CN110264342A - A kind of business audit method and device based on machine learning - Google Patents

Abstract

The embodiment of the present invention provides a business review method and device based on machine learning, which relates to the technical field of financial technology. The method includes: using historical business review data to train a random forest model, so that the random forest model learns the review behavior of the auditor, and then Input the feature vector of the business applicant into each decision tree in the random forest model, obtain the classification results output by each decision tree in the random forest model, and then according to the classification results output by each decision tree in the random forest model and the random forest The weight corresponding to each decision tree in the model determines the review result of the business request. Compared with the expert model, the random forest model is trained based on historical business audit data, not just the experience of professional auditors, so it is less dependent on human experience, reduces the influence of subjective factors, and improves the generality of the audit model. capability and versatility.

Description

A business review method and device based on machine learning

technical field

Embodiments of the present invention relate to the technical field of financial technology (Fintech), in particular to a method and device for business review based on machine learning.

Background technique

With the development of computer technology, more and more technical applications (such as: artificial intelligence, cloud computing, blockchain, etc.) in the financial field, the traditional financial industry is gradually transforming into financial technology (Fintech), but due to the Security and real-time requirements also put forward higher requirements for technology. In the financial industry, supply chain loans solve the financing difficulties and guarantee difficulties of upstream and downstream enterprises, and by breaking through the upstream and downstream financing bottlenecks, they can also reduce supply chain financing costs and improve the competitiveness of core enterprises and supporting enterprises. At present, the expert model is mainly used to review loans. After collecting the experience of business personnel who have been engaged in loan review for many years, they have been sorted out and summarized into a set of fixed business logic rules. The rule engine is deployed in the review system to realize automatic review. This method relies too much on the experience of business personnel and has strong subjective factors.

Contents of the invention

Due to the problems of relying too much on the experience of business personnel and strong subjective factors in the scheme of auditing business by using the expert model, the embodiments of the present invention provide a business audit method and device based on machine learning.

On the one hand, the embodiment of the present invention provides a business review method based on machine learning, including:

Obtain business requests from business applicants;

Extracting the feature vector of the service applicant according to the service request;

Input the feature vector of the business applicant into each decision tree in the random forest model, and obtain the classification result output by each decision tree in the random forest model, and the random forest model uses historical business audit data as a training sample acquired by training

According to the classification result output by each decision tree in the random forest model, the review result of the service request is determined.

Optionally, the determining the review result of the business request according to the classification result output by each decision tree in the random forest model includes:

The review result of the service request is determined according to the classification result output by each decision tree in the random forest model and the weight corresponding to each decision tree in the random forest model.

Optionally, determining the review result of the business request according to the classification result output by each decision tree in the random forest model and the weight corresponding to each decision tree in the random forest model includes:

Add the weights of the decision trees with the same classification results in the random forest model to determine the classification weight of each classification result;

The classification result with the largest classification weight is taken as the review result.

Optionally, the business request is a supply chain business request, and the historical business audit data includes chain enterprise characteristic data, core enterprise characteristic data, and historical audit records of auditors.

Optionally, the random forest model is obtained by training historical business audit data as training samples, including:

Obtain historical business audit data;

determining a set of feature vectors according to the historical business audit data;

Extracting N sub-feature vector sets from the feature vector set, where N is a preset positive integer;

Using the N sub-feature vector sets to train and obtain N decision trees;

Combine N decision trees into a random forest model.

On the one hand, an embodiment of the present invention provides a machine learning-based business review device, including:

An acquisition module, configured to acquire a business request from a business applicant;

An extraction module, configured to extract the feature vector of the service applicant according to the service request;

The classification module is used to input the feature vector of the business applicant into each decision tree in the random forest model, and obtain the classification result output by each decision tree in the random forest model, and the random forest model is based on historical business The audit data is obtained from training samples;

The processing module is configured to determine the review result of the business request according to the classification result output by each decision tree in the random forest model.

Optionally, the processing module is specifically configured to:

The review result of the service request is determined according to the classification result output by each decision tree in the random forest model and the weight corresponding to each decision tree in the random forest model.

