Bank lending data matching method based on deep learning
Through the bank lending data matching method based on deep learning, a lending data matching model is built and precise matching is solved, and the problem of low lending data processing efficiency in the existing technology is achieved, and efficient matching and transaction rate improvement is achieved.
Patent Information
- Application Number
- CN202510023795.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-07
- Publication Date
- 2025-05-06
AI Technical Summary
Existing bank lending data processing technology is difficult to achieve accurate matching of the needs of both lenders, resulting in low matching efficiency and insufficient transaction rate.
A bank lending data matching method based on deep learning is adopted, and a loan data matching model is constructed through a strategy gradient deep learning algorithm, a large amount of historical excellent lending user information is used for training, a routing module is established to match the needs of loan users, and further accurate matching is made through the fund party's custom preference module.
It achieves accurate matching between loan users and asset institutions, avoids matching omissions, increases transaction rates, and hands over the decision-making power to loan users, improving user experience.
Smart Images
Figure CN119941389A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of bank loan data processing, and in particular to a bank loan data matching method based on deep learning. Background Art
[0002] With the rapid development of Internet finance in my country, the number of lending entities on the Internet has increased significantly. Many bank-type funding parties have cooperated by accessing traffic platforms and other asset-side platforms to assist in lending, and by paying the asset-side technical service fees. However, because the funding parties now simply use regions, age, degree, etc. to divide customers into simple customer groups to lend to customers, it is difficult to use this method to fully utilize customer portraits for comprehensive and accurate screening, resulting in a large number of rejected users. However, to accurately profile loan customers, a large amount of customer data is required, resulting in low matching efficiency, which makes it impossible for funding parties to fully utilize users attracted by the platform, and borrowers cannot quickly find suitable funding institutions. Summary of the invention
[0003] The present invention aims to provide a bank loan data matching method based on deep learning to solve the problem that the current needs of borrowers and lenders cannot be fully matched through simple customer group division.
[0004] In order to solve the above problems, the technical solution adopted by the present invention is: a bank loan data matching method based on deep learning, comprising the following steps:
[0005] Step 1: Obtain a large amount of loan user information with loan needs through the loan flow platform;
[0006] Step 2: All the information of the loan user is directly input into the routing module. The routing module establishes a loan data matching model based on the loan user's needs, and obtains a list of asset institutions by running the loan data matching model. The establishment process of the loan data matching model includes: firstly, a deep learning model is constructed by using a policy gradient deep learning algorithm, and then a large amount of historical excellent loan user information with no overdue and bad debt is input into the deep learning model for training. When the deep learning model converges, a loan data matching model is obtained, and the converged loan data matching model is written into the routing module through machine language;
[0007] Step 3: Input the asset institution list into the preference module of the funding party's custom preference. The preference module outputs the funding party list, which is provided to the loan user for selection.
[0008] The principle and beneficial effect of this method is that this method no longer uses a simple routing strategy to avoid screening out some loan customers who are not the main determining factors, resulting in matching omissions. For example, loan customers are rejected because age, marital status, address discrepancy, and failure to pass a certain bank's preference are not the main determining factors. This solution is to run a loan data matching model in the routing module, with the loan user as the target to first match the list of asset institutions that meet the loan user, so as to avoid missing the funding party that may complete the loan for the loan user, and then input the list of asset institutions into the preference module to further accurately match all funding parties that may lend to the loan user; finally, the loan user decides to choose the funding party, so that the model can quickly complete the matching of a loan user with multiple asset institutions, and further hand over the decision-making power to the loan user, thereby increasing the transaction rate.
[0009] Preferably, the gradient deep learning algorithm in step 2 is a PPO deep reinforcement learning algorithm. Compared with some traditional policy gradient algorithms, PPO can more effectively utilize the collected data and reduce the number of samples required to obtain a good strategy.
[0010] The PPO (Proximal Policy Optimization) deep reinforcement learning algorithm is a policy gradient method proposed by John Schulman et al. in 2017.
[0011] Preferably, the loan user information input into the routing module is a standardized parameter after standardization. For example, if the user information is [bachelor's degree, married, Beijing, 20000], the vector formed after standardization is [2, 1, 1, 0.2]. Standardizing and transforming the loan user information makes the calculation process more expressive.
[0012] Preferably, the parameters input into the routing module include at least user attribute variables, loan amount, annualized interest rate, and response time. User attribute variables, loan amount, annualized interest rate, and response time parameters can match eligible funding parties more quickly.
[0013] Preferably, in step 2, the loss function of the loan data matching model is constructed by using the non-performing loan rate, response time, and overdue rate parameters. The loss function is the expectation of the reward, and the larger the expectation, the better. Therefore, the inverse value of the loss function is taken as the training function of the gradient descent, that is, the repayment ability is used as the parameter of the loss function to converge the loan data matching model.
[0014] Preferably, in step 2, the loan data matching model is trained regularly for loan customers with no overdue payments or bad debts, so as to obtain a loan data matching model with better convergence. BRIEF DESCRIPTION OF THE DRAWINGS
[0015] Figure 1 This is a flowchart of a bank loan data matching method based on deep learning;
[0016] Figure 2 Matching data processing function module diagram for bank loan data. DETAILED DESCRIPTION
[0017] Example 1
[0018] The following is further described in detail by specific implementations, but the implementations of the present invention are not limited thereto. Unless otherwise specified, the technical means used in the following implementations are conventional means well known to those skilled in the art.
