data of biased approved users to loan transaction data that is more representative of the characteristics of all possible future customers, and uses the corrected data to build a model to predict the loan of the user to be tested. This method may include operations S201 to S204.

[0050] In operation S201, obtain the user's authorization to access user information.

[0051] In operation S202, after obtaining authorization from the user under test to obtain user information, the user information corresponding to the user under test is obtained.

[0052] In operation S203, the risk information of the user to be tested is predicted by the risk identification model based on the user information; wherein, the risk identification model is constructed based on the loan transaction data of approved users and the loan transaction data of rejected users after being marked, and the loan transaction data of rejected users is marked by importance sampling and using the loan transaction data of approved users.

[0053] In operation S204, the loan grant information for the user under test is determined based on the risk information.

[0054] In embodiments of this disclosure, the user's consent or authorization can be obtained before acquiring the user's information. For example, a request to acquire user information can be sent to the user before operation S202. If the user agrees or authorizes the acquisition of user information, operation S202 is executed.

[0055] In this embodiment of the disclosure, user information may include, for example, credit card account, loan period, credit card score, auto loan, personal loan, mortgage, education loan, total account, savings account, years of employment, age, whether there are any overdue payments, gender, marital status, employment status, etc. Loan granting information may include, for example, granting or denying a loan to the user being tested.

[0056] In this embodiment, transfer learning is used to label loan transactions of rejected users. Transfer learning can be categorized into four types based on the transfer subject. The first is sample transfer: although data from the source domain cannot be directly applied to the target domain as a whole, some data from the source domain can be reapplied to the target domain. After adjusting the weights of this source domain data to match the data in the target domain, the transfer is performed. The second is feature transfer: finding some representative high-quality features and using feature transformation to represent the features of the source and target domains with common features, ensuring that the data in the source and target domains have the same distribution, and then performing traditional machine learning. The third is parameter transfer: assuming that the models in the source and target domains have common parameters or prior distributions, these parameters can be shared between the source and target domains, thus achieving good accuracy when transferring the original model to new data. The fourth is relation transfer: transferring similar relations, corresponding some relations in the source domain to relations in the target domain, and further transferring them. The two main techniques used in sample transfer learning include sample reweighting and importance sampling. Preferably, the transfer learning technique used in this embodiment is importance sampling.

[0057] In this embodiment of the disclosure, the risk identification model constructed based on the loan transaction data of approved users and the loan transaction data of labeled rejected users adopts a semi-supervised learning method.

[0058] Specifically, model construction methods can include supervised learning and unsupervised learning. Supervised learning is very common, and widely used classification algorithms such as logistic regression and decision trees belong to supervised learning. These algorithms are suitable for scenarios where the training data samples contain labeled information. Unsupervised learning, on the other hand, is for learning tasks where the training data samples do not contain labeled information, such as clustering algorithms for grouping. In these scenarios, the training samples are unlabeled. However, in real-world scenarios, a large number of data samples are often missing labels, with only a small portion containing labels. In such cases, labeled data samples are extremely valuable and rare. In loan application scenarios, loan transaction data of approved users can correspond to labeled data samples, while loan transaction data of rejected users can correspond to unlabeled data samples. Furthermore, the number of users who receive loans is relatively small compared to the total number of applicants; that is, only a small portion of loan recipients have labeled data.

[0059] In realizing the concept of this disclosure, the applicant discovered that although unlabeled data (sample transaction data of rejected users) does not directly contain labeling information, if this data and labeled data (sample transaction data of approved users) are obtained from the same data source through independent and identically distributed sampling, then the information about the data distribution contained in the unlabeled data can be very helpful for model building. Therefore, in order to better utilize unlabeled sample data to build a model, the embodiments of this disclosure introduce semi-supervised learning. Semi-supervised learning is an important research problem in the field of machine learning, and it is a method that combines supervised and unsupervised learning. Semi-supervised learning uses labeled samples while introducing a large number of unlabeled samples to jointly build a machine learning model, so that the machine learning model no longer depends on external interaction and can automatically use unlabeled data to improve learning performance. When using semi-supervised learning, a large amount of human and material resources are no longer needed to obtain or label data. At the same time, semi-supervised learning has relatively high accuracy and performance, so it is playing an increasingly important role.

