WO2019080407A1

WO2019080407A1 - Credit evaluation method, apparatus and device, and computer readable storage medium

Info

Publication number: WO2019080407A1
Application number: PCT/CN2018/075663
Authority: WO
Inventors: 李天平
Original assignee: 深圳壹账通智能科技有限公司
Priority date: 2017-10-25
Filing date: 2018-02-07
Publication date: 2019-05-02
Also published as: CN107886425A

Abstract

Disclosed is a credit evaluation method. The method comprises: acquiring historical credit data, and constructing an AI risk control engine according to the historical credit data; when a loan request is received, acquiring pre-loan user data of a loan user corresponding to the loan request, and inputting the pre-loan user data to a pre-set expert engine and the AI risk control engine respectively for credit evaluation; and acquiring an expert evaluation conclusion of the pre-set expert engine and an AI evaluation conclusion of the AI risk control engine, and generating a corresponding decision and suggestion according to the expert evaluation conclusion and the AI evaluation conclusion. Further disclosed are a credit evaluation apparatus and device, and a computer readable storage medium. In the present application, a credit decision and suggestion are acquired by means of double-engine evaluation, which greatly reduces adverse influence of limitations of human thought on engine construction and a credit evaluation process, improving the evaluation capability of a risk control model and effectively reducing the rate of bad loans.

Description

Credit evaluation method, device, device and computer readable storage medium

This application claims the priority of the Chinese Patent Application filed on October 25, 2017, the Chinese Patent Application No. 201711010536.1, entitled "Credit Evaluation Method, Apparatus, Equipment, and Computer Readable Storage Medium", the entire contents of which are hereby incorporated by reference. The citation is incorporated in the application.

Technical field

The present application relates to the field of financial credit, and in particular, to a credit evaluation method, apparatus, device, and computer readable storage medium.

Background technique

The traditional method of credit approval is to manually review the borrowing request of the borrowing user by the staff of the financial institution. In the process of reviewing, the staff needs to check the large amount of data information of the borrowing user to determine the risk of the loan transaction, for example, the borrowing user borrows How much money can be paid, whether it can be repaid within the loan period, etc. With the development of science and technology, financial institutions began to use computer risk control systems to evaluate credit requests. The wind control model is an important part of the entire risk control system. It is mainly used to identify, analyze and evaluate user data. Thereby the conclusion of the evaluation is obtained.

The current wind control model mainly uses the static rule engine of drools, and introduces expert rules to evaluate the credit request (that is, convert the human experience into a computer logic algorithm, and then use the computer to identify the credit request). However, the human experience is not necessarily correct. The human experience is susceptible to subjective factors, which affects the accuracy of the risk control model assessment, making the wind control model unable to give a reasonable loan plan, or even difficult to identify conscious fraud. Thereby increasing the bad debt rate of the loan.

Application content

The main purpose of the present application is to provide a credit evaluation method, apparatus, device and computer readable storage medium, which aim to improve the evaluation capability of the wind control model and reduce the loan bad debt rate.

To achieve the above object, the present application provides a credit evaluation method, and the credit evaluation method includes the following steps:

Obtain historical credit data and construct an AI wind control engine based on the historical credit data;

Upon receiving the loan request, acquiring the borrowing request corresponding to the user's pre-lending data of the borrowing user, and inputting the user pre-lending data into the preset expert engine and the AI wind control engine for credit evaluation;

Obtaining an expert evaluation conclusion of the preset expert engine and an AI evaluation conclusion of the AI wind control engine, and generating corresponding decision suggestions based on the expert evaluation conclusion and the AI evaluation conclusion.

Optionally, the historical credit data includes historical pre-lending data and historical post-loan data.

The step of acquiring historical credit data and constructing an AI wind control engine according to the historical credit data includes:

The historical pre-lending data and the historical post-lost data are respectively used as an input training set and an output training set, and the initial machine learning model is analyzed and trained based on the input training set and the output training set to obtain an AI wind control engine.

Optionally, the initial machine learning model is cold initiated by expert rules.

Optionally, the input element in the input training set includes a multi-dimensional vector x of dimension m, and the output element in the output training set includes a loan non-performing rate y,

The analyzing and training the initial machine learning model based on the input training set and the output training set, and the steps of obtaining the AI wind control engine include:

Combining the input element and the output element to define m multi-dimensional training vector groups

{(x ¹ , y ¹ ), (x ² , y ² ), ..., (x ^m , y ^m )};

Learning the multidimensional training vector in the multi-dimensional training vector group based on the gradient lifting tree GBDT algorithm to obtain a corresponding decision tree

T(x; θ _m ),

Where θ _m is a parameter of the decision tree;

Linearly combining the decision trees to obtain a linear combination tree

The cost function of the linear combination number is

f(x)=l(h(x,D),Y),

Wherein the h(x, D) is a hypothesis function, which is a decision result of the decision tree, and the first-order expansion of the Taylor formula of the cost function f(x) is:

among them,

The weighted sum of the first t decision trees,

For the fastest falling iteration formula,

To assume the gradient of the function;

The AI wind control engine is obtained according to the linear combination tree and the cost function.

