Method and system for constructing scoring model of small and micro enterprise operators

Abstract

The invention discloses a method for constructing a scoring model of small and micro enterprise operators, and the method comprises the steps: extracting small and micro credit applicants based on existing rich samples, carrying out the clustering analysis, and extracting the features of an operator group; through a logistic regression algorithm, an operator feature scoring model is firstly carried out on a small and micro credit applicant, a subsequent credit approval process is determined according to a rating result, and the accuracy of a credit rating model is assisted to be improved; according to the method, a comprehensive scoring model is constructed in a mode of combining logistic regression and rule identification, so that a financial institution can identify actual operators of small and micro enterprises more accurately, and the accuracy and scientificity in a credit approval process are improved. Through the small and micro enterprise operator identification scoring model, whether the applicant accords with the operator characteristics is analyzed and predicted, so that the whole credit approval accuracy is improved, and the bank is helped to reduce the risk.

Description

Technical Field

[0001] The present application relates to the field of financial credit risk technology, and in particular to a method, system, computer equipment and computer program product for constructing a scoring model for small and micro business operators. Background Art

[0002] As the proportion of small and micro enterprises in economic development gradually increases, how to effectively assess the credit risk of small and micro enterprises has become a major challenge facing financial institutions. At present, in the process of small and micro loan approval, the applicant must be a "legal representative" and the "legal representative" must hold a certain number of shares in the enterprise, so that the "legal representative" is assumed to be the "actual controller" or "actual operator". At the same time, in the credit rating model for small and micro enterprises, the relevant information of the legal representative is used as an important part of the credit approval, with a weight of more than 50%. If the "legal representative" is not the actual operator of the enterprise, it is equivalent to the rating object of the "credit rating model" for small and micro enterprise credit approval being wrong, and the accuracy of the credit rating will be disturbed and affected by the verification.

[0003] The operators of small and micro enterprises control the daily operations of small and micro enterprises and determine their survival. Existing banking and financial institutions often pay more attention to the "actual controller" in the process of enterprise credit approval. The industry mainly uses various complex situations such as equity penetration, control relationship, and actual control type to identify and identify the "actual controller". For example, the disclosed invention application CN114510693A-A method and device for constructing a tool for identifying the actual controller of an enterprise, by analyzing equity investment information as a training sample to build a model for identifying the actual controller of the enterprise, and then assisting in business decisions such as customer access, credit approval, risk monitoring and asset preservation. However, in the scenario of small and micro loans, it is entirely possible to set the "applicant" to be the "legal representative" and require the "legal representative" to hold a certain number of shares in the enterprise, so as to default the "legal representative" as the "actual controller" or "actual operator". Since the equity of small and micro enterprises is relatively simple, but the actual operator has a significant impact on the enterprise, the existing solution may have the problem of "approval object" error. Summary of the invention

[0004] The present application provides a method, apparatus, computer equipment and storage medium for constructing a scoring model for small and micro business operators, aiming to solve the problem that in the prior art, in the credit approval process, the legal representative is regarded as the actual operator, which may lead to deviation in risk assessment, thereby affecting the accuracy and reliability of credit decisions.

[0005] In a first aspect, a method for constructing a scoring model for small and micro business operators is provided, the method comprising:

[0006] Obtain target samples required for modeling, wherein the target samples are based on the existing information of microfinance loan applicants that have been manually approved, and the target samples include two types of samples: those that have been approved and those that have not been approved after manual approval;

[0007] Based on the target sample, the telecommunication behavior, bank flow, social security and provident fund data of small and micro enterprise credit applicants are collected to establish the basic data set of the model;

[0008] Filtering and obtaining a plurality of sample scoring indicators according to the basic data set, obtaining scoring rules for the plurality of sample scoring indicators, and calculating the scores of the plurality of sample scoring indicators according to the scoring rules;

[0009] Based on the multiple sample scoring indicators and the score of each scoring indicator, a scoring model is constructed by a logistic regression algorithm.

[0010] In the above scheme, the rule scoring indicators may optionally include: applicant's age, applicant's gender, network access time, number of active days in the past 12 months, average traffic usage in the past 12 months, social security payment years, provident fund payment base, and transaction amount between the applicant and the company in the past 12 months.