[0060] Therefore, this embodiment of the disclosure uses importance sampling and marks the loan transaction data of rejected users using the loan transaction data of approved users, so that the loan transaction data of rejected users has a similar distribution to the loan transaction data of approved users. This results in loan transaction data that is more representative of the characteristics of all customers who may apply for loans in the future. As a result, the risk identification model built based on the loan transaction data of approved users and the marked loan transaction data of rejected users makes the loan prediction of the target user more consistent with the actual situation and the prediction more accurate.

[0061] According to the loan prediction method provided in this disclosure, the loan transaction data of the rejected user is marked by importance sampling and using the loan transaction data of the approved user. Then, a risk identification model is constructed based on the loan transaction data of the approved user and the marked loan transaction data of the rejected user to identify the risk information of the user to be tested. The loan grant information of the user to be tested is then determined based on the risk identification information. Finally, the loan grant information is used to determine whether to grant or reject the loan. This method can improve the accuracy of loan prediction and reduce financial risk.

[0062] The following section continues with reference to the accompanying drawings. Figure 1 The loan forecasting method shown is explained in detail.

[0063] Figure 3 The flowchart illustrating a method for marking loan transaction data of rejected users according to an embodiment of the present disclosure is shown.

[0064] like Figure 3 As shown, the process of sampling by importance and marking the loan transaction data of rejected users by using the loan transaction data of approved users may include, for example, operations S301 to S305.

[0065] In operation S301, obtain loan transaction data for approved users and loan transaction data for rejected users.

[0066] In this embodiment of the disclosure, consent or authorization from historical approving and rejecting users can be obtained before acquiring their loan transaction data. For example, before operation S301, a request to acquire loan transaction data can be sent to historical approving and rejecting users. If historical approving and rejecting users consent or authorize the acquisition of loan transaction data, operation S302 is performed.

[0067] In operation S302, the first probability of an approved or rejected user being a fulfilling customer is predicted using the first model and based on the loan transaction data of approved users and rejected users.

[0068] In this embodiment of the disclosure, the loan transaction data of approved users and the loan transaction data of rejected users are labeled to form the sample data for constructing the risk identification model. The loan transaction data of approved users can be denoted as labeled sample data, and the loan transaction data of rejected users can be denoted as unlabeled sample data. When labeling the unlabeled sample data using labeled sample data, it is necessary to calculate the similarity between the labeled and unlabeled sample data, which can also be referred to as the similarity between the users corresponding to the labeled and unlabeled sample data respectively.

[0069] In realizing the concept of this disclosure, the applicant discovered that when calculating the similarity between rejecting users and approving users, directly calculating the Euclidean distance of each sample feature space would ignore the differences in importance between the features. Therefore, the embodiments of this disclosure can use the difference in the predicted probability of sample data fulfillment to replace the calculation of the Euclidean distance of the sample feature space.

[0070] Therefore, this embodiment of the disclosure needs to predict, through a first model, the probability that an approving user or a rejecting user will be a fulfilling customer. Specifically, a logistic regression model can be constructed based on the loan transaction data of approved users to obtain the first model. Each transaction data point (including approved and rejected users) is input into the first model to predict the probability that the user corresponding to the loan transaction data will be a fulfilling customer.

[0071] In operation S303, a second model is used to predict the second probability of approving or rejecting a loan for a customer based on the loan transaction data of approved users and rejected users.