Optionally, when receiving the loan request, acquiring the loan request corresponding to the user's pre-lending data of the borrowing user, and inputting the user pre-lending data into the preset expert engine and the AI wind control engine for credit The steps of the assessment include:

Upon receiving the loan request, generating a corresponding data acquisition request according to the user identifier and the authorization license included in the loan request, and sending the data acquisition request to the data management system to obtain corresponding user pre-lending data;

Receiving user pre-lending data fed back by the data management system, and inputting the pre-lending data of the user into a preset expert engine and the AI wind control engine for credit evaluation.

Optionally, the step of acquiring the expert evaluation conclusion of the preset expert engine and the AI evaluation conclusion of the AI wind control engine, and generating corresponding decision suggestions based on the expert evaluation conclusion and the AI evaluation conclusion includes:

Obtaining an expert evaluation conclusion of the preset expert engine and an AI evaluation conclusion of the AI wind control engine, and performing a weighting operation based on the expert evaluation conclusion and the AI evaluation conclusion to obtain a credit line of the borrowing user;

A corresponding loan recommendation is generated according to the credit line, wherein the loan recommendation includes a loan amount and a loan period.

Optionally, after the step of acquiring the expert evaluation result of the preset expert engine and the AI evaluation result of the AI wind control engine, and generating a corresponding decision suggestion based on the expert evaluation conclusion and the AI evaluation conclusion, include:

Obtaining user loan data of the borrowing user, and combining the user pre-lending data and the user post-loan data to obtain an iterative training sample;

The AI wind control engine is iteratively trained by the iterative training samples to adjust the AI wind control engine.

In addition, in order to achieve the above object, the present application further provides a credit evaluation apparatus, where the credit evaluation apparatus includes:

An engine construction module, configured to acquire historical credit data, and construct an AI wind control engine according to the historical credit data;

a credit evaluation module, configured to acquire, before receiving the loan request, the pre-lending data of the borrowing user corresponding to the borrowing request, and input the pre-lending data of the user into a preset expert engine and the AI wind control engine for credit Evaluation

The suggestion generating module is configured to obtain an expert evaluation conclusion of the preset expert engine and an AI evaluation conclusion of the AI wind control engine, and generate corresponding decision suggestions based on the expert evaluation conclusion and the AI evaluation conclusion.

Further, in order to achieve the above object, the present application further provides a credit evaluation apparatus including a processor, a memory, and a credit evaluation program stored on the memory and executable by the processor, wherein When the credit evaluation program is executed by the processor, the following steps are implemented:

In addition, in order to achieve the above object, the present application further provides a computer readable storage medium having a credit evaluation program stored thereon, wherein when the credit evaluation program is executed by a processor, the following steps are implemented:

Upon receiving the loan request, acquiring the loan request corresponding to the user's pre-lending data of the borrowing user, and inputting the user pre-lending data into the preset expert engine and the AI wind control engine for credit evaluation;

DRAWINGS

1 is a schematic structural diagram of hardware of a credit evaluation apparatus involved in an embodiment of the present application;

2 is a schematic flow chart of a first embodiment of a credit evaluation method according to the present application;

3 is the user pre-credit data of the borrowing user corresponding to the borrowing request, and input the user pre-lending data into the preset expert engine and the AI wind control engine respectively when receiving the borrowing request. Schematic diagram of the detailed process of credit evaluation;

4 is a schematic flow chart of a second embodiment of a credit evaluation method according to the present application;

FIG. 5 is a schematic diagram of functional modules of a first embodiment of a credit evaluation apparatus according to the present application.

The implementation, functional features and advantages of the present application will be further described with reference to the accompanying drawings.

Detailed ways

It is understood that the specific embodiments described herein are merely illustrative of the application and are not intended to be limiting.

The credit evaluation method according to the embodiment of the present application is mainly applied to a credit evaluation device. Referring to FIG. 1, FIG. 1 is a schematic structural diagram of hardware of a credit evaluation apparatus involved in an embodiment of the present application. The credit evaluation apparatus in the embodiment of the present application may include a processor 1001 (for example, a CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is configured to implement connection communication between the components; the user interface 1003 may include a display, an input unit such as a keyboard; the network interface 1004 may optionally include a standard wired interface and a wireless interface. (such as WI-FI interface); the memory 1005 may be a high-speed RAM memory, or may be a non-volatile memory, such as a disk storage, and the memory 1005 may alternatively be a storage device independent of the processor 1001 described above. . It will be understood by those skilled in the art that the hardware structure of the credit evaluation device shown in FIG. 1 does not constitute a limitation of the credit evaluation device, and may include more or less components than those illustrated, or combine some components, or different. Parts layout.