[0011] In the above solution, optionally, the scoring rules for obtaining the multiple sample scoring indicators include:

[0012] Obtain a customer characteristic value, where the customer characteristic value is any one of the multiple sample scoring indicators;

[0013] Determine a plurality of gradings corresponding to the customer characteristic values, and determine an evaluation index corresponding to each grading;

[0014] Based on the evaluation index, the operator characteristic score corresponding to each grade is obtained.

[0015] In the above solution, optionally, calculating the scores of the multiple sample scoring indicators includes:

[0016] Obtain the customer characteristic value t, and based on the indicator scoring rule, calculate the rule score including:

[0017] score t =score lower +(score upper -score lower ) / (median upper -median lower )*(t-median lower );

[0018] Among them, when the customer's characteristic value t is greater than the median in the current tier and less than the median in the next tier, score lower Indicates the operator characteristic score of this tier, score upper Indicates the operator characteristic score of the next tier, median lower Indicates the median in this bin, median upper Indicates the median in the next bin; when the customer's characteristic value t is less than the median in this bin and greater than the median in the previous bin, score lower Indicates the operator characteristic score of the previous tier, score upper Indicates the operator characteristic score of this tier, median lower Indicates the median of the previous bin, median upper Indicates the median in this bin.

[0019] In the above scheme, optionally, the step of constructing a scoring model by a logistic regression algorithm includes:

[0020] The logarithmic ratio of customer feature values ​​is calculated using a logistic regression algorithm;

[0021] Determine the values ​​of constants A and B in the scorecard model and calculate the final score for the customer based on these constants.

[0022] In the above scheme, optionally, the probability that the customer is not an operator is expressed as p, and the normal probability is 1-p; the logarithmic ratio:

[0023]

[0024] The probability p that the customer is not an operator can be expressed as:

[0025]

[0026] The score scale set by the scorecard is defined by expressing the score as a linear expression of the logarithm of the ratio, which is expressed as follows:

[0027] Score = AB * log (lr);

[0028] Among them, A and B are constants. The negative sign in the formula can make the probability of not being an operator lower and the sample score higher. Low score represents low probability and high score represents high probability. The logistic regression scorecard model calculates the ratio as follows:

[0029] log(lr)=β0+β1x1+β2x2…+β n x n ;

[0030] Among them, β i(i=1,2,…,n) are the parameters obtained after fitting the model;

[0031] The constructed logistic regression scorecard model is used to score the samples, and the score result of the logistic regression scorecard model is Score logistic .

[0032] In the above scheme, optionally, the final scoring result of the model is used in the small and micro enterprise credit approval process of banks or other financial institutions to assist in identifying whether the applicant is the actual operator of the enterprise and making corresponding credit decisions based on the identification result.

[0033] In a second aspect, a device for constructing a scoring model for small and micro business operators is provided, the device comprising:

[0034] Sample acquisition module: used to acquire target samples required for modeling. The target samples are based on the information of existing microfinance applicants that have been manually approved. The target samples include two types of samples: those that have been approved and those that have not been approved after manual approval.

[0035] Aggregation data set module: used to aggregate the telecommunication behavior, bank flow, social security and provident fund data of small and micro enterprise credit applicants based on the target sample to establish the basic data set of the model;

[0036] Calculation module: used to obtain multiple sample scoring indicators based on the basic data set, obtain scoring rules for the multiple sample scoring indicators, and calculate the scores of the multiple sample scoring indicators based on the scoring rules;

[0037] Model building module: used to build a scoring model through a logistic regression algorithm based on the multiple sample scoring indicators and the score of each scoring indicator.

[0038] In a third aspect, a computer device includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0039] Obtain target samples required for modeling, wherein the target samples are based on the existing information of microfinance loan applicants that have been manually approved, and the target samples include two types of samples: those that have been approved and those that have not been approved after manual approval;

[0040] Based on the target sample, the telecommunication behavior, bank flow, social security and provident fund data of small and micro enterprise credit applicants are collected to establish the basic data set of the model;

[0041] Filtering and obtaining a plurality of sample scoring indicators according to the basic data set, obtaining scoring rules for the plurality of sample scoring indicators, and calculating the scores of the plurality of sample scoring indicators according to the scoring rules;

[0042] Based on the multiple sample scoring indicators and the score of each scoring indicator, a scoring model is constructed by a logistic regression algorithm.

