CN108876076A - The personal credit methods of marking and device of data based on instruction - Google Patents

Abstract

The embodiment of the invention discloses the personal credit methods of marking and device of a kind of data based on instruction.Method includes：Sample account group is obtained, and selects positive sample and negative sample from sample account group according to preset rules；Branch mailbox processing is carried out to the first pre-set level group, and the corresponding WOE value of each branch mailbox is obtained according to the accounting of negative sample in each branch mailbox；The estimator of the parameter of prebuild Logic Regression Models is obtained according to the corresponding WOE value of each branch mailbox；Penalty term is configured to the estimator of the Logic Regression Models parameter, the penalty term is for configuration signal data and other contributions of non-signaling data to the Logic Regression Models；It is scored according to personal credit of the Logic Regression Models to user, obtains the personal credit scoring of user.The embodiment of the present invention is based on signaling data and improves to Logic Regression Models, occupies biggish specific gravity to guarantee signaling data in Logic Regression Models, compared with prior art, has the advantages that scoring is more accurate.

Description

Personal credit scoring method and device based on instruction data

technical field

The embodiments of the present invention relate to the technical field of credit risk management, and in particular to a personal credit scoring method and device based on instruction data.

Background technique

As of the end of September 2015, the credit information system of the central bank has included 870 million natural persons and 21.02 million enterprises and other organizations. Business and communication payment records, etc. In fact, mobile companies have detailed user behavior data, user background information, and can also grasp signaling data of user location information, call communication records, payment and consumption records, and these data have already been incorporated into the national credit system. As the field of credit investigation with the most practical value in the application of big data, mobile companies have been constantly exploring.

In the process of implementing the embodiment of the present invention, the inventor found that the personal credit scoring method established by the communication operator mainly considered the user's basic information, service order information, consumption ability, communication behavior, historical arrears and downtime records, contact circles, etc. However, because the factors considered are not focused, the actual scoring results are not accurate.

Contents of the invention

An object of the embodiments of the present invention is to solve the problem in the prior art that scoring results are inaccurate due to lack of emphasis on factors considered in scoring.

The embodiment of the present invention proposes a personal credit scoring method based on instruction data, including:

Obtain sample account groups, and select positive samples and negative samples from the sample account groups according to preset rules;

Perform binning processing on the first preset index group, and obtain the WOE value corresponding to each bin according to the proportion of negative samples in each bin;

Obtain the estimator of the parameters of the pre-built logistic regression model according to the WOE value corresponding to each bin;

Configuring a penalty item for the estimator of the logistic regression model parameter, the penalty item is used to configure the contribution of signaling data and other non-signaling data to the logistic regression model;

Score the user's personal credit according to the logistic regression model to obtain the user's personal credit score.

Optionally, selecting positive samples and negative samples from sample account groups according to preset rules includes:

Use the entropy method to determine the degree of dispersion of the second preset indicator group for each sample account;

Select positive samples and negative samples from the sample account groups according to the degree of dispersion of the second preset index group of each sample account.

Optionally, the configuration penalty item for the estimator of the logistic regression model parameter includes:

Analyzing the first preset index group to obtain a first index group related to signaling data and a second index group unrelated to signaling data;

Constructing a penalty term between the coefficients of the indicators in the second indicator group and the coefficients of the indicators in the first indicator group;

The penalty term is configured to the estimator of the logistic regression model parameter.

Optionally, the penalty item is

Wherein, ψ ₁ is the penalty coefficient, and β _j is the coefficient of the jth index in the second index group, is the coefficient of the k _nth index in the first index group.

Optionally, the scoring the user's personal credit according to the logistic regression model includes:

Under preset constraint conditions, converting the logistic regression model into a scoring model;

Using the data corresponding to the user's second preset index group as the input of the scoring model, and obtaining the scoring value corresponding to each sub-bin for each index in the second preset index group;

Obtain the user's personal credit score according to the score value of each indicator corresponding to each bin.