With continued reference to FIG. 1, the memory 1005 as a computer readable storage medium of FIG. 1 can include an operating system, a network communication module, and a credit evaluation program. The network communication module is mainly used to connect to the database and perform data communication with the database; and the processor 1001 can call the credit evaluation program stored in the memory 1005 and execute various embodiments of the credit evaluation method of the embodiment of the present application.

The application provides a credit evaluation method.

Referring to FIG. 2, FIG. 2 is a schematic flowchart of a first embodiment of a credit evaluation method according to the present application.

In this embodiment, the credit evaluation method includes the following steps:

Step S10: Obtain historical credit data, and construct an AI wind control engine according to the historical credit data;

In this embodiment, a credit evaluation method is proposed, which adopts a dual-core wind control engine to evaluate and make decisions on borrowing requests, reduces the adverse effects of human thinking limitations on the wind control model, and improves the accuracy of the risk control system, thereby reducing credit bad debts. rate.

In this embodiment, it is first necessary to construct an AI wind control engine, which is different from a conventional expert engine. For the expert engine, the risk control expert or related staff will analyze the experience (rule) after analyzing a large amount of loan data, and then convert these experience summaries into computer logic algorithms, thus forming the evaluation rules of the expert engine. It is called expert rule); but human experience is not necessarily correct. Human experience is easily influenced by subjective factors, which makes the expert rules have certain defects and reduces the evaluation ability of the risk control engine using expert rules. Therefore, the AI wind control engine construction process in this embodiment does not completely adopt the expert rule, but constructs the AI wind control engine by means of machine learning. Among them, machine learning means not relying on humans to sum up experience and input logic. Humans only need to input a large amount of sample data (including input sample data and output sample data) to the computer, and then the computer itself summarizes the data conversion logic or mapping. Convert the relationship and summarize the corresponding logic code to get a data conversion rule to form the corresponding evaluation engine. Specifically, in this embodiment, the database is first queried to obtain historical credit data, and the historical credit data is a historical lending record of the financial institution, including behavior data provided by the user before the loan (such as credit status, personal financial resources, bank transaction records, social security payment status). , call data, etc.), and the repayment history and post-loan behavior data of the user after the loan; then the behavior data of the user before the loan as the input training set, the post-loan repayment record and the post-loan behavior data as the output training set And inputting the input training set and the output training set to an initial computer learning model, and analyzing and training the initial machine learning model based on the input training set and the output training set, and automatically calculating the two data structure forms and data contents by a computer The aspects are compared and analyzed, the conversion rules are summarized, the transformation logic is summarized, and the AI wind control engine is constructed.

Further, since machine learning often requires a large amount of training data to demonstrate excellent learning ability, but for credit data (especially in the evaluation of personal credit), the number of users is limited and available to users. The effective information is more limited, so the expert rule can be used as the cold start zero of the initial machine learning model, that is, the expert logic can be used as the initial logic of the initial machine learning model, and then the historical credit data is used to continuously train and construct AI wind control engine.

It should be noted that the process of constructing the AI wind control engine in this embodiment adopts algorithms including, but not limited to, logistic regression, decision tree, GBDT, SVM, and the like. For example, for GBDT (Gradient Lifting Tree), the sample set includes real estate, income, age, education, etc., and a combination of m vector groups can be defined, where x is a multidimensional vector and y is the non-performing rate of the loan. That is

{(x ¹ , y ¹ ), (x ² , y ² ),...,(x ^m ,y ^m )} 1

Depending on the combination, different weak models (short decision trees) can be learned and then linearly combined into a single function:

In Equation 2, T(x; θ _m ) represents a decision tree, which can be regarded as a regression tree;

θ _m is a parameter of the decision tree;

m is the number of decision trees.

The cost function of this GBDT is:

f(x)=l(h(x,D),Y) 3

In Equation 3, h(x, D) is a hypothesis function, which is the decision result of the decision tree;

The first-order expansion of the Taylor formula of f(x) is:

In the 4th formula,

Is the weighted sum of the top t decision trees;

Is the steepest drop iteration formula;

Assume the gradient of the function.

It can be seen from the above that the idea of the loss function of GBDT is similar to the logistic regression, except that the latter function uses a bijective sigmoid function, and this embodiment uses CART here. It can be seen that the method of calculating the GBDT parameter is similar to the logistic regression, and the gradient descent method can also be used. According to the above, a CART-based linear combination tree GDBT (such as Equation 2) and a corresponding cost function (such as Equation 3) can be obtained, that is, the AI wind control engine.