[0072] In this embodiment, importance sampling is used to label loan transaction data of rejected users based on loan transaction data of approved users. Importance sampling can obtain samples from a specified distribution from another distribution. By randomly weighting an average of a relatively simple distribution function (corresponding to the distribution of loan transaction data of approved users), the expected value of the target distribution function is approximately calculated. This relatively simple distribution function can be called the importance density function. The weight values ​​are approximately proportional to the likelihood ratio of the two distributions. Generally, the purpose of sampling is to evaluate the expected value of a function on a certain distribution, that is:

[0073]

[0074] If samples following the p(x) distribution are not easy to generate, then a new distribution q(x) can be introduced, which allows for convenient sampling:

[0075]

[0076]

[0077] This transforms the expectation of f(x) under the distribution p(x) into the expectation of g(x) under the distribution q(x), which can be called the importance weight w(x) = p(x) / q(x). Here, the original distribution p(x) is the distribution of the accepted sample data (the loan transaction data corresponding to approved users), and the target distribution q(x) is the distribution of the total sample (the loan transaction data corresponding to approved users and the loan transaction data of the labeled rejected samples). The weight can be chosen as the ratio of the probability that the training sample is a sample of the total sample to the probability that the training sample is an accepted sample.

[0078] Therefore, a logistic regression model can be built by combining loan transaction data of approved users (labeled samples) and loan transaction data of rejected users (unlabeled samples) to obtain a second model. The predicted label is whether it is labeled or not. The loan transaction data corresponding to the user is input into the second model to predict the second probability of the user being a loan approved customer. Then, the importance features of the loan transaction data of approved users are transferred to the loan transaction data of rejected users through importance assessment.

[0079] In operation S304, the similarity between the rejecting user and the approving user is calculated using the third model and based on the first and second probabilities.

[0080] In this embodiment of the disclosure, the results of importance sampling are used in the label propagation model so that the loss function can better represent the overall sample.

[0081] In operation S305, the loan transaction data of the approved users with a similarity greater than a preset value are used to mark the loan transaction data of the corresponding rejected users.

[0082] In this embodiment, a similarity greater than a preset value between the loan transaction data of an approved user and the loan transaction data of an rejected user can mean that the distribution of the approved user's loan transaction data is similar to that of the rejected user's loan transaction data. The characteristics of the approved user's loan transaction data will significantly influence the prediction of the rejected user's loan transaction data. Therefore, the loan transaction data of the approved user (with a similarity greater than the preset value) is used to mark the corresponding rejected user's loan transaction data.

[0083] According to the embodiment of this disclosure, the method of labeling loan transaction data of rejected users by sampling importance and using loan transaction data of approved users is proposed. The Euclidean distance of the sample feature space is calculated based on the difference in the probability of the sample data being predicted to be fulfilled, and the similarity is calculated by combining the probability of each user being predicted to be a fulfilling customer. On the basis of fully considering the importance features, more accurate and appropriate loan transaction data of approved users are selected for labeling the loan transaction data of rejected users. This can effectively transfer the important features of loan transaction data of approved users to loan transaction data of rejected users. In this way, the loan transaction data can better represent the characteristics of all customers who may apply for loans in the future, making the loan prediction more in line with the actual situation and the prediction more accurate.

[0084] Figure 4 The flowchart illustrates a method for calculating the similarity between a rejecting user and an approving user based on a first probability and a second probability according to an embodiment of the present disclosure.

[0085] like Figure 4 As shown, the method may include, for example, operations S401 to S404.

[0086] In operation S401, the loan transaction data of approved users and the loan transaction data of rejected users are input into the third model to construct a fully connected graph, where each node of the fully connected graph represents the loan transaction data of an approved user or a rejected user.

[0087] In this embodiment of the disclosure, a Gaussian similarity function is typically used to construct such a fully connected graph, which builds a graph for all the data. The nodes of the graph are data points, containing labeled data (loan transaction data of approved users) and unlabeled data (loan transaction data of rejected users).

[0088] In operation S402, the first node is determined. The first node represents the node that approves the user's loan transaction data.

[0089] In this embodiment of the disclosure, a node that can represent the loan transaction data of the approving user is arbitrarily determined from the fully connected graph as the first node.

[0090] In operation S403, the second node is determined. The second node is the node closest to the first node and represents the loan transaction data of the rejected user.

[0091] In operation S404, the similarity between the rejected user corresponding to the second node and the approved user corresponding to the first node is calculated based on the first probability and the second probability corresponding to the first node and the first probability corresponding to the second node.

[0092] Generally, when using Euclidean distance to calculate similarity, the edge between nodes i and j represents the similarity between the two points, i.e., the edge weight is:

[0093]

[0094] Where, x j It is the distance x i One of the k most recent points.