Optionally, the processing module is specifically configured to:

Add the weights of the decision trees with the same classification results in the random forest model to determine the classification weight of each classification result;

The classification result with the largest classification weight is taken as the review result.

Optionally, the business request is a supply chain business request, and the historical business audit data includes chain enterprise characteristic data, core enterprise characteristic data, and historical audit records of auditors.

Optionally, the classification module is specifically used for:

Obtain historical business audit data;

determining a set of feature vectors according to the historical business audit data;

Extracting N sub-feature vector sets from the feature vector set, where N is a preset positive integer;

Using the N sub-feature vector sets to train and obtain N decision trees;

Combine N decision trees into a random forest model.

On the one hand, an embodiment of the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and operable on the processor. When the processor executes the program, a machine learning-based business Steps in the audit method.

On the one hand, an embodiment of the present invention provides a computer-readable storage medium, which stores a computer program executable by a computer device, and when the program is run on the computer device, the computer device executes a machine learning-based service Steps in the audit method.

In the embodiment of the present invention, since the random forest model is trained with historical business audit data, the random forest model learns the auditor's audit behavior, and then the feature vector of the business applicant is input into each decision tree in the random forest model to obtain a random forest The classification results output by each decision tree in the model, and then determine the review results of the business request based on the classification results output by each decision tree in the random forest model. Compared with the expert model, the random forest model is trained based on historical business audit data, not just the experience of professional auditors, so it is less dependent on human experience, reduces the influence of subjective factors, and improves the generality of the audit model. capability and versatility.

Description of drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings that need to be used in the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For Those of ordinary skill in the art can also obtain other drawings based on these drawings without any creative effort.

FIG. 1 is a schematic diagram of an application scenario provided by an embodiment of the present invention;

FIG. 2 is a schematic flowchart of a machine learning-based business review method provided by an embodiment of the present invention;

FIG. 3 is a schematic flowchart of a method for training a random forest model provided by an embodiment of the present invention;

FIG. 4 is a schematic flowchart of a machine learning-based business review method provided by an embodiment of the present invention;

FIG. 5 is a schematic structural diagram of a machine learning-based business review device provided by an embodiment of the present invention;

FIG. 6 is a schematic structural diagram of a computer device provided by an embodiment of the present invention.

Detailed ways

In order to make the object, technical solution and beneficial effects of the present invention more clear, the present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain the present invention, not to limit the present invention.

For the convenience of understanding, the terms involved in the embodiments of the present invention are explained below.

Supply Chain Finance: Supply Chain Finance is to take the core enterprises in the supply chain and their related upstream and downstream chain enterprises as a whole, and formulate an overall loan based on goods rights and cash flow control according to the transaction relationship and industry characteristics of the enterprises in the supply chain. A funding model for financial solutions.

Core enterprises: Enterprises in the supply chain that have mastered core technologies, core capabilities, and core links.

Chain affiliated enterprises: upstream and downstream chain affiliated enterprises of core enterprises in the supply chain.

The business review method based on machine learning in the embodiment of the present invention can be applied to the application scenario shown in Figure 1, which includes a terminal device 101 and an audit server 102, wherein the terminal device 101 can be a smart phone, a tablet computer or portable personal computer, etc. The verification server 102 may be a business verification server of a financial institution such as a bank. A user submits a service request on the terminal device 101 , and the terminal device 101 sends the service request to the audit server 102 . The audit server 102 includes a trained random forest model for business audit. The review server 102 extracts the feature vector of the business applicant according to the business request, inputs the feature vector of the business applicant into each decision tree in the random forest model, and obtains the classification result output by each decision tree in the random forest model. The classification result output by each decision tree in the model determines the review result of the business request. The review server 102 sends the review result to the terminal device 101 , and the user can view the review result of the service request from the terminal device 101 .

Based on the application scenario diagram shown in Figure 1, the embodiment of the present invention provides a process of a business review method based on machine learning, and the process of the method can be executed by a business review device based on machine learning, and the business review device based on machine learning Can be audit server 102 among Fig. 1, as shown in Fig. 2, comprise the following steps:

Step S201, acquiring the service request of the service applicant.