Of course, for the trained AI wind control engine, the staff can make manual adjustments to improve the AI wind control engine.

Step S20: Upon receiving the loan request, obtain the user loan pre-lending data corresponding to the borrowing user, and input the user pre-lending data into the preset expert engine and the AI wind control engine for credit evaluation;

In this embodiment, after the AI wind control engine is trained, the AI wind control engine can be used for loan evaluation. It should be noted that, in this embodiment, not only the AI wind control engine is used for evaluation, but the AI wind control engine and the preset expert engine are simultaneously used, and the credit evaluation is performed by using the dual-core wind control method. Among them, the meaning of the preset expert engine is as described above. It is the experience (regular) summary of the analysis of a large amount of loan data by the risk control expert or related staff, and then the summary of these experiences is converted into the logic algorithm of the computer, thus constituting the Expert engine evaluation rules.

Specifically, the financial institution's risk control model obtains the relevant pre-lending data of the borrowing user when receiving the borrowing request from the borrowing user (the borrowing request includes the borrowing amount and the borrowing period, etc.) (pre-loan data includes the borrower's Credit status, personal financial resources, bank record, social security payment, call data, etc., and input these risk control data into the preset expert engine and AI wind control engine respectively, and respectively through the preset expert engine and AI wind control engine Evaluate the loan request.

Step S30: Obtain an expert evaluation conclusion of the preset expert engine and an AI evaluation conclusion of the AI wind control engine, and generate a corresponding decision suggestion based on the expert evaluation conclusion and the AI evaluation conclusion.

In this embodiment, after the loan request is evaluated by the preset expert engine and the AI wind control engine respectively, the corresponding expert evaluation conclusion and the AI evaluation conclusion are obtained, and the combination of the expert evaluation and the AI evaluation conclusion can be used to obtain the decision suggestion ( Such as the amount of the loan, the loan period, etc.).

In the specific implementation, the expert evaluation conclusion and the AI evaluation conclusion may be represented by a score, and the score ranges from 0 to 100; the higher the score, the lower the bad debt rate of the borrowing user's loan request; When the loan request is evaluated by the preset expert engine and the AI wind control engine, and the expert score and the AI score are obtained, the expert score and the AI score need to be weighted and run. a comprehensive evaluation value, the meaning of the comprehensive evaluation value is similar to the evaluation score obtained by the single-core evaluation, and the range is also between 0 and 100. The higher the score, the lower the bad debt rate of the borrowing user's loan request; Different score segments, which correspond to different credit lines, the credit limit represents the recommended maximum amount and maximum loan period for lending to the borrower; by calculating the comprehensive evaluation value of the borrower, the borrower can be determined. The credit line, which generates the corresponding loan proposal (including the loan amount and the loan period).

Of course, in the specific implementation, for the expert evaluation conclusion and the AI evaluation conclusion, the evaluation conclusion can be divided into two points (ie, agreeing to release the money and rejecting the two points of the loan), and the generation process of the decision suggestion can also be implemented in other ways. It does not constitute a limitation on the credit evaluation method of the present embodiment.

Further, in this embodiment, after generating the decision proposal, the wind control model may also generate a corresponding credit evaluation report. The credit evaluation report includes the pre-lending data of the borrowing users used in the evaluation process, as well as the expert evaluation conclusions obtained from the pre-lending data evaluation, the AI evaluation conclusions, and the final decision-making recommendations. In the credit evaluation report, since a large amount of data is involved, the display manner of the data can be customized. For example, the pre-lending data includes the total expenditure of the borrower in the first three months, and the monthly expenditure includes hydropower expenditure, telephone business expenditure, credit card consumption, etc. For each monthly expenditure, a pie chart can be used. The way to display, to facilitate decision-makers to understand the monthly investment in different things for borrowing users; and for the three-month total expenditure situation, you can use a bar graph to display, so that decision-makers can understand the borrowing users every month. The trend of spending. Of course, the display mode of different types of data can be configured by the decision maker according to the actual situation.

In this embodiment, the historical credit data is obtained, and the AI wind control engine is constructed according to the historical credit data; when the borrowing request is received, the borrowing request is obtained corresponding to the user's pre-lending data of the borrowing user, and the user is The pre-lending data is respectively input into a preset expert engine and the AI wind control engine for credit evaluation; obtaining an expert evaluation conclusion of the preset expert engine and an AI evaluation conclusion of the AI wind control engine, and based on the expert evaluation conclusion And the AI assessment conclusions generate corresponding decision suggestions. In the above manner, the embodiment constructs the AI wind control engine by the machine learning method, and simultaneously evaluates the loan request through the AI wind control engine and the expert rule engine, and obtains the decision suggestion by using the dual engine evaluation method, thereby greatly reducing the limitation of human thinking. The adverse effects of sex on the engine construction and credit evaluation process, improve the assessment ability of the risk control model, and effectively reduce the bad debt rate of loans.