The embodiment of the present invention proposes a personal credit scoring device based on instruction data, including:

The obtaining module is used to obtain sample account groups, and select positive samples and negative samples from the sample account groups according to preset rules;

A binning module, configured to bin the first preset index group, and obtain the WOE value corresponding to each bin according to the proportion of negative samples in each bin;

A modeling module, configured to obtain an estimator of a parameter of a pre-built logistic regression model according to the WOE value corresponding to each bin;

A configuration module, configured to configure a penalty item for the estimator of the logistic regression model parameter, and the penalty item is used to configure the contribution of signaling data and other non-signaling data to the logistic regression model;

The scoring module is configured to score the user's personal credit according to the logistic regression model, and obtain the user's personal credit score.

Optionally, the acquisition module is configured to use an entropy method to judge the degree of dispersion of the second preset index group of each sample account; according to the degree of dispersion of the second preset index group of each sample account, the sample Select positive samples and negative samples from the account group.

Optionally, the configuration module is configured to analyze the first preset index group, obtain a first index group related to signaling data and a second index group unrelated to signaling data; construct the second index group The coefficient of the index in the second index group and the penalty item of the coefficient of the index in the first index group; the penalty item is configured as the estimator of the logistic regression model parameter.

Optionally, the penalty item is

Wherein, ψ ₁ is the penalty coefficient, and β _j is the coefficient of the jth index in the second index group, is the coefficient of the k _nth index in the first index group.

Optionally, the scoring module is configured to convert the logistic regression model into a scoring model under preset constraints; use the data corresponding to the user's second preset index group as the input of the scoring model to obtain the Each index in the second preset index group corresponds to the scoring value of each bin; and the user's personal credit score is obtained according to the scoring value of each index corresponding to each bin.

It can be seen from the above technical solution that a personal credit scoring method and device based on instruction data proposed in the embodiment of the present invention improves the logistic regression model based on signaling data, so as to ensure that signaling data occupies a larger proportion in the logistic regression model , compared with the prior art, has the advantage of more accurate scoring.

Description of drawings

The features and advantages of the present invention will be more clearly understood by referring to the accompanying drawings, which are schematic and should not be construed as limiting the invention in any way. In the accompanying drawings:

Fig. 1 shows a schematic flow chart of a personal credit scoring method based on instruction data provided by an embodiment of the present invention;

Fig. 2 shows a schematic flow chart of a personal credit scoring method based on instruction data provided by another embodiment of the present invention;

Fig. 3 shows a schematic structural diagram of a personal credit scoring device based on instruction data provided by an embodiment of the present invention.

Detailed ways

In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments It is a part of embodiments of the present invention, but not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts belong to the protection scope of the present invention.

Fig. 1 shows a schematic flow chart of a personal credit scoring method based on instruction data provided by an embodiment of the present invention. Referring to Fig. 1, the method can be implemented by a processor, and specifically includes the following distinguishing technical features:

110. Obtain sample account groups, and select positive samples and negative samples from the sample account groups according to preset rules;

It should be noted that the sample account here can be the user's mobile phone number, user number and other unique identification information of a user of an enterprise; then, based on some division rules, positive samples and negative samples are selected, that is, good user How many and bad users are there, or what is the percentage of the total sample.

Among them, there are many division rules, for example: the golden section method.

120. Perform binning processing on the first preset index group, and obtain the WOE value corresponding to each bin according to the proportion of negative samples in each bin;

It should be noted that the binning method is a relatively mature technology and will not be repeated here.

130. Obtain the parameter estimator of the pre-built logistic regression model according to the WOE value corresponding to each bin;

140. Configure a penalty item for the estimator of the logistic regression model parameter, and the penalty item is used to configure the contribution of signaling data and other non-signaling data to the logistic regression model;

It should be noted that in order to focus the scoring factors of the model on signaling data, the relationship between other non-signaling data and signaling data needs to be limited to ensure that the proportion of signaling data in the logistic regression model is greater than that of other non-signaling data. The proportion of non-signaling data.

150. Score the user's personal credit according to the logistic regression model, and acquire the user's personal credit score.

It should be noted that after the model is established, the relevant information of the user's first preset index group is used as the input of the model to obtain the user's personal credit score.

It can be seen that this embodiment improves the logistic regression model based on the signaling data to ensure that the signaling data occupies a larger proportion in the logistic regression model, and has the advantage of more accurate scoring compared with the prior art.