Step S202, extracting the feature vector of the business applicant according to the business request.

The business applicant can be an individual or an enterprise, and the business request can be a loan request, such as a personal loan request, an enterprise loan request, a supply chain loan request, and the like.

When the business applicant is an individual, the feature vector of the business applicant can be a personal feature, such as personal basic information, personal credit records, personal business records, personal assets, and the like. When the business applicant is an enterprise, the feature vector of the business applicant may be the characteristics of the enterprise, such as enterprise credit records, enterprise qualifications, and the like.

Step S203, input the feature vector of the business applicant into each decision tree in the random forest model, and obtain the classification result output by each decision tree in the random forest model.

The random forest model is obtained by training with historical business audit data as training samples. When the random forest model is used to audit individual business requests, the historical business audit data includes personal characteristic data and auditor's historical audit records. When the random forest model is used to audit supply chain business requests, historical business audit data includes chain enterprise characteristic data, core enterprise characteristic data, and historical audit records of auditors.

Step S204, according to the classification result output by each decision tree in the random forest model, determine the review result of the business request.

In a possible implementation, the classification results output by the decision tree include approval and failure, and when the classification result output in the random forest model is that the number of decision trees that pass the review is greater than the number of decision trees that fail the review, it is determined that The review result of the business request is approved. When the output classification result of the random forest model is that the number of decision trees that pass the review is less than the number of decision trees that fail the review, it is determined that the review result of the business request is not passed the review.

In a possible implementation manner, the review result of the service request is determined according to the classification result output by each decision tree in the random forest model and the weight corresponding to each decision tree in the random forest model. Specifically, the weight corresponding to each decision tree is determined according to the importance of feature vectors in the decision tree, and the sum of the weights corresponding to all decision trees is 1.

In a specific implementation, the weights of the decision trees with the same classification results in the random forest model can be added to determine the classification weight of each classification result, and the classification result with the largest classification weight can be used as the review result.

Exemplarily, the classification results of the decision tree are set to include approval and approval, and the weights of the decision trees whose classification results are approval are added to obtain the classification weight of approval. Add up the weights of the decision trees whose classification results fail the review to obtain the classification weight of the failed review. When the classification weight of the approved category is greater than the category weight of the failed category, the result of the review is determined to be approved;

Since the random forest model is trained with historical business audit data, the random forest model learns the audit behavior of the auditor, and then the feature vector of the business applicant is input into each decision tree in the random forest model to obtain each decision tree in the random forest model The output classification results, and then according to the classification results output by each decision tree in the random forest model, determine the review results of the business request. Compared with the expert model, the random forest model is trained based on historical business audit data, not just the experience of professional auditors, so it is less dependent on human experience, reduces the influence of subjective factors, and improves the generality of the audit model. capability and versatility.

The following describes the process of using historical business audit data to train and obtain a random forest model, as shown in Figure 3, including the following steps:

Step S301, acquiring historical business review data.

Historical business audit data includes chain enterprise characteristic data, core enterprise characteristic data, and historical audit records of auditors. Among them, chain enterprise characteristic data includes chain enterprise basic information (for example, registration time, enterprise size, number of employees, principal operating business industry, etc.), financial reports of chain enterprises (for example, debt ratio, debt amount, external guarantee amount, turnover, operating profit, etc.), pledged accounts receivable amount, pledged accounts receivable number, payable Accounts receivable feature data (such as accounts receivable period, etc.). The characteristic data of the core enterprise include the basic information of the core enterprise (for example, time of incorporation, enterprise size, number of employees, main business industry, whether it is a listed company, etc.), core enterprise operating financial reports, the number of recent major negative public opinion news, accounts receivable Characteristic data (such as accounts receivable period, etc.).

In step S302, a set of feature vectors is determined according to historical business review data.