Referring to FIG. 3, FIG. 3 is the user pre-credit data of the borrowing user corresponding to the borrowing request when the borrowing request is received, and the user pre-lending data is input into the preset expert engine and the AI respectively. A schematic diagram of the detailed process of the credit evaluation by the risk control engine.

Based on the embodiment shown in FIG. 2 above, in this embodiment, step S20 includes:

Step S21: When receiving the loan request, generate a corresponding data acquisition request according to the user identifier and the authorization permission included in the loan request, and send the data acquisition request to the data management system to obtain the corresponding user loan data;

Step S22: Receive user pre-lending data fed back by the data management system, and input the user pre-lending data into the preset expert engine and the AI wind control engine for credit evaluation.

In this embodiment, the pre-lending data of the borrowing user often involves the privacy of the borrowing user, and the pre-lending data needs to be obtained from other data sources. For example, the borrowing user's bank card consumption record needs to be obtained from the banking system. The borrower's billing data needs to be obtained from the telecommunications company; if the risk control model wants to obtain these pre-lending data, it needs to obtain the license and authorization of the borrowing user first. Specifically, when the borrowing user sends a loan request, the borrowing request includes the identity of the borrowing user (such as identity information such as the ID number), and the license information of the borrowing user (such as an electronic signature) to indicate the borrowing. Users allow the risk control model (financial institution) to obtain their own relevant pre-lending data for credit evaluation. Upon receiving the loan request, the wind control model generates a corresponding data acquisition request according to the user identification and the authorization, and sends the data acquisition request to the relevant data management system (such as a banking system, a telecommunications company, etc.). When receiving the data acquisition request, the data management system first obtains the user identifier therein, and determines which data acquisition request is related data of which user needs to be acquired; then, the authorization license is verified, and the authorized license is determined. The authenticity determines whether the risk control model requesting the data acquisition is a legal acquirer; when the authorization verification passes, the data management system will query the corresponding user pre-lending data and return the data to the wind control model. When the wind control model receives the user pre-loan data fed back by the data management system, the user pre-lending data can be separately input into the preset expert engine and the AI wind control engine for credit evaluation.

Referring to FIG. 4, FIG. 4 is a schematic flowchart of a second embodiment of a credit evaluation method according to the present application.

Based on the embodiment shown in FIG. 2, in this embodiment, after step S30, the method further includes:

Step S40: acquiring user loan data of the borrowing user, and combining the user pre-lending data and the user loan data to obtain an iterative training sample;

Step S50: Perform iterative training on the AI wind control engine through the iterative training samples to adjust the AI wind control engine.

In this embodiment, considering the development of time, different economic cycles and social factors may affect the AI rule (evaluation logic) of the AI engine, so that the effective interval of some factors generates a swing, such as a monthly income of 6000. Yuan, under different price conditions, the weight factor of the data is not the same; in order to make the AI engine adapt to the economic cycle and the actual situation of the society, the AI engine needs to be constantly revised and adjusted. Specifically, after determining the loan to the borrowing user, the risk control model will continue to obtain the post-loan data of the borrowing user (such as repayment records, consumption records, etc.); and then combine the pre-lending data obtained in step S20 with the post-loan data. The iterative training sample is continuously adjusted and adjusted by the AI engine of the iterative training sample pair, and the adjusted content includes changing the threshold of the parameter, changing the priority of some wind control data or evaluation factor, changing the weight of the data, etc.; for example, the AI engine In the strategy algorithm, the data weighting level is 1 to 5. For the data provided by the user when borrowing, the A data weighting level is 5, which means that the A data has a large reference value for judging whether the user can repay the loan on time; Some post-loan data found that the reference value of A data is not as big as expected. At this time, the AI engine will correct the original policy algorithm and reduce the weight level of A data. After adjusting the AI engine, the possibility of overdue or bad debts of the borrowing user may be given in the new loan cycle, and the actual post-loan data of the new cycle will be used for feedback correction. Through this iterative training method, the AI engine is continuously improved, and the evaluation accuracy is gradually improved.

In addition, the present application also provides a credit evaluation device.

Referring to FIG. 5, FIG. 5 is a schematic diagram of functional modules of a first embodiment of a credit evaluation apparatus according to the present application.