The above steps are described in detail below:

First, the method for selecting positive/negative samples in step 110 may include the following steps:

Use the entropy method to determine the degree of dispersion of the second preset indicator group for each sample account;

Select positive samples and negative samples from the sample account groups according to the degree of dispersion of the second preset index group of each sample account.

Secondly, step 140 specifically includes:

Analyzing the first preset index group to obtain a first index group related to signaling data and a second index group unrelated to signaling data;

Constructing a penalty term between the coefficients of the indicators in the second indicator group and the coefficients of the indicators in the first indicator group;

The penalty term is configured to the estimator of the logistic regression model parameter.

The form of the penalty term is

Wherein, ψ ₁ is the penalty coefficient, and β _j is the coefficient of the jth index in the second index group, is the coefficient of the k _nth index in the first index group.

Step 150 specifically includes:

Under preset constraint conditions, converting the logistic regression model into a scoring model;

Using the data corresponding to the user's second preset index group as the input of the scoring model, and obtaining the scoring value corresponding to each sub-bin for each index in the second preset index group;

Obtain the user's personal credit score according to the score value of each indicator corresponding to each bin.

Wherein, the constraint condition involved in the converting step may be a scoring range, for example: 1-100.

It can be seen that this embodiment utilizes an adaptive logistic regression model based on signaling data to score personal credit, and adaptively selects indicators and coefficients that are effective for credit scoring, ensuring that the personal credit scoring model remains stable and reflects signaling when screening indicators. The important role of data reduces the error of model coefficients and makes the scoring model more reasonable.

Fig. 2 shows a schematic flow chart of a personal credit scoring method based on instruction data provided by another embodiment of the present invention. Referring to Fig. 2, the design principle of the present invention will be described in detail below:

1. Design idea

This solution is mainly optimized for the traditional logistic regression personal credit scoring model. Here, an adaptive logistic regression model based on signaling data is used for personal credit scoring. The main process of the whole program is to first use the entropy method to extract positive and negative samples as standard sample data for subsequent scoring modeling, and select indicators related to measuring personal credit as input variables for modeling, including basic information, consumption ability, and credit records , network connections, indicators of behavioral preferences, and indicators of signaling data. Then preprocess the indicators of the standard sample data by binning and calculating their WOE values, and then establish an adaptive logistic regression model based on signaling data, use the signaling data to perform adaptive training on the model, and automatically select the ones that are effective for credit scoring Indicators and coefficients, and finally transform the regression model into a personal credit scoring table for personal credit scoring.

210. Extract standard sample data for scoring modeling;

To establish a personal credit evaluation system, some standard samples must be selected first, as a reference system, which can distinguish good users from bad users, and the subsequent scoring model will be analyzed based on these data.

This technical solution uses the entropy method combined with indicators of arrears to score users, and the scores are sorted from high to low. The higher the score, the higher the user's arrears, and the probability of default increases accordingly. Therefore, the top 1% users are taken as bad users, that is, positive samples; 10% of the total number of users are randomly selected from the remaining users as good users, that is, negative samples. The specific steps are as follows:

1. Select the total number of outages in the past three months, the total amount of arrears in the past three months, and the type of customer account period as indicators. These indicators balance the default situation of users in arrears. Due to the inconsistent value range of the indicators, in order to avoid too much emphasis on a single indicator, it is necessary to standardize the indicators. The standardization formula is as follows:

Among them, U _ij ,i=1,2,...,m,j=1,2,3 is the i-th record of the j-th indicator in the original data, m is the total number of users, and V _ij is the standardized data.

2. By calculating the entropy value, it can be used to judge the degree of dispersion of the three indicators of the total number of shutdowns in the past three months, the total amount of arrears in the past three months, and the type of customer account period. The greater the degree of dispersion, the greater the impact of this indicator on the comprehensive evaluation .

First, calculate the entropy value of the indicator, which measures the degree of dispersion of the indicator. The calculation formula is as follows:

Among them, r _ij represents the proportion of the index j in the i-th record

Then, calculate the weight of the indicators, and measure the coefficients that should be multiplied when calculating the total score of the three indicators:

Wherein, h _j is the difference coefficient of the jth index h _j =1-e _j , j=1,2,3.