[0043] In a fourth aspect, a computer program product comprises a computer program / instruction, wherein when the computer program / instruction is executed by a processor, the following steps are implemented:

[0044] Obtain target samples required for modeling, wherein the target samples are based on the existing information of microfinance loan applicants that have been manually approved, and the target samples include two types of samples: those that have been approved and those that have not been approved after manual approval;

[0045] Based on the target sample, the telecommunication behavior, bank flow, social security and provident fund data of small and micro enterprise credit applicants are collected to establish the basic data set of the model;

[0046] Filtering and obtaining a plurality of sample scoring indicators according to the basic data set, obtaining scoring rules for the plurality of sample scoring indicators, and calculating the scores of the plurality of sample scoring indicators according to the scoring rules;

[0047] Based on the multiple sample scoring indicators and the score of each scoring indicator, a scoring model is constructed by a logistic regression algorithm.

[0048] Compared with the prior art, this application has at least the following beneficial effects:

[0049] This application is based on further analysis and research of existing technical problems. It is recognized that in the credit approval process, the existing technology regards the legal representative as the actual operator, which may lead to deviations in risk assessment, thereby affecting the accuracy and reliability of credit decisions. Based on the existing rich samples, this solution extracts small and micro loan applicants, performs cluster analysis, and extracts the characteristics of the operator group; through the logistic regression algorithm, the operator characteristic scoring model is first performed on the small and micro loan applicants, and the subsequent credit approval process is determined according to the rating results, which assists in providing the accuracy of the credit rating model; this application combines logistic regression and rule recognition to construct a comprehensive scoring model, which enables financial institutions to more accurately identify the actual operators of small and micro enterprises, and improves the accuracy and scientificity of the credit approval process. Through the small and micro enterprise operator identification scoring model, it analyzes and predicts whether the applicant meets the operator characteristics, thereby improving the accuracy of the entire credit approval and helping banks reduce risks. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 A flowchart of a method for constructing a scoring model for small and micro business operators provided in one embodiment of the present application;

[0051] Figure 2 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0052] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0053] In one embodiment, Figure 1 As shown, a method for constructing a scoring model for small and micro business operators is provided, comprising the following steps:

[0054] Obtain target samples required for modeling, wherein the target samples are based on the existing information of microfinance loan applicants that have been manually approved, and the target samples include two types of samples: those that have been approved and those that have not been approved after manual approval;

[0055] Based on the target sample, the telecommunication behavior, bank flow, social security and provident fund data of small and micro enterprise credit applicants are collected to establish the basic data set of the model;

[0056] Filtering and obtaining a plurality of sample scoring indicators according to the basic data set, obtaining scoring rules for the plurality of sample scoring indicators, and calculating the scores of the plurality of sample scoring indicators according to the scoring rules;

[0057] Based on the multiple sample scoring indicators and the score of each scoring indicator, a scoring model is constructed by a logistic regression algorithm.

[0058] In this embodiment, the rule scoring indicators include: applicant age, applicant gender, network access time, active days in the past 12 months, average traffic usage in the past 12 months, social security payment years, provident fund payment base, and transaction amount between the applicant and the company in the past 12 months.

[0059] In this embodiment, the scoring rules for obtaining the multiple sample scoring indicators include:

[0060] Obtain a customer characteristic value, where the customer characteristic value is any one of the multiple sample scoring indicators;

[0061] Determine a plurality of gradings corresponding to the customer characteristic values, and determine an evaluation index corresponding to each grading;

[0062] Based on the evaluation index, the operator characteristic score corresponding to each grade is obtained.

[0063] In this embodiment, calculating the scores of the multiple sample scoring indicators includes:

[0064] Obtain the customer characteristic value t, and based on the indicator scoring rule, calculate the rule score including:

[0065] score t =score lower +(score upper -score lower ) / (median upper -median lower )*(t-median lower );

[0066] Among them, when the customer's characteristic value t is greater than the median in the current tier and less than the median in the next tier, score lower Indicates the operator characteristic score of this tier, score upper Indicates the operator characteristic score of the next tier, median lower Indicates the median in this bin, median upper Indicates the median in the next bin; when the customer's characteristic value t is less than the median in this bin and greater than the median in the previous bin, score lower Indicates the operator characteristic score of the previous tier, score upper Indicates the operator characteristic score of this tier, median lower Indicates the median of the previous bin, median upper Indicates the median in this bin.