[0095] As can be seen from the foregoing description, the embodiments of this disclosure use the difference in the probability of predicted performance of sample data to replace the calculation of Euclidean distance in the sample feature space to calculate similarity. Based on this, the embodiments of this disclosure are analogous to the calculation method of Euclidean distance to calculate similarity.

[0096] Figure 5 The flowchart illustrates a method for calculating the similarity between a rejecting user corresponding to a second node and an approving user corresponding to a first node, according to an embodiment of the present disclosure, based on a first probability and a second probability corresponding to a first node and a first probability corresponding to a second node.

[0097] like Figure 5 As shown, the method may include, for example, operations S501 to S502.

[0098] In operation S501, the difference between the first probability corresponding to the first node and the first probability corresponding to the second node is calculated.

[0099] In operation S502, the similarity between the rejected user corresponding to the second node and the approved user corresponding to the first node is calculated based on the difference and the reciprocal of the second probability corresponding to the first node.

[0100] Specifically, analogous to the calculation of Euclidean distance, the edge weight between node i and node j is:

[0101] x j It is the distance x i One of the nearest k points

[0102] W ij =0, otherwise

[0103] Where, p gd (x i Let p be the first probability that the approved user corresponding to the loan transaction data of the approved user represented by the i-th node is predicted to be a fulfilling customer. gd (x j Let be the first probability that the user whose loan transaction data represents the user who refused the loan is predicted to be a customer who fulfills the loan agreement.

[0104] When x iand x j When dealing with loan transaction data for approved and rejected users, the reciprocal of the second probability can be used to weight the loss function between the loan transaction data represented by node i and node j. Since the probability of an approved user being accepted is often greater than the probability of being rejected, the reciprocal of the second probability is usually greater than 1. Furthermore, the higher the probability of acceptance, the smaller the weight for importance estimation. This way, loan transaction data for approved users, whose distribution is inconsistent with that of rejected users, will not affect x. j The label prediction has a significant impact. Conversely, when the predicted probability of acceptance based on the loan transaction data of an approved user is low, it means that the probability of that historically approved user being rejected is relatively high. This indicates that the distribution of the loan transaction data of this approved user is quite similar to the distribution of the loan transaction data of the rejected user, which will affect x. j The label prediction has a significant impact. Therefore, the edge weights (i.e., similarity) for graph label propagation are updated:

[0105]

[0106] in, Let σ be the second probability, and σ be a preset coefficient.

[0107] Based on this similarity, the new energy function can be:

[0108]

[0109] According to the similarity calculation method provided in this disclosure embodiment, based on the graph labeling model, the difference between the first probabilities can fully consider the differences in importance, and the reciprocal of the second probability is used to weight the loss function between the loan transaction data represented by node i and node j, so as to ensure that the loan transaction data of the approved user closest to the loan transaction distribution of a rejected user can be selected, thereby improving the accuracy of similarity calculation and ensuring the accuracy of data labeling.

[0110] Figure 6 A flowchart illustrating a loan prediction method according to another embodiment of this disclosure is shown schematically.

[0111] like Figure 6 As shown, before labeling the data, the loan prediction method may include, for example, preprocessing the loan transaction data of approved users and the loan transaction data of rejected users, which may include operations S601 to S603.

[0112] In operation S601, the loan transaction data of approved users and the loan transaction data of rejected users are deleted to make up for the missing data.

[0113] In this embodiment, missing values ​​in loan transaction data for approved users and rejected users can be processed separately. For example, the missing values ​​can be: first, columns with a missing percentage greater than 50% are deleted; then, missing values ​​for the remaining features are imputed. For columns with no more than 5 possible values, they are randomly imputed according to the proportion of each value; otherwise, the mean is used. Since the distribution differences between accepted and rejected samples may be significant, the imputation process is performed separately. Then, features with a single overall value are deleted; if the proportion of a certain value exceeds 99%, it is deleted.

[0114] In operation S602, highly correlated features are deleted from the loan transaction data of approved users and the loan transaction data of rejected users, respectively.