Specifically, after the historical business audit data is acquired, the historical business audit data is preprocessed. Preprocessing can include the following methods: Method 1. Since the collected historical business audit data may have errors, outliers and missing situations, operations such as assigning default values and removing samples can be performed on such data to avoid the impact of such data training results. Method 2. Standardize the specified feature columns. For example, the company's operating income is a one-dimensional continuous distribution of digital data, and interval classification can be performed here, such as below 3 million, 3 million to 10 million, 10 million to 50 million, 5000 More than 4 interval categories of more than 10,000, and then mark them. Method 3: Organize historical business review data into a multi-dimensional data matrix for use in training.

Step S303, extracting N sub-feature vector sets from the feature vector set, where N is a preset positive integer.

Specifically, the bootstrap method is used to extract N sub-feature vector sets from the feature vector set.

Step S304, using N sub-feature vector sets to train to obtain N decision trees, and forming the N decision trees into a random forest model.

For each sub-feature vector set, the feature vectors in the sub-feature vector set are used to train to obtain a decision tree, wherein the decision tree may be a CART (Classification and Regression) tree. According to the importance of the eigenvectors in the sub-eigenvector set, the weight of the decision tree is set, and the sum of the weights of the N decision trees is 1.

Use historical business review data to train the random forest model, and set the weight of each decision tree in the random forest model according to the importance of the feature vector. Therefore, when using the random forest model to review business requests, combine the classification results of the decision tree and the decision tree. Weights can effectively provide the accuracy of audit results.

In order to better explain the embodiment of the present invention, the following uses supply chain loans as an implementation scenario to describe a machine learning-based business review method provided by the embodiment of the present invention, which is executed by a machine-learning-based business review device, as shown in Figure 4 As shown, the method includes the following steps:

It is assumed that bank A has more loan business in the field of supply chain finance, and the data warehouse of the business system has accumulated loan application records and loan review history data of supply chain business. Extract feature vector set Dt from business system data, intra-bank risk data and PBOC credit data. Use the bootstrap method to extract N sub-feature vector sets {D1, D2, ..., DN} from the feature vector set Dt. Use N sub-feature vector sets to train and obtain N decision trees, respectively {T1, T2, ..., TN}, according to the importance of the feature vectors in the sub-feature vector sets, set the weights of N decision trees as {a1, a2,...,a N}, the sum of the weights of N decision trees is 1, and N decision trees form a random forest model.

There is a property decoration business between the core enterprise C Real Estate Company in the supply chain and the chain enterprise B Decoration Company. The decoration company needs business capital turnover, and applies for a loan of 3.5 million yuan from Bank A through the accounts receivable of this business trade background. After receiving the loan request from Decoration Company B, Bank A inquires the credit records of Decoration Company B from the credit investigation platform of the People's Bank of China, including historical loan registration records, asset mortgage records, asset pledge records, enterprise external guarantee records, etc. Register the data source to obtain the enterprise characteristic data of C real estate company and B decoration company, such as enterprise scale, establishment time, enterprise registered capital, operating financial report, account receivable characteristic data, etc. Extract the eigenvectors of C Real Estate Company from the above data related to this loan, and input the eigenvectors into N decision trees, each decision tree outputs a classification result, where the classification results include approval and approval. . Add up the weights of the decision trees whose classification results are approved to obtain the approved classification weights. Add up the weights of the decision trees whose classification results fail the review to obtain the classification weight of the failed review. When the classification weight of the approved category is greater than the category weight of the failed category, the result of the review is determined to be approved;

Since the random forest model is trained with historical business audit data, the random forest model learns the audit behavior of the auditor, and then the feature vector of the business applicant is input into each decision tree in the random forest model to obtain each decision tree in the random forest model The output classification results, and then according to the classification results output by each decision tree in the random forest model, determine the review results of the business request. Compared with the expert model, the random forest model is trained based on historical business audit data, not just the experience of professional auditors, so it is less dependent on human experience, reduces the influence of subjective factors, and improves the generality of the audit model. capability and versatility.