[0067] In this embodiment, the steps of constructing a scoring model using a logistic regression algorithm include:

[0068] The logarithmic ratio of customer feature values ​​is calculated using a logistic regression algorithm;

[0069] Determine the values ​​of constants A and B in the scorecard model and calculate the final score for the customer based on these constants.

[0070] In this embodiment, the probability that the customer is not an operator is expressed as p, and the normal probability is 1-p; the logarithmic ratio:

[0071]

[0072] The probability p that the customer is not an operator can be expressed as:

[0073]

[0074] The score scale set by the scorecard is defined by expressing the score as a linear expression of the logarithm of the ratio, which is expressed as follows:

[0075] Score = AB * log (lr);

[0076] Among them, A and B are constants. The negative sign in the formula can make the probability of not being an operator lower and the sample score higher. Low score represents low probability and high score represents high probability. The logistic regression scorecard model calculates the ratio as follows:

[0077] log(lr)=β0+β1x1+β2x2…+β n x n ;

[0078] Among them, β i (i=1,2,…,n) are the parameters obtained after fitting the model;

[0079] The constructed logistic regression scorecard model is used to score the samples, and the score result of the logistic regression scorecard model is Score logistic .

[0080] In this embodiment, the final scoring result of the model is used in the small and micro enterprise credit approval process of banks or other financial institutions to assist in identifying whether the applicant is the actual operator of the enterprise and making corresponding credit decisions based on the identification result.

[0081] Based on the existing rich samples, this embodiment extracts small and micro loan applicants, performs cluster analysis, and extracts the characteristics of the operator group; through the logistic regression algorithm, the operator characteristic scoring model is first performed on the small and micro loan applicants, and the subsequent credit approval process is determined according to the rating results, which assists in providing the accuracy of the credit rating model; this application constructs a comprehensive scoring model by combining logistic regression and rule recognition, so that financial institutions can more accurately identify the actual operators of small and micro enterprises, and improve the accuracy and scientificity of the credit approval process. Through the small and micro enterprise operator identification scoring model, it is analyzed and predicted whether the applicant meets the operator characteristics, thereby improving the accuracy of the entire credit approval and helping banks reduce risks.

[0082] This embodiment extracts the characteristics of small and micro business operators through a large number of samples, constructs a small and micro business operator identification model through a logistic regression model algorithm, calculates the probability that an "applicant" qualifies as a small and micro business operator, helps to relatively accurately assess the credit risk of credit approval, and improves the accuracy of small and micro business loan approval.

[0083] In one embodiment, a scoring model construction method based on logistic regression and rules is provided to solve the problem of inaccurate rating of existing small and medium-sized business operators by identifying the operators of small and micro-enterprise credit applicants.

[0084] Based on the above purpose, one or more embodiments of this specification provide a scoring model construction method based on logistic regression and rules, including:

[0085] 1. Through existing data resources, we screen out small and micro loan applicants to form the samples needed for modeling. As the existing samples, customers have been approved by the bank and most of them have been investigated offline, so the proportion of qualified operators is relatively high, which serves as the basis for building the model;

[0086] 2. Based on the samples, gather data on telecommunications behavior, bank flow, social security, provident fund, etc. to establish a data foundation for subsequent indicator calculation and model construction;

[0087] 3. Screen and obtain the rule scoring indicators;

[0088] Some specific indicators are as follows:

[0089] Table 1

[0090]

[0091]

[0092]

[0093] 4. Obtain the scoring rules of the rule scoring indicators;

[0094] The scoring calculation rules for the scoring indicators are divided into the following steps:

[0095] S1 obtains a customer characteristic value t, where the customer characteristic value t is any one of the indicators in Table 1: for example, taking the customer's age as an example.