In this embodiment, the credit evaluation apparatus includes:

An engine construction module 10, configured to acquire historical credit data, and construct an AI wind control engine according to the historical credit data;

The credit evaluation module 20 is configured to acquire, before receiving the loan request, the pre-lending data of the borrowing user corresponding to the borrowing request, and input the pre-lending data of the user into the preset expert engine and the AI wind control engine respectively. Credit evaluation;

The suggestion generating module 30 is configured to obtain an expert evaluation conclusion of the preset expert engine and an AI evaluation result of the AI wind control engine, and generate corresponding decision suggestions based on the expert evaluation conclusion and the AI evaluation conclusion

Further, the historical credit data includes historical pre-lending data and historical post-loan data, and the engine construction module 10 is further configured to separately use the historical pre-lending data and historical post-loan data as an input training set and an output training set. And based on the above input training set and output training set, the initial machine learning model is analyzed and trained to obtain an AI wind control engine.

Further, the initial machine learning model is cold initiated by expert rules.

Further, the input element in the input training set includes a multi-dimensional vector x of dimension m, and the output element in the output training set includes a loan non-performing rate y.

The engine construction module 10 is further configured to combine the input element and the output element to define m multi-dimensional training vector groups.

{(x ¹ , y ¹ ), (x ² , y ² ), ..., (x ^m , y ^m )};

Learning the multi-dimensional training vector in the multi-dimensional training vector group based on the gradient lifting tree GBDT algorithm to obtain a corresponding decision tree

T(x; θ _m ),

Where θ _m is a parameter of the decision tree;

Linearly combining the decision trees to obtain a linear combination tree

The cost function of the linear combination number is

f(x)=l(h(x,D),Y),

among them,

The weighted sum of the first t decision trees,

For the fastest falling iteration formula,

To assume the gradient of the function;

Further, the credit evaluation module 20 includes:

a data obtaining unit, configured to generate a corresponding data acquisition request according to the user identifier and the authorization permission included in the loan request, and send the data acquisition request to the data management system to obtain a corresponding User pre-lending data;

The credit evaluation unit is configured to receive user pre-lending data fed back by the data management system, and input the user pre-lending data into the preset expert engine and the AI wind control engine for credit evaluation.

Further, the suggestion generating module 30 includes:

a weighting operation unit, configured to obtain an expert evaluation conclusion of the preset expert engine and an AI evaluation result of the AI wind control engine, and perform a weighting operation based on the expert evaluation conclusion and the AI evaluation conclusion to obtain the borrowing user credits;

The suggestion generating unit is configured to generate a corresponding loan proposal according to the credit limit, wherein the loan recommendation includes a loan amount and a loan period.

Further, the credit evaluation device further includes:

a data combination module, configured to obtain data of the user's credits of the borrowing user, and combine the user pre-lending data and the user's post-loan data to obtain an iterative training sample;

An iterative training module is configured to iteratively train the AI wind control engine through the iterative training samples to adjust the AI wind control engine.

The modules in the above credit evaluation apparatus correspond to the steps in the foregoing credit evaluation method embodiment, and the functions and implementation processes thereof are not repeated herein.

In addition, the present application also provides a computer readable storage medium.

A credit evaluation program is stored on the computer readable storage medium of the present application, and the computer readable storage medium stores a credit evaluation program, wherein when the credit evaluation program is executed by the processor, the steps of the credit evaluation method as described above are implemented.

For the method implemented when the credit evaluation program is executed, reference may be made to various embodiments of the credit evaluation method of the present application, and details are not described herein again.

The above is only a preferred embodiment of the present application, and is not intended to limit the scope of the patent application, and the equivalent structure or equivalent process transformations made by the specification and the drawings of the present application, or directly or indirectly applied to other related technical fields. The same is included in the scope of patent protection of this application.

Claims

A credit evaluation method, characterized in that the credit evaluation method comprises the following steps:

Obtain historical credit data and construct an AI wind control engine based on the historical credit data;

Upon receiving the loan request, acquiring the borrowing request corresponding to the user's pre-lending data of the borrowing user, and inputting the user pre-lending data into the preset expert engine and the AI wind control engine for credit evaluation;

Obtaining an expert evaluation conclusion of the preset expert engine and an AI evaluation conclusion of the AI wind control engine, and generating corresponding decision suggestions based on the expert evaluation conclusion and the AI evaluation conclusion.
The credit evaluation method according to claim 1, wherein said historical credit data includes historical pre-lending data and historical post-loan data.