Based on the same technical concept, the embodiment of the present invention provides a machine learning-based business review device, as shown in Figure 5, the device 500 includes:

An acquisition module 501, configured to acquire a business request from a business applicant;

An extraction module 502, configured to extract the feature vector of the business applicant according to the business request;

The classification module 503 is used to input the feature vector of the business applicant into each decision tree in the random forest model, and obtain the classification result output by each decision tree in the random forest model, and the random forest model is based on the history The business review data is obtained from training samples;

The processing module 504 is configured to determine the review result of the business request according to the classification result output by each decision tree in the random forest model.

Optionally, the processing module 504 is specifically configured to:

The review result of the service request is determined according to the classification result output by each decision tree in the random forest model and the weight corresponding to each decision tree in the random forest model.

Optionally, the processing module 504 is specifically configured to:

Add the weights of the decision trees with the same classification results in the random forest model to determine the classification weight of each classification result;

The classification result with the largest classification weight is taken as the review result.

Optionally, the business request is a supply chain business request, and the historical business audit data includes chain enterprise characteristic data, core enterprise characteristic data, and historical audit records of auditors.

Optionally, the classification module 503 is specifically configured to:

Obtain historical business audit data;

determining a set of feature vectors according to the historical business audit data;

Extracting N sub-feature vector sets from the feature vector set, where N is a preset positive integer;

Using the N sub-feature vector sets to train and obtain N decision trees;

Combine N decision trees into a random forest model.

Based on the same technical concept, an embodiment of the present invention provides a computer device, as shown in FIG. 6 , including at least one processor 601, and a memory 602 connected to at least one processor. The processor is not limited in the embodiment of the present invention. As for the specific connection medium between 601 and the memory 602, the bus connection between the processor 601 and the memory 602 in FIG. 6 is taken as an example. The bus can be divided into address bus, data bus, control bus and so on.

In the embodiment of the present invention, the memory 602 stores instructions that can be executed by at least one processor 601, and at least one processor 601 can execute the instructions included in the aforementioned machine learning-based business review method by executing the instructions stored in the memory 602. step.

Among them, the processor 601 is the control center of the computer equipment, which can use various interfaces and lines to connect various parts of the computer equipment, and conduct business operations by running or executing instructions stored in the memory 602 and calling data stored in the memory 602. review. Optionally, the processor 601 may include one or more processing units, and the processor 601 may integrate an application processor and a modem processor. The tuner processor mainly handles wireless communication. It can be understood that the foregoing modem processor may not be integrated into the processor 601 . In some embodiments, the processor 601 and the memory 602 can be implemented on the same chip, and in some embodiments, they can also be implemented on independent chips.

The processor 601 may be a general-purpose processor, such as a central processing unit (CPU), a digital signal processor, an application specific integrated circuit (Application Specific Integrated Circuit, ASIC), a field programmable gate array or other programmable logic devices, discrete gates or transistors Logic devices and discrete hardware components can implement or execute the methods, steps and logic block diagrams disclosed in the embodiments of the present invention. A general purpose processor may be a microprocessor or any conventional processor or the like. The steps of the methods disclosed in connection with the embodiments of the present invention may be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules in the processor.

The memory 602, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer-executable programs and modules. The memory 602 may include at least one type of storage medium, such as flash memory, hard disk, multimedia card, card-type memory, random access memory (Random Access Memory, RAM), static random access memory (Static Random Access Memory, SRAM), Programmable Read Only Memory (PROM), Read Only Memory (ROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Magnetic Memory, Disk, discs and more. Memory 602 is, but is not limited to, any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a computer. The memory 602 in the embodiment of the present invention may also be a circuit or any other device capable of implementing a storage function, and is used for storing program instructions and/or data.

Based on the same technical idea, an embodiment of the present invention provides a computer-readable storage medium, which stores a computer program executable by a computer device. When the program is run on the computer device, the computer device executes a machine-based Learn the steps of the business audit methodology.

Those skilled in the art should understand that the embodiments of the present invention may be provided as methods or computer program products. Accordingly, the present invention can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.