[0096] In this embodiment, samples, namely customer characteristic values ​​t, are obtained based on the customer database. Each customer characteristic value t can be used to evaluate the distinctiveness of the customer's operator characteristics, thereby classifying the customers. Financial institutions can adjust the provision of different financial services to customers of different categories based on the customer's operator classification, so as to reduce the risks in providing services to each customer.

[0097] S2 determines a number of grades corresponding to the customer characteristic value t, and determines an evaluation index corresponding to each grade.

[0098] In this embodiment, determining the multiple tiers corresponding to the customer characteristic value t specifically includes:

[0099] The customer characteristic value t is classified by using a classification method combining equal frequency classification with expert experience method;

[0100] The classification results are obtained, including: the number of grades and the threshold value of the customer characteristic value t in each grade.

[0101] Methods for grading customer characteristic value t include: equal frequency grading, equal interval grading and chi-square grading. Taking customer age as customer characteristic value t as an example, using equal interval grading means making the threshold interval of grading the same. Assuming that the ages of 1,000 customers are graded equally, the youngest customer is 18 years old and the oldest is 78 years old, and they are divided into 5 grades, namely 18 to 30, 30 to 42, 42 to 54, 54 to 66 and 66 to 78 years old. Because there are fewer young and elderly customers, the distribution of the number of customers under such grading rules shows obvious characteristics of low on both sides and high in the middle, with large fluctuations. Using chi-square grading means sorting the customer ages, calculating the chi-square value according to the characteristic frequency and characteristic value of each pair of adjacent characteristics, finding the smallest pair of all chi-square values ​​for feature merging, and repeating the above process. The process is continued until the calculated chi-square value is not lower than the preset threshold or the number of groups has reached a certain condition. The chi-square classification process is very complicated, and the obtained classification interval is not consistent with the actual scene. This embodiment uses the equal frequency classification combined with the expert experience classification method to classify the customer characteristic value t. Assuming that the total number of customers is 1,000, they are divided into 5 classes. After arranging them in order of age, every 200 customers are divided into one class. Referring to Table 2, because the age distribution of customers is uneven, the classification method of equal frequency classification combined with the expert experience method is used for classification. The upper and lower limits of the classification thresholds of young customers and elderly customers, whose number of customers is relatively small, are quite different. However, the upper and lower limits of the classification thresholds of middle-aged customers are relatively small because the number of customers is relatively large. Such a classification result is closer to the actual scene.

[0102] Table 2

[0103] Lower threshold (≥) Upper threshold value (<) First gear 0 25 Second gear 25 29 Third gear 29 36 Fourth gear 36 50 Fifth gear 50

[0104] In this embodiment, the evaluation indicators include: the median of the customer characteristic value t in each tier, the weight of evidence value WOE i , Maximum weight of evidence WOE max and minimum weight of evidence WOE min Weight of Evidence (WOE) i It is expressed as:

[0105] WOE i =Ln(Bad i / Bad) / (Good i / Good);

[0106] Among them, Bad i represents the number of bad samples in the i-th bin, Bad represents the number of bad samples in all samples, Good i It represents the number of good samples in the i-th bin, and Good represents the number of good samples in all samples.

[0107] Refer to Table 3 to obtain the weight of evidence for each level.

[0108] Table 3

[0109] Lower threshold (≥) Upper threshold value (<) <![CDATA[WOE i ]]> First gear 0 25 0.5387 Second gear 25 29 0.2090 Third gear 29 36 -0.3424 Fourth gear 36 50 -0.0560 Fifth gear 50 0.2589

[0110] Refer to Table 4 to find the maximum weight of evidence value WOE max , and the minimum weight of evidence WOE min .

[0111] Table 4

[0112] Lower threshold (≥) Lower threshold (<) <![CDATA[WOE i ]]> <![CDATA[WOE max ]]> <![CDATA[WOE min ]]> First gear 0 25 0.5387 0.5387 -0.3424 Second gear 25 29 0.2090 0.5387 -0.3424 Third gear 29 36 -0.3424 0.5387 -0.3424 Fourth gear 36 50 -0.0560 0.5387 -0.3424 Fifth gear 50 0.2589 0.5387 -0.3424

[0113] Refer to Table 5 to find the median of the customer characteristic value t in each tier. i , assuming that the first age category of customers is 0 to 25 years old, the median age of these customers is 19, and the medians of other categories are calculated accordingly.