[0115] In this embodiment of the disclosure, the deletion of highly correlated features can be measured using the Pearson coefficient to measure the degree of correlation between features. The formula for the Pearson correlation coefficient can be as follows:

[0116]

[0117] In operation S603, normalization and variance filtering are performed on the loan transaction data of approved users and the loan transaction data of rejected users.

[0118] In this embodiment of the disclosure, in order to compare features with different value ranges, it is necessary to convert dimensional values ​​into dimensionless values. This allows features of different magnitudes to be compared, so mapping the data to the range [0, 1] helps improve the model's performance. The specific formula for normalization can be:

[0119]

[0120] Furthermore, feature selection can be achieved by filtering out features with low variance. Variance reflects the dispersion of a feature. If the variance of a feature is extremely small, it means that most sample data have the same value, and then this feature is not very meaningful to the model. For example, features with variance below 0.01 can be deleted.

[0121] According to the data preprocessing method provided in this disclosure, by employing missing value deletion and filling, deletion of highly correlated features, normalization, and low variance filtering, useless data in the original loan transaction data can be removed, and the original data can be transformed into standard structured data. This reduces the computational load of subsequent data labeling and avoids interference from useless data in loan prediction, thereby improving the accuracy of loan prediction.

[0122] Figure 7 A flowchart illustrating a loan prediction method according to yet another embodiment of this disclosure is shown.

[0123] like Figure 7 As shown, after operation S204, the loan forecasting method may include, for example, operation S701.

[0124] In operation S201, obtain the user's authorization to access user information.

[0125] In operation S202, after obtaining authorization from the user under test to obtain user information, the user information corresponding to the user under test is obtained.

[0126] In operation S203, the risk information of the user to be tested is predicted by the risk identification model based on the user information; wherein, the risk identification model is constructed based on the loan transaction data of approved users and the loan transaction data of rejected users after being marked, and the loan transaction data of rejected users is marked by importance sampling and using the loan transaction data of approved users.

[0127] In operation S204, the loan grant information for the user under test is determined based on the risk information.

[0128] When operating S701, the risk identification model is corrected based on the transaction data of the user to be tested.

[0129] In this embodiment of the disclosure, a risk identification model is used to predict loan eligibility for new loan applicants and generate new approval tags to determine whether to grant a loan. Loan transaction data generated from newly approved and available repayment behaviors are analyzed to obtain performance tags, which demonstrate the creditworthiness of the customer.

[0130] Whether new loan transaction data passes the label and whether it is a good customer is continuously revised and iterated by regularly adding new loan transaction data to the risk identification model.

[0131] The loan prediction method provided in this disclosure continuously adds the number of loan transactions generated by new users to the risk identification model for continuous correction and iteration, resulting in a robust risk identification model that always adapts to business changes, thereby further improving the accuracy of loan prediction probability.

[0132] Figure 8 A block diagram of a loan prediction apparatus according to an embodiment of the present disclosure is shown schematically.

[0133] like Figure 8 As shown, the loan default detection device 800 may include a first acquisition module 810, a second acquisition module 820, a prediction module 830, and a determination module 840.

[0134] The first acquisition module 810 is used to acquire the authorization of the user to be tested to acquire user information.

[0135] The second acquisition module 820 is used to acquire the user information of the user under test after obtaining authorization from the user under test to acquire user information.

[0136] The prediction module 830 is used to predict the risk information of the user to be tested by using a risk identification model and based on user information. The risk identification model is constructed based on the loan transaction data of approved users and the loan transaction data of rejected users after being marked. The loan transaction data of rejected users is marked by importance sampling and using the loan transaction data of approved users.

[0137] The determination module is used to determine loan grant information for the user under test based on risk information.

[0138] Figure 9 A block diagram of a loan prediction apparatus according to another embodiment of the present disclosure is shown schematically.

[0139] like Figure 9 As shown, the loan default detection device 800 may also include, for example, a marking module 850.

[0140] The tagging module 850 is used to tag loan transaction data of rejected users by sampling for importance and using loan transaction data of approved users.