The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It should be understood that each procedure and/or block in the flowchart and/or block diagram, and a combination of procedures and/or blocks in the flowchart and/or block diagram can be realized by computer program instructions. These computer program instructions may be provided to a general purpose computer, special purpose computer, embedded processor, or processor of other programmable data processing equipment to produce a machine such that the instructions executed by the processor of the computer or other programmable data processing equipment produce a An apparatus for realizing the functions specified in one or more procedures of the flowchart and/or one or more blocks of the block diagram.

These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing apparatus to operate in a specific manner, such that the instructions stored in the computer-readable memory produce an article of manufacture comprising instruction means, the instructions The device realizes the function specified in one or more procedures of the flowchart and/or one or more blocks of the block diagram.

These computer program instructions can also be loaded onto a computer or other programmable data processing device, causing a series of operational steps to be performed on the computer or other programmable device to produce a computer-implemented process, thereby The instructions provide steps for implementing the functions specified in the flow chart or blocks of the flowchart and/or the block or blocks of the block diagrams.

While preferred embodiments of the invention have been described, additional changes and modifications to these embodiments can be made by those skilled in the art once the basic inventive concept is appreciated. Therefore, it is intended that the appended claims be construed to cover the preferred embodiment as well as all changes and modifications which fall within the scope of the invention.

Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and equivalent technologies thereof, the present invention also intends to include these modifications and variations.

Claims (10)

1. A business review method based on machine learning, characterized in that, comprising: Obtain business requests from business applicants; Extracting the feature vector of the service applicant according to the service request; Input the feature vector of the business applicant into each decision tree in the random forest model, and obtain the classification result output by each decision tree in the random forest model, and the random forest model uses historical business audit data as a training sample acquired by training According to the classification result output by each decision tree in the random forest model, the review result of the service request is determined. 2. The method according to claim 1, wherein the determination of the review result of the business request according to the classification result output by each decision tree in the random forest model comprises: The review result of the service request is determined according to the classification result output by each decision tree in the random forest model and the weight corresponding to each decision tree in the random forest model. 3. the method for claim 2, is characterized in that, described according to the classification result of each decision tree output in the described random forest model and the corresponding weight of each decision tree in the described random forest model, determine the described Review results of business requests, including: Add the weights of the decision trees with the same classification results in the random forest model to determine the classification weight of each classification result; The classification result with the largest classification weight is taken as the review result. 4. The method according to any one of claims 1 to 3, wherein the business request is a supply chain business request, and the historical business audit data includes chain enterprise characteristic data, core enterprise characteristic data, auditor's Historical audit records. 5. The method according to claim 4, wherein the random forest model is obtained by training historical business audit data as a training sample, comprising: Obtain historical business audit data; determining a set of feature vectors according to the historical business audit data; Extracting N sub-feature vector sets from the feature vector set, where N is a preset positive integer; Using the N sub-feature vector sets to train and obtain N decision trees; Combine N decision trees into a random forest model. 6. A business review device based on machine learning, characterized in that it comprises: An acquisition module, configured to acquire a business request from a business applicant; An extraction module, configured to extract the feature vector of the service applicant according to the service request; The classification module is used to input the feature vector of the business applicant into each decision tree in the random forest model, and obtain the classification result output by each decision tree in the random forest model, and the random forest model is based on historical business The audit data is obtained from training samples; The processing module is configured to determine the review result of the business request according to the classification result output by each decision tree in the random forest model. 7. The device according to claim 6, wherein the processing module is specifically used for: The review result of the service request is determined according to the classification result output by each decision tree in the random forest model and the weight corresponding to each decision tree in the random forest model. 8. The device according to claim 7, wherein the processing module is specifically used for: Add the weights of the decision trees with the same classification results in the random forest model to determine the classification weight of each classification result; The classification result with the largest classification weight is taken as the review result. 9. A computer device, comprising a memory, a processor, and a computer program stored in the memory and operable on the processor, characterized in that, when the processor executes the program, any one of claims 1 to 5 is realized The steps of the method are claimed. 10. A computer-readable storage medium, characterized in that it stores a computer program that can be executed by a computer device, and when the program is run on the computer device, the computer device executes any one of claims 1-5. steps of the method described above.