Finally, calculate the entropy score of each user based on the weight of the indicator and the indicator value

3. Sort the S _i scores from high to low. The higher the score, the more serious the default in arrears. The users who get the top 1% of the score are regarded as bad users, that is, positive samples; 10% of the number of users are regarded as good users, that is, negative samples. The collection of positive and negative samples is the standard sample data, which is used to establish the credit scoring model in the future.

220. Select indicators related to measuring personal credit and perform preprocessing such as binning;

Select indicators that can comprehensively evaluate the user's credit status, and at the same time, in order to facilitate the subsequent scoring and form a scoring table to facilitate the evaluation of credit scores, it is necessary to bin the indicators to obtain the WOE value.

In order to comprehensively evaluate the user's credit situation, in addition to extracting the five major indicators of the user's basic information, consumption ability, credit record, personal relationship, and behavior preference from the perspective of traditional scoring, the user's signaling data is also added. The signaling data here is mainly considered location information. Considering the user's permanent location during the day and night, users who reside in high-end office buildings and CBDs during the day and in high-end residential areas at night have a relatively high-quality credit status.

The basic information of the user mainly includes information such as brand, online time, and identity; the consumption ability is a measure of the consumption level, consumption level, and consumption activity of the user in communication consumption, which mainly includes the account balance, the cost included in the main package, and the total number of calls last month. The number of times, the average recharge amount in the last three natural months, etc.; the credit record is used to measure the user's performance ability, including the total amount of arrears in the last three natural months, the number of single downtime days in the previous natural month, the number of double downtime days in the previous natural month, etc.; The network relationship is used to measure the strength of the user's social relationship. The network relationship is evaluated from the social influence and the credit score of the people around, including the number of high-frequency peer numbers, the average duration of high-frequency peer numbers, the number of close people, and the average number of close people. Consumption level, etc.; behavior preference is used to measure the activity of users using apps and application preferences, including APP type preference top1, the number of times of making friends in the community, the traffic of making friends in the community, the number of times of using e-commerce shopping, the number of times of using stock apps, etc. The signaling data of users mainly select the number of times that the permanent location is high-end office buildings and CBDs from 10:00 to 17:00 on weekdays and the number of times that the permanent location is high-end residential areas from 22:00 to 6:00 the next day.

In order to facilitate the subsequent scoring and form a scoring table to facilitate the evaluation of credit scores, it is necessary to bin the indicators. For continuous indicators, a reasonable binning should make the amount of data in each box more balanced, and should not be too much or too little , and the proportion of negative samples in each bin should show a monotonous upward or downward trend. The WOE value is used here, which can not only measure the trend of each bin, but also be the variable input of the subsequent regression model. The calculation formula is as follows:

For discrete indicators, when the value of the indicator is not many, it can be directly divided into bins according to its value and calculate the WOE value; when there are many values, some values can be combined, and then the corresponding value can be calculated WOE value.

230. Using the signaling data to perform adaptive training on the scoring model;

240. Automatically select indicators and coefficients that are effective for credit scoring

First, an adaptive logistic regression model based on signaling data is established for personal credit scoring.

Logistic regression is widely used in credit scoring models. Its structure is simple, and the role of coefficients is easy to explain in business.

The probability that a user is a bad user can be expressed by P, and the logistic regression model can be expressed as

Among them, x _i (i=1,2,...,s) is the index, since the value of P is between 0 and 1, after the logit transformation, the value range can be transformed into any real value, what needs to be solved is β=(β ₀ ,β ₁ ,...,β _s ) ^T .

When using logistic regression prediction, all indicators can be used to enter the model, but some indicators that do not contribute much to the prediction will also enter the model, resulting in a larger deviation in model prediction. The solution at this time is to do variable screening such as forward method, Back-off method, stepwise regression and other methods to eliminate indicators with insignificant effects.

However, when these traditional regression models do stepwise regression, variable selection and parameter estimation are separated into two stages, which leads to the instability of model selection. The logistic regression model based on signaling data adaptation makes use of signaling data to simultaneously perform variable selection and coefficient estimation adaptively, effectively reducing the bias of model coefficient estimation.