[0141] Figure 10 A block diagram of a marking module 850 according to an embodiment of the present disclosure is shown schematically.

[0142] like Figure 10 As shown, the marking module 850 may include, for example, an acquisition unit 851, a first prediction unit 852, a second prediction unit 853, a calculation unit 854, and a marking unit 855.

[0143] The acquisition unit 851 is used to acquire loan transaction data of approved users and loan transaction data of rejected users.

[0144] The first prediction unit 852 is used to predict, through the first model and based on the loan transaction data of approved users and loan transaction data of rejected users, the first probability that an approved user or a rejected user is a fulfilling customer.

[0145] The second prediction unit 853 is used to predict the second probability of a loan approval or rejection of a customer by using the second model and based on the loan transaction data of approved users and loan transaction data of rejected users.

[0146] The calculation unit 854 is used to calculate the similarity between the rejecting user and the approving user using a third model and based on a first probability and a second probability.

[0147] The tagging unit 855 is used to tag the loan transaction data of the corresponding rejected user using the loan transaction data of the approved user with a similarity greater than a preset value.

[0148] Figure 11 A block diagram of a loan prediction apparatus according to yet another embodiment of the present disclosure is shown schematically.

[0149] like Figure 11 As shown, the loan default detection device 800 may also include a preprocessing module 860.

[0150] The preprocessing module 860 is used to preprocess the loan transaction data of approved users and rejected users. Specifically, this may include: deleting data from both approved and rejected users to fill in the gaps; removing highly correlated features from both sets of data; and normalizing and filtering the data based on variance.

[0151] Figure 12 A block diagram of a loan prediction apparatus according to yet another embodiment of the present disclosure is shown schematically.

[0152] like Figure 12 As shown, the loan default detection device 800 may also include, for example, a correction module 870.

[0153] The correction module 870 is used to correct the risk identification model based on the transaction data corresponding to the user under test. Any one or more of the modules, sub-modules, units, and sub-units according to the embodiments of this disclosure, or at least part of the functions of any one or more of them, can be implemented in one module. Any one or more of the modules, sub-modules, units, and sub-units according to the embodiments of this disclosure can be implemented by splitting them into multiple modules. Any one or more of the modules, sub-modules, units, and sub-units according to the embodiments of this disclosure can be at least partially implemented as hardware circuits, such as field-programmable gate arrays (FPGAs), programmable logic arrays (PLAs), systems-on-a-chip, systems-on-a-substrate, systems-on-package, application-specific integrated circuits (ASICs), or implemented by hardware or firmware in any other reasonable manner by integrating or packaging circuits, or implemented in any one of software, hardware, and firmware, or in a suitable combination of any of these. Alternatively, one or more of the modules, sub-modules, units, and sub-units according to the embodiments of this disclosure can be at least partially implemented as computer program modules, which can perform corresponding functions when the computer program module is run.

[0154] For example, any multiple modules such as the first acquisition module 810, the second acquisition module 820, the prediction module 830, the determination module 840, the marking module 850, the preprocessing module 860, and the correction module 870 can be combined into one module / unit / subunit, or any one of these modules / units / subunits can be split into multiple modules / units / subunits. Alternatively, at least some of the functionality of one or more of these modules / units / subunits can be combined with at least some of the functionality of other modules / units / subunits and implemented in one module / unit / subunit. According to embodiments of this disclosure, at least one of the first acquisition module 810, the second acquisition module 820, the prediction module 830, the determination module 840, the marking module 850, the preprocessing module 860, and the correction module 870 can be at least partially implemented as hardware circuits, such as field-programmable gate arrays (FPGAs), programmable logic arrays (PLAs), systems-on-a-chip, systems-on-a-substrate, systems-on-package, application-specific integrated circuits (ASICs), or any other reasonable means of integrating or packaging circuits, or implemented in software, hardware, or firmware, or in any suitable combination of any of these three implementation methods. Alternatively, at least one of the first acquisition module 810, the second acquisition module 820, the prediction module 830, the determination module 840, the marking module 850, the preprocessing module 860, and the correction module 870 can be at least partially implemented as computer program modules, which can perform corresponding functions when the computer program module is run.