[0114] Table 5

[0115]

[0116]

[0117] S3 obtains the operator characteristic score corresponding to each tier based on the evaluation index.

[0118] In this embodiment, referring to Table 5, based on the median of the customer characteristic value t in each tier, the weight of evidence value WOE i , Maximum weight of evidence WOE max and minimum weight of evidence WOE min The operator characteristic score corresponding to each tier is obtained, and the operator characteristic score corresponding to each tier is expressed as

[0119] score i =min(max(100 / (WOE min -WOE max )*(WOE i -WOE max ),0),100).

[0120] Table 6

[0121] Lower threshold (≥) Lower threshold (<) <![CDATA[WOE max ]]> <![CDATA[WOE min ]]> <![CDATA[score i ]]> First gear 0 25 0.5387 -0.3424 0 Second gear 25 29 0.5387 -0.3424 37 Third gear 29 36 0.5387 -0.3424 100 Fourth gear 36 50 0.5387 -0.3424 67 Fifth gear 50 0.5387 -0.3424 32

[0122] The above is the scoring rule for obtaining the rule scoring indicator.

[0123] 5. Obtain the rule score of the rule scoring indicator according to the scoring rule;

[0124] In this example, a customer characteristic value t is obtained, and the customer characteristic value t is any one of the indicators in Table 1: Taking the customer age as an example, based on the above indicator scoring rules, the calculation rule score is as follows:

[0125] score t =score lower +(score upper -score lower ) / (median upper -median lower )*(t-median lower ).

[0126] Among them, when the customer's characteristic value t is greater than the median in the current tier and less than the median in the next tier, score lower Indicates the operator characteristic score of this tier, score upper Indicates the operator characteristic score of the next tier, median lower Indicates the median in this bin, median upper Indicates the median in the next bin; when the customer's characteristic value t is less than the median in this bin and greater than the median in the previous bin, score lower Indicates the operator characteristic score of the previous tier, score upper Indicates the operator characteristic score of this tier, median lower Indicates the median of the previous bin, median upper Indicates the median in this bin. Refer to Table 7:

[0127] Table 7

[0128] Customer age Operator Characteristics Rating Results 18 0 20 23.11 25 24.67 29 58 36 87.31 45 64.31 50 50.85 56 34.70 62 32

[0129] 6. Build a logistic regression scorecard model;

[0130] According to the basic principle of logistic regression, if the probability that a customer is not an operator is expressed as p, then the normal probability is 1-p. Therefore, we can get the probability, also called logarithmic ratio:

[0131]

[0132] Then the probability p that the customer is not an operator can be expressed as:

[0133]

[0134] The score scale set by the scorecard can be defined by expressing the score as a linear expression of the logarithm of the ratio, which can be expressed as follows:

[0135] Scor = AB * log (lr) re;

[0136] Among them, A and B are constants. The negative sign in the formula can make the sample score higher as the probability of not being an operator is lower. Usually, this setting is in line with the ideal direction of change, that is, low scores represent low probability and high scores represent high probability. The logistic regression scorecard model calculates the ratio as follows:

[0137] log(lr)=β0+β1x1+β2x2…+β n x n ;

[0138] Among them, β i (i=1,2,…,n) are the parameters obtained after fitting the model.

[0139] The values ​​of constants A and B can be obtained by substituting two known or assumed scores into the calculation. After parameters A and B are determined, the ratio and probability, as well as the corresponding scores, can be calculated to form the final logistic regression scorecard model.

[0140] The constructed logistic regression scorecard model is used to score the samples, and the score result of the logistic regression scorecard model is Score logistic .

[0141] It should be understood that the above method for constructing a logistic regression scorecard model is illustrative and is not intended to limit the implementation of the present invention. All methods for constructing a logistic regression scorecard model in the prior art can be used in this embodiment.

[0142] 7. Calculate customer scores for customer stratification:

[0143] Based on the methods described in steps 4, 5, and 6 above, customer scores are calculated and customer stratification is performed.