The step of acquiring historical credit data and constructing an AI wind control engine according to the historical credit data includes:

The historical pre-lending data and the historical post-lost data are respectively used as an input training set and an output training set, and the initial machine learning model is analyzed and trained based on the input training set and the output training set to obtain an AI wind control engine.
The credit evaluation method according to claim 2, wherein said initial machine learning model is cold-started by an expert rule.
The credit evaluation method according to claim 2, wherein the input element in the input training set comprises a multidimensional vector x of dimension m, and the output element of the output training set comprises a loan non-performing rate y,

The analyzing and training the initial machine learning model based on the input training set and the output training set, and the steps of obtaining the AI wind control engine include:

Combining the input element and the output element to define m multi-dimensional training vector groups

{(x 1 , y 1 ), (x 2 , y 2 ), ..., (x m , y m )};

Learning the multidimensional training vector in the multi-dimensional training vector group based on the gradient lifting tree GBDT algorithm to obtain a corresponding decision tree

T(x; θ m ),

Where θ m is a parameter of the decision tree;

Linearly combining the decision trees to obtain a linear combination tree

The cost function of the linear combination number is

f(x)=l(h(x,D),Y),

Wherein the h(x, D) is a hypothesis function, which is a decision result of the decision tree, and the first-order expansion of the Taylor formula of the cost function f(x) is:

among them,
The weighted sum of the first t decision trees,

For the fastest falling iteration formula,

To assume the gradient of the function;

The AI wind control engine is obtained according to the linear combination tree and the cost function.
The credit evaluation method according to claim 1, wherein when the borrowing request is received, the borrowing request is obtained corresponding to the user's pre-lending data of the borrowing user, and the user's pre-lending data is separately input into the preset. The steps of the expert engine and the AI risk control engine for credit evaluation include:

Upon receiving the loan request, generating a corresponding data acquisition request according to the user identifier and the authorization license included in the loan request, and sending the data acquisition request to the data management system to obtain corresponding user pre-lending data;

Receiving user pre-lending data fed back by the data management system, and inputting the pre-lending data of the user into a preset expert engine and the AI wind control engine for credit evaluation.
The credit evaluation method according to claim 1, wherein said obtaining an expert evaluation result of said preset expert engine and an AI evaluation result of said AI wind control engine are based on said expert evaluation conclusion and AI evaluation Conclusion The steps to generate a corresponding decision suggestion include:

Obtaining an expert evaluation conclusion of the preset expert engine and an AI evaluation conclusion of the AI wind control engine, and performing a weighting operation based on the expert evaluation conclusion and the AI evaluation conclusion to obtain a credit line of the borrowing user;

A corresponding loan recommendation is generated according to the credit line, wherein the loan recommendation includes a loan amount and a loan period.
The credit evaluation method according to claim 1, wherein said obtaining an expert evaluation result of said preset expert engine and an AI evaluation result of said AI wind control engine are based on said expert evaluation conclusion and AI evaluation Conclusion After the steps to generate the corresponding decision suggestions, it also includes:

Obtaining user loan data of the borrowing user, and combining the user pre-lending data and the user post-loan data to obtain an iterative training sample;

The AI wind control engine is iteratively trained by the iterative training samples to adjust the AI wind control engine.
A credit evaluation apparatus, characterized in that the credit evaluation apparatus comprises:

An engine construction module, configured to acquire historical credit data, and construct an AI wind control engine according to the historical credit data;

a credit evaluation module, configured to acquire, before receiving the loan request, the pre-lending data of the borrowing user corresponding to the borrowing request, and input the pre-lending data of the user into a preset expert engine and the AI wind control engine for credit Evaluation

The suggestion generating module is configured to obtain an expert evaluation conclusion of the preset expert engine and an AI evaluation conclusion of the AI wind control engine, and generate corresponding decision suggestions based on the expert evaluation conclusion and the AI evaluation conclusion.
The credit evaluation apparatus according to claim 8, wherein said historical credit data includes historical pre-lending data and historical post-loan data.

The engine construction module is further configured to use the historical pre-lending data and the historical post-lost data as an input training set and an output training set, respectively, and analyzing and training the initial machine learning model based on the input training set and the output training set. Get the AI wind control engine.
The credit evaluation apparatus according to claim 9, wherein said initial machine learning model is cold-started by an expert rule.
The credit evaluation apparatus according to claim 9, wherein said input element in said input training set comprises a multidimensional vector x of dimension m, and said output element of said output training set comprises loan non-performing rate y,

The engine construction module is further configured to combine the input element and the output element to define m multi-dimensional training vector groups

{(x 1 , y 1 ), (x 2 , y 2 ), ..., (x m , y m )};

Learning the multidimensional training vector in the multi-dimensional training vector group based on the gradient lifting tree GBDT algorithm to obtain a corresponding decision tree

T(x; θ m ),

Where θ m is a parameter of the decision tree;

Linearly combining the decision trees to obtain a linear combination tree

The cost function of the linear combination number is

f(x)=l(h(x,D),Y),

Wherein the h(x, D) is a hypothesis function, which is a decision result of the decision tree, and the first-order expansion of the Taylor formula of the cost function f(x) is:

among them,
The weighted sum of the first t decision trees,

For the fastest falling iteration formula,

To assume the gradient of the function;

The AI wind control engine is obtained according to the linear combination tree and the cost function.
The credit evaluation apparatus according to claim 8, wherein said credit evaluation module comprises:

a data obtaining unit, configured to generate a corresponding data acquisition request according to the user identifier and the authorization permission included in the loan request, and send the data acquisition request to the data management system to obtain a corresponding User pre-lending data;