[0155] It should be noted that the loan prediction device part in the embodiments of this disclosure corresponds to the loan prediction method part in the embodiments of this disclosure, and their specific implementation details and the resulting technical effects are the same, which will not be repeated here.

[0156] Figure 13 A block diagram of an electronic device suitable for implementing the methods described above, according to embodiments of the present disclosure, is illustrated schematically. Figure 13 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments disclosed herein.

[0157] like Figure 13As shown, an electronic device 1300 according to an embodiment of the present disclosure includes a processor 1301, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1302 or a program loaded from a storage portion 1308 into a random access memory (RAM) 1303. The processor 1301 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 1301 may also include onboard memory for caching purposes. The processor 1301 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present disclosure.

[0158] RAM 1303 stores various programs and data required for the operation of electronic device 1300. Processor 1301, ROM 1302, and RAM 1303 are interconnected via bus 1304. Processor 1301 performs various operations of the method flow according to embodiments of the present disclosure by executing programs in ROM 1302 and / or RAM 1303. It should be noted that the programs may also be stored in one or more memories other than ROM 1302 and RAM 1303. Processor 1301 may also perform various operations of the method flow according to embodiments of the present disclosure by executing programs stored in said one or more memories.

[0159] According to embodiments of this disclosure, the electronic device 1300 may further include an input / output (I / O) interface 1305, which is also connected to a bus 1304. The electronic device 1300 may also include one or more of the following components connected to the I / O interface 1305: an input section 1306 including a keyboard, mouse, etc.; an output section 1307 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 1308 including a hard disk, etc.; and a communication section 1309 including a network interface card such as a LAN card, modem, etc. The communication section 1309 performs communication processing via a network such as the Internet. A drive 1310 is also connected to the I / O interface 1305 as needed. A removable medium 1311, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 1310 as needed so that computer programs read from it can be installed into the storage section 1308 as needed.

[0160] According to embodiments of this disclosure, the method flow according to embodiments of this disclosure can be implemented as a computer software program. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable storage medium, the computer program containing program code for performing the methods shown in the flowchart. In such embodiments, the computer program can be downloaded and installed from a network via communication section 1309, and / or installed from removable medium 1311. When the computer program is executed by processor 1301, it performs the functions defined in the system of embodiments of this disclosure. According to embodiments of this disclosure, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0161] This disclosure also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs that, when executed, implement the method according to the embodiments of this disclosure.

[0162] According to embodiments of this disclosure, the computer-readable storage medium can be a non-volatile computer-readable storage medium. Examples include, but are not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this disclosure, the computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0163] For example, according to embodiments of this disclosure, a computer-readable storage medium may include one or more memories other than ROM 1302 and / or RAM 1303 described above.

[0164] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions. Those skilled in the art will understand that the features recited in the various embodiments and / or claims of this disclosure can be combined and / or combined in various ways, even if such combinations or combinations are not expressly described in this disclosure. In particular, the features described in the various embodiments and / or claims of this disclosure may be combined and / or combined in various ways without departing from the spirit and teachings of this disclosure. All such combinations and / or combinations fall within the scope of this disclosure.

Claims

1. A loan forecasting method, comprising: Obtain authorization from the user being tested to access their user information; After obtaining authorization from the user under test to acquire user information, the user information of the user under test is acquired. The risk identification model is used to predict the risk information of the user to be tested based on the user information; wherein, the risk identification model is constructed based on the loan transaction data of approved users and the loan transaction data of rejected users after being marked, and the loan transaction data of rejected users is marked by importance sampling and using the loan transaction data of approved users; The loan grant information for the user under test is determined based on the risk information. The step of marking the loan transaction data of the rejected users by sampling importance and using the loan transaction data of the approved users includes: Obtain the loan transaction data of the approved users and the loan transaction data of the rejected users; The first model is used to predict the first probability that an approved or rejected user is a fulfilling customer, based on the loan transaction data of the approved user and the loan transaction data of the rejected user. The second model is used to predict the second probability of an approved or rejected user being a customer whose loan is approved, based on the loan transaction data of the approved user and the loan transaction data of the rejected user. The loan transaction data of the approved users and the loan transaction data of the rejected users are input into the third model to construct a fully connected graph, wherein each node of the fully connected graph represents the loan transaction data of an approved user or the loan transaction data of a rejected user. Determine the first node, which represents the node that approves the user's loan transaction data; Determine the second node, which is the node closest to the first node and represents the loan transaction data of the rejected user; Calculate the similarity between the rejecting user corresponding to the second node and the approving user corresponding to the first node based on the first probability and the second probability corresponding to the first node; The loan transaction data of approved users with a similarity greater than a preset value are used to mark the loan transaction data of corresponding rejected users.