[0144] Based on the characteristic variables of each customer, the operator characteristic scoring result of each customer is obtained. If the comprehensive result of the operator characteristic scoring results obtained by the customer based on all customer characteristic values ​​t is higher than 85 points, the customer is classified as an A-level customer who meets the operator characteristics; if the comprehensive result of the operator characteristic scoring results obtained by the customer based on all customer characteristic values ​​t is lower than 65 points, the customer is classified as a low-probability operator, that is, a D-level customer. Among them, the comprehensive result of the operator characteristic scoring results obtained by the customer based on all customer characteristic values ​​t is obtained, and the comprehensive result of the operator characteristic scoring results obtained by each customer based on all customer characteristic values ​​t is obtained according to the operator characteristic scoring result obtained by each customer characteristic value t and the scoring weight corresponding to the operator characteristic scoring result obtained by each customer characteristic value t. For example, when the customer's characteristic value t is the customer's age, if the operator characteristic scoring result obtained by the current customer based on the customer age as the characteristic value t is 60 points, and the weight of the customer age in all characteristic values ​​t is 0.2, then considering multiple characteristic values ​​t, the comprehensive result of the customer in the characteristic value t of customer age is 12 points. The same method is used to calculate the comprehensive result of other different characteristic values ​​t. If the comprehensive result of all characteristic values ​​t is higher than 85 points, the customer is classified as a high-probability operator and is a Class A customer.

[0145] The customer classification also includes: if the customer obtains a comprehensive score of 75-85 points for the operator characteristics based on all customer characteristics t, the customer is classified as a Class B customer; if the customer obtains a comprehensive score of 65-75 points for the operator characteristics based on all customer characteristics t, the customer is classified as a Class C customer. Based on the customer's operator characteristics classification results, financial institutions will provide exclusive services for customers of different levels to reduce service risks.

[0146] Through the small and micro enterprise operator identification scoring model, it is possible to analyze and predict whether the applicant meets the operator characteristics, thereby improving the accuracy of the entire credit approval process and helping banks reduce risks.

[0147] The main purpose of this embodiment is to calculate the "operator" identification scoring model for the "applicant" using telecommunications, bank flow, social security, provident fund and other data when approving credit for small and micro enterprises. If the scoring model scores high, the subsequent approval of small and micro enterprise credit access strategy, credit rating, quota strategy, etc. will continue; if the scoring model scores average, the factors affecting the low score will be analyzed, and the subsequent approval will be carried out after supplementing the information according to different reasons; if the scoring model scores low, it is highly likely that the applicant is not a small and micro enterprise operator, and the credit account manager needs to conduct an offline investigation or approval rejection.

[0148] In one embodiment, a system for constructing a scoring model for small and micro business operators is provided, including the following program modules:

[0149] Sample acquisition module: used to acquire target samples required for modeling. The target samples are based on the information of existing microfinance applicants that have been manually approved. The target samples include two types of samples: those that have been approved and those that have not been approved after manual approval.

[0150] Aggregation data set module: used to aggregate the telecommunication behavior, bank flow, social security and provident fund data of small and micro enterprise credit applicants based on the target sample to establish the basic data set of the model;

[0151] Calculation module: used to obtain multiple sample scoring indicators based on the basic data set, obtain scoring rules for the multiple sample scoring indicators, and calculate the scores of the multiple sample scoring indicators based on the scoring rules;

[0152] Model building module: used to build a scoring model through a logistic regression algorithm based on the multiple sample scoring indicators and the score of each scoring indicator.

[0153] The specific implementation content of each module can be found in the above-mentioned definition of the construction method of the small and micro enterprise operator scoring model, which will not be repeated here.

[0154] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Figure 2 As shown. The computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities, and the communication interface is used to communicate with an external terminal in a wired or wireless manner. The wireless method can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The computer device implements the above-mentioned method for constructing a scoring model for small and micro business operators by loading and running a computer program. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad provided on the computer device housing, or an external keyboard, touchpad or mouse, etc.

[0155] Those skilled in the art will understand that Figure 2 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0156] In one embodiment, a computer program product is also provided, including a computer program / instruction, which, when executed by a processor, involves all or part of the process in the above-mentioned embodiment method.