[0019] like Figure 1 As shown in FIG. 1 , the bank loan data matching method based on deep learning includes the following steps:
[0020] Step 1 S1: Obtain a large amount of loan user information with loan needs through the loan flow platform;
[0021] Step 2 S2: All the information of the loan user is standardized and transformed and then input into the routing module. The input parameters at least include user attribute variables, loan amount, annualized interest rate, and response time.
[0022] The routing module establishes a loan data matching model based on loan user needs, and obtains a list of asset institutions by running the loan data matching model; the establishment process of the loan data matching model includes: firstly, using the PPO deep reinforcement learning algorithm to build a deep learning model,
[0023] The loss function of the loan data matching model is constructed with the parameters of non-performing loan rate, response time, and overdue rate. A large amount of historical excellent loan user information with no overdue and bad debts is input into the deep learning model for training. When the deep learning model converges, the loan data matching model is obtained, and the converged loan data matching model is written into the routing module through machine language.
[0024] Regularly train the loan data matching model for loan customers with no overdue or bad debts, so as to obtain a loan data matching model with better convergence.
[0025] Step 3 S3: Input the asset institution list into the preference module of the funding party's custom preference, and the preference module outputs the funding party list, which is provided to the loan user for selection.
[0026] This method no longer uses a simple routing strategy to avoid screening out some loan customers who are not the main determining factors, resulting in matching omissions. For example, loan customers are rejected because age, marital status, address discrepancy, and failure to pass a certain bank's preference are not the main determining factors. This solution is to run a loan data matching model in the routing module, with the loan user as the target to first match the list of asset institutions that meet the loan user, so as to avoid missing the funding party that may complete the loan for the loan user, and then input the list of asset institutions into the preference module to further accurately match all funding parties that may lend to the loan user; finally, the loan user decides to choose the funding party, so that the model can quickly complete the matching of a loan user with multiple asset institutions, and further hand over the decision-making power to the loan user, thereby increasing the transaction rate.
[0027] Example 2
[0028] like Figure 2 As shown, the bank loan data matching data processing function module compiles the loan data matching steps into a computer executable software program according to the data transmission method and function of the function module.
[0029] The bank loan data matching data processing function module is a virtual function on the trading system software, which is used to execute the bank loan data matching method based on deep learning. The bank loan data matching data processing function module includes a request processing module, a routing module and a preference module.
[0030] The request processing module collects loan requests from loan users from the network interface and transmits the loan request data to the routing module. In the routing module, the routing module performs standard parameterization on the loan information in the loan request data.
[0031] The loan data matching model is run in the routing module. The loan data matching model first uses the policy gradient deep learning algorithm to build a deep learning model, and then inputs a large amount of historical excellent loan user information with no overdue and bad debts into the deep learning model for training. When the deep learning model converges, the loan data matching model is obtained, and the converged loan data matching model is written into the routing module through machine language.
[0032] The routing module outputs the matching loan users and asset institutions to form an asset institution list. Further accurate matching is performed by running the preference module to filter the preferences of all asset institutions in the asset institution list to obtain a funding party list, which is a list of all funding parties that can provide loans to loan users.
[0033] Loan users can further select funding parties that better suit their preferences and ultimately achieve an accurate match between borrowers and lenders.
[0034] The above is only an embodiment of the present invention, and the common knowledge such as the known specific technical solutions and / or characteristics in the solution is not described in detail here. It should be pointed out that for those skilled in the art, without departing from the technical solution of the present invention, several modifications and improvements can be made, which should also be regarded as the protection scope of the present invention, and these will not affect the effect of the implementation of the present invention and the practicality of the patent. The scope of protection required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the specification can be used to interpret the content of the claims.
Claims
1. A bank loan data matching method based on deep learning, characterized in that: The following steps are included: Step 1: Obtain a large amount of loan user information with loan needs through the loan flow platform; Step 2: All the information of the loan user is directly input into the routing module. The routing module establishes a loan data matching model based on the loan user's needs, and obtains a list of asset institutions by running the loan data matching model. The establishment process of the loan data matching model includes: firstly, a deep learning model is constructed by using a policy gradient deep learning algorithm, and then a large amount of historical excellent loan user information with no overdue and bad debt is input into the deep learning model for training. When the deep learning model converges, a loan data matching model is obtained, and the converged loan data matching model is written into the routing module through machine language; Step 3: Input the asset institution list into the preference module of the funding party's custom preference. The preference module outputs the funding party list, which is provided to the loan user for selection.
2. The message queue traffic switching method based on multiple servers according to claim 1 is characterized in that ,The gradient deep learning algorithm described in step 2 is the PPO deep reinforcement learning algorithm.
3. The message queue traffic switching method based on multiple servers according to claim 2 is characterized in that: The loan user information input into the routing module is standardized parameters after standardization and deformation.
4. The message queue traffic switching method based on multiple servers according to claim 3 is characterized in that: The parameters input into the routing module include at least user attribute variables, loan amount, annualized interest rate, and response time.
5. The message queue traffic switching method based on multiple servers according to claim 4 is characterized in that: In the step 2, the loss function of the loan data matching model is constructed by using the non-performing loan rate, response time, and overdue rate parameters.
6. The message queue traffic switching method based on multiple servers according to claim 5 is characterized in that: In the step 2, the loan data matching model is continuously trained on a regular basis for loan customers with no overdue payments or bad debts.