The credit evaluation unit is configured to receive user pre-lending data fed back by the data management system, and input the user pre-lending data into the preset expert engine and the AI wind control engine for credit evaluation.
The credit evaluation apparatus according to claim 8, wherein the suggestion generation module comprises:

a weighting operation unit, configured to obtain an expert evaluation conclusion of the preset expert engine and an AI evaluation result of the AI wind control engine, and perform a weighting operation based on the expert evaluation conclusion and the AI evaluation conclusion to obtain the borrowing user credits;

The suggestion generating unit is configured to generate a corresponding loan proposal according to the credit limit, wherein the loan recommendation includes a loan amount and a loan period.
The credit evaluation apparatus according to claim 8, wherein said credit evaluation apparatus further comprises:

a data combination module, configured to obtain data of the user's credits of the borrowing user, and combine the user pre-lending data and the user's post-loan data to obtain an iterative training sample;

An iterative training module is configured to iteratively train the AI wind control engine through the iterative training samples to adjust the AI wind control engine.
A credit evaluation device, characterized in that the credit evaluation device comprises a processor, a memory, and a credit evaluation program stored on the memory and executable by the processor, wherein the credit evaluation program is When the processor executes, the following steps are implemented:

Obtain historical credit data and construct an AI wind control engine based on the historical credit data;

Upon receiving the loan request, acquiring the borrowing request corresponding to the user's pre-lending data of the borrowing user, and inputting the user pre-lending data into the preset expert engine and the AI wind control engine for credit evaluation;

Obtaining an expert evaluation conclusion of the preset expert engine and an AI evaluation conclusion of the AI wind control engine, and generating corresponding decision suggestions based on the expert evaluation conclusion and the AI evaluation conclusion.
A credit evaluation apparatus according to claim 15, wherein said historical credit data includes historical pre-lending data and historical post-loan data.

The step of acquiring historical credit data and constructing an AI wind control engine according to the historical credit data includes:

The historical pre-lending data and the historical post-lost data are respectively used as an input training set and an output training set, and the initial machine learning model is analyzed and trained based on the input training set and the output training set to obtain an AI wind control engine.
The credit evaluation device of claim 16 wherein said initial machine learning model is cold initiated by expert rules.
The credit evaluation method according to claim 16, wherein the input element in the input training set comprises a multi-dimensional vector x of dimension m, and the output element of the output training set comprises a loan non-performing rate y,

The analyzing and training the initial machine learning model based on the input training set and the output training set, and the steps of obtaining the AI wind control engine include:

Combining the input element and the output element to define m multi-dimensional training vector groups

{(x 1 , y 1 ), (x 2 , y 2 ), ..., (x m , y m )};

Learning the multidimensional training vector in the multi-dimensional training vector group based on the gradient lifting tree GBDT algorithm to obtain a corresponding decision tree

T(x; θ m ),

Where θ m is a parameter of the decision tree;

Linearly combining the decision trees to obtain a linear combination tree

The cost function of the linear combination number is

f(x)=l(h(x,D),Y),

Wherein the h(x, D) is a hypothesis function, which is a decision result of the decision tree, and the first-order expansion of the Taylor formula of the cost function f(x) is:

among them,
The weighted sum of the first t decision trees,

For the fastest falling iteration formula,

To assume the gradient of the function;

The AI wind control engine is obtained according to the linear combination tree and the cost function.
The credit evaluation method according to claim 15, wherein when the borrowing request is received, the borrowing request acquires data of the user's pre-lending data corresponding to the borrowing user, and inputs the pre-lending data of the user into the preset The steps of the expert engine and the AI risk control engine for credit evaluation include:

Upon receiving the loan request, generating a corresponding data acquisition request according to the user identifier and the authorization license included in the loan request, and sending the data acquisition request to the data management system to obtain corresponding user pre-lending data;

Receiving user pre-lending data fed back by the data management system, and inputting the pre-lending data of the user into a preset expert engine and the AI wind control engine for credit evaluation.
A computer readable storage medium, wherein the computer readable storage medium stores a credit evaluation program, wherein when the credit evaluation program is executed by a processor, the following steps are implemented:

Obtain historical credit data and construct an AI wind control engine based on the historical credit data;

Upon receiving the loan request, acquiring the borrowing request corresponding to the user's pre-lending data of the borrowing user, and inputting the user pre-lending data into the preset expert engine and the AI wind control engine for credit evaluation;

Obtaining an expert evaluation conclusion of the preset expert engine and an AI evaluation conclusion of the AI wind control engine, and generating corresponding decision suggestions based on the expert evaluation conclusion and the AI evaluation conclusion.