[0157] The technical features of the above embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

Claims

1. A method for constructing a scoring model for small and micro business operators, characterized in that: The method comprises: Obtain target samples required for modeling, wherein the target samples are based on the existing information of microfinance loan applicants that have been manually approved, and the target samples include two types of samples: those that have been approved and those that have not been approved after manual approval; Based on the target sample, the telecommunication behavior, bank flow, social security and provident fund data of small and micro enterprise credit applicants are collected to establish the basic data set of the model; Filtering and obtaining a plurality of sample scoring indicators according to the basic data set, obtaining scoring rules for the plurality of sample scoring indicators, and calculating the scores of the plurality of sample scoring indicators according to the scoring rules; Based on the multiple sample scoring indicators and the score of each scoring indicator, a scoring model is constructed by a logistic regression algorithm.

2. The method according to claim 1, characterized in that The rule scoring indicators include: applicant's age, applicant's gender, network access time, number of active days in the past 12 months, average traffic usage in the past 12 months, social security payment years, provident fund payment base, and the transaction amount between the applicant and the company in the past 12 months.

3. The method according to claim 1, characterized in that: The scoring rules for obtaining the multiple sample scoring indicators include: Obtain a customer characteristic value, where the customer characteristic value is any one of the multiple sample scoring indicators; Determine a plurality of gradings corresponding to the customer characteristic values, and determine an evaluation index corresponding to each grading; Based on the evaluation index, the operator characteristic score corresponding to each grade is obtained.

4. The method according to claim 3, characterized in that Calculating the scores of the multiple sample scoring indicators includes: Obtain the customer characteristic value t, and based on the indicator scoring rule, calculate the rule score including: score t =score lower +(score upper -score lower ) / (median upper -median lower )*(t-median lower ); Among them, when the customer's characteristic value t is greater than the median in the current tier and less than the median in the next tier, score lower Indicates the operator characteristic score of this tier, score upper Indicates the operator characteristic score of the next tier, median lower Indicates the median in this bin, median upper Indicates the median in the next bin; when the customer's characteristic value t is less than the median in this bin and greater than the median in the previous bin, score lower Indicates the operator characteristic score of the previous tier, score upper Indicates the operator characteristic score of this tier, median lower Indicates the median of the previous bin, median upper Indicates the median in this bin.

5. The method according to claim 1, characterized in that The steps of constructing a scoring model using a logistic regression algorithm include: The logarithmic ratio of customer feature values ​​is calculated using a logistic regression algorithm; Determine the values ​​of constants A and B in the scorecard model and calculate the final score for the customer based on these constants.

6. The method according to claim 5, characterized in that The probability that the customer is not an operator is expressed as p, and the normal probability is 1-p; logarithmic ratio: The probability p that the customer is not an operator can be expressed as: The score scale set by the scorecard is defined by expressing the score as a linear expression of the logarithm of the ratio, which is expressed as follows: Score = AB * log (lr); Among them, A and B are constants. The negative sign in the formula can make the probability of not being an operator lower and the sample score higher. Low score represents low probability and high score represents high probability. The logistic regression scorecard model calculates the ratio as follows: log(lr)=β0+β1x1+β2x2…+β n x n ; Among them, β i (i=1,2,…,n) are the parameters obtained after fitting the model; The constructed logistic regression scorecard model is used to score the samples, and the score result of the logistic regression scorecard model is Score logistic .

7. The method according to claim 1, characterized in that The final scoring result of the model is used in the small and micro enterprise credit approval process of banks or other financial institutions to assist in identifying whether the applicant is the actual operator of the enterprise and to make corresponding credit decisions based on the identification result.

8. A system for constructing a scoring model for small and micro business operators, characterized in that: include: Sample acquisition module: used to acquire target samples required for modeling. The target samples are based on the existing information of microfinance applicants that have been manually approved. The target samples include two types of samples: those that have been approved and those that have not been approved after manual approval. Aggregation data set module: used to aggregate the telecommunication behavior, bank flow, social security and provident fund data of small and micro enterprise credit applicants based on the target sample to establish the basic data set of the model; Calculation module: used to obtain multiple sample scoring indicators based on the basic data set, obtain scoring rules for the multiple sample scoring indicators, and calculate the scores of the multiple sample scoring indicators based on the scoring rules; Model building module: used to build a scoring model through a logistic regression algorithm based on the multiple sample scoring indicators and the score of each scoring indicator.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the method according to claims 1-7 are implemented.

Images