2. The loan forecasting method according to claim 1, wherein, Based on the loan transaction data of the approved users, a logistic regression model is constructed to obtain the first model.

3. The loan forecasting method according to claim 1, wherein, Based on the loan transaction data of approved users and the loan transaction data of rejected users, a logistic regression model is constructed to obtain the second model.

4. The loan forecasting method according to claim 1, wherein, The third model includes the label propagation model.

5. The loan forecasting method according to claim 1, wherein, The similarity between the rejecting user corresponding to the second node and the approving user corresponding to the first node is calculated based on the first probability and the second probability corresponding to the first node, including: Calculate the difference between the first probability corresponding to the first node and the first probability corresponding to the second node; The similarity between the rejecting user corresponding to the second node and the approving user corresponding to the first node is calculated based on the difference and the reciprocal of the second probability corresponding to the first node.

6. The loan forecasting method according to claim 1, further comprising: Preprocessing of loan transaction data for approved users and loan transaction data for rejected users includes: The loan transaction data of the approved users and the loan transaction data of the rejected users are deleted and then filled in; Delete highly correlated features from the loan transaction data of the approved users and the loan transaction data of the rejected users, respectively; and Normalize and perform variance filtering on the loan transaction data of approved users and rejected users.

7. The loan forecasting method according to claim 1, further comprising: The risk identification model is corrected based on the transaction data of the user to be tested.

8. A loan forecasting device, comprising: The first acquisition module is used to obtain the authorization of the user under test to acquire user information; The second acquisition module is used to acquire the user information of the user under test after obtaining the authorization of the user under test to acquire user information; The prediction module is used to predict the risk information of the user to be tested by using a risk identification model and based on the user information; wherein, the risk identification model is constructed based on the loan transaction data of approved users and the loan transaction data of rejected users after being marked, and the loan transaction data of rejected users is marked by importance sampling and using the loan transaction data of approved users; The determination module is used to determine the loan grant information of the user under test based on the risk information; The step of marking the loan transaction data of the rejected users by sampling importance and using the loan transaction data of the approved users includes: Obtain the loan transaction data of the approved users and the loan transaction data of the rejected users; The first model is used to predict the first probability that an approved or rejected user is a fulfilling customer, based on the loan transaction data of the approved user and the loan transaction data of the rejected user. The second model is used to predict the second probability of an approved or rejected user being a customer whose loan is approved, based on the loan transaction data of the approved user and the loan transaction data of the rejected user. The loan transaction data of the approved users and the loan transaction data of the rejected users are input into the third model to construct a fully connected graph, wherein each node of the fully connected graph represents the loan transaction data of an approved user or the loan transaction data of a rejected user. Determine the first node, which represents the node that approves the user's loan transaction data; Determine the second node, which is the node closest to the first node and represents the loan transaction data of the rejected user; Calculate the similarity between the rejecting user corresponding to the second node and the approving user corresponding to the first node based on the first probability and the second probability corresponding to the first node; The loan transaction data of approved users with a similarity greater than a preset value are used to mark the loan transaction data of corresponding rejected users.

9. An electronic device, comprising: One or more processors; Storage device for storing one or more programs. Wherein, when the one or more programs are executed by the one or more processors, the one or more processors perform the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, It stores executable instructions that, when executed by a processor, cause the processor to perform the method according to any one of claims 1 to 7.

11. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the method according to any one of claims 1 to 7.

Images