Here, the Adaptive-Lasso method is firstly used to solve the logistic regression model. Given data (X ⁽ⁱ⁾ ,y ⁽ⁱ⁾ ), i=1,2,...,n, where X ⁽ⁱ⁾ =(x _i1 ,...,x _is ), means that in the sample data The WOE value vector of the i-th data, a total of n, x _i1 represents the WOE value corresponding to the first index of the i-th data, y ⁽ⁱ⁾ represents the target variable, if the i-th data is a positive sample, then y ^{( i)} =1; if the i-th data is a negative sample, then y ⁽ⁱ⁾ =0. Then under the Adaptive-Lasso method, the estimator of β=(β ₀ ,β ₁ ,...,β _s ) ^T is defined as

The first part of the formula (2) represents the goodness of model fitting, which is the part of the general logistic regression model when solving, the second part Then it represents the penalty item of the coefficient, and λ _n is the penalty parameter. and in Indicates the estimated value of β _j obtained by the least squares estimation of formula (1). When the coefficient of |β _j | is large, Given a small penalty, a small deviation can be obtained; and when the |β _j | coefficient is small, Given a larger penalty, the coefficient is approximately 0, realizing the function of variable selection.

At the same time, the process of solving needs to use the indicators of signaling data to adaptively control the coefficients of other indicators to ensure that the indicators of signaling data contribute a higher weight, so it is necessary to add a penalty item on the basis of the Adaptive-Lasso method.

Record the number of times that the permanent location is high-end office buildings and CBDs from 10:00 to 17:00 on weekdays and the number of times that the permanent location is high-end residential areas from 22:00 to 6:00 the next day. The two indicators are in all indicators x _i (i=1 ,2,...,s) the subscripts are k ₁ ,k ₂ , namely Indicates the number of times the permanent location is high-end office buildings and CBDs from 10:00 to 17:00 on weekdays, Representation index corresponding coefficients.

In order to guarantee the signaling data aspect indicators and Contributing higher weights requires controlling for differences between _βj . Consider adding a penalty used to control indicators and The coefficient value of , restricted by The size of the guaranteed index and The coefficient of must be greater than the coefficients of other indicators, that is, to ensure that the indicators of signaling data contribute a higher weight to the model, and ψ ₁ is the penalty coefficient.

In summary, the adaptive logistic regression model based on signaling data β=(β ₀ ,β ₁ ,...,β _s ) The estimator of ^T is defined as

250. Transform the regression model into a scoring model

Converting regression coefficients into credit scores is a process of scale compilation. In order to facilitate the use of business personnel and the differences between scores have business significance, the following three requirements are usually required:

1. The score is controlled within a certain range, such as between 0-900 points.

2. For a specific score, there is a certain proportional relationship between good users and bad users, which is adopted here.

For measurement, if the score is 600, the ratio of good users to bad users is 50:1.

3. The increase in the rating value should reflect the change in the proportion of good users and bad users. If the rating value does not increase by 50 points, the odds will also double.

The current credit scoring formula commonly used in the industry is as follows:

score=offest+factor×ln(odds),

In order to satisfy the above three conditions, the equation needs to satisfy the following two equations

a. score=offest+factor×ln(odds)

b. score+pdo=offest+factor×ln(2×odds)

Among them, pdo indicates the value that the score value needs to be increased by doubling the odds. then there is

factor=pdo/ln(2), offset=score-factor×ln(odds).

Thus, the final scoring equation is obtained as:

score=offest+factor×ln(odds)

If the ratio of good users to bad users is 50:1 when the score is 600 points, and the odds doubles, the score increases by 50 points. Then there are:

factor＝50/ln(2)＝72.13，offest＝600-72.13×ln(50)＝317.83

Then the final scoring equation is obtained: score=317.83+72.13×ln(odds).

Since the left side of the logistic regression equation shows -logit(P)=ln(odds), the estimator of β will be obtained in step 4 Substituting into the scoring equation, we get:

Here x _i represents the binned WOE value corresponding to the i-th variable (indicator) value, is the regression model coefficient obtained from formula (3).

Therefore, according to the scoring formula, the scoring value corresponding to each bin of each variable can be obtained.

Among them, WOE represents the WOE value corresponding to the binning of the variable.

For the method implementation, for the sake of simple description, it is expressed as a series of action combinations, but those skilled in the art should know that the implementation of the present invention is not limited by the described action order, because according to the implementation of the present invention , certain steps may be performed in other order or simultaneously. Secondly, those skilled in the art should also know that the implementations described in the specification belong to preferred implementations, and the actions involved are not necessarily required by the implementations of the present invention.

Fig. 3 shows a schematic structural diagram of a personal credit scoring device based on instruction data provided by an embodiment of the present invention. Referring to Fig. 2, the device includes: an acquisition module 310, a binning module 320, a modeling module 330, a configuration module 340 and Scoring module, where:

An acquisition module 310, configured to acquire sample account groups, and select positive samples and negative samples from the sample account groups according to preset rules;

The binning module 320 is configured to bin the first preset index group, and obtain the WOE value corresponding to each bin according to the proportion of negative samples in each bin;

Modeling module 330, for obtaining the estimator of the parameter of pre-built logistic regression model according to the WOE value corresponding to each sub-bin;

The configuration module 340 is configured to configure a penalty item for the estimator of the logistic regression model parameter, and the penalty item is used to configure the contribution of signaling data and other non-signaling data to the logistic regression model;

Scoring module 350, configured to score the user's personal credit according to the logistic regression model, and obtain the user's personal credit score.

After receiving the instruction to start scoring, the acquisition module 310 acquires the account sample group from the pre-established database, then divides the account sample group, and sends the division result to the binning module 320; the binning module 320 combines the first preset The indicator group performs binning processing on the received positive/negative samples, then obtains the WOE data corresponding to each binning, and then sends it to the modeling module 330, and the modeling module 330 combines the received data with the pre-established Analyze the model, obtain the estimated quantity of unknown parameters in the model, and then send the model to the configuration module 340; the configuration module 340 configures the penalty item for the model to limit the contribution of the signaling index and the non-signaling index to the model, and will establish The completed model is sent to the scoring module 350; the scoring model 350 scores the user based on the completed model.

It can be seen that this embodiment utilizes an adaptive logistic regression model based on signaling data to score personal credit, and adaptively selects indicators and coefficients that are effective for credit scoring, ensuring that the personal credit scoring model remains stable and reflects signaling when screening indicators. The important role of data reduces the error of model coefficients and makes the scoring model more reasonable.

The functional modules of the device are described in detail below:

The acquisition module 310 is used to judge the degree of dispersion of the second preset indicator group of each sample account by using the entropy method; and select positive indicators from the sample account group according to the degree of dispersion of the second preset indicator group of each sample account. samples and negative samples.

The configuration module 340 is configured to analyze the first preset indicator group, obtain a first indicator group related to signaling data and a second indicator group unrelated to signaling data; construct indicators in the second indicator group and the penalty item of the coefficient of the index in the first index group; configure the penalty item to the estimator of the logistic regression model parameter.

The scoring module 350 is configured to transform the logistic regression model into a scoring model under preset constraints; use the data corresponding to the second preset indicator group of the user as the input of the scoring model to obtain the second preset indicator Each indicator in the group corresponds to the scoring value of each sub-box; the user's personal credit score is obtained according to the scoring value of each indicator corresponding to each sub-bin.

It can be seen that, compared with the existing logistic regression personal credit scoring technology, this technical solution uses an adaptive logistic regression model based on signaling data to improve the traditional logistic regression personal credit scoring model, ensuring that the personal credit scoring model is effective in screening indicators. The time remains stable and reflects the important role of signaling data, reducing the error of model coefficients and making the scoring model more reasonable.

In summary, compared with the traditional personal credit scoring method, the benefits brought by this program are compared as follows:

As for the device implementation, since it is basically similar to the method implementation, the description is relatively simple, and for related parts, please refer to the part of the description of the method implementation.

It should be noted that among the various components of the device of the present invention, the components are logically divided according to the functions to be realized, but the present invention is not limited thereto, and each component can be re-divided as required or a combination.

The various component implementations of the present invention can be implemented in hardware, or in software modules running on one or more processors, or in a combination thereof. In this device, the PC realizes the remote control of the equipment or device through the Internet, and precisely controls each operation step of the device or device. The present invention can also be implemented as an apparatus or an apparatus program (for example, a computer program and a computer program product) for performing a part or all of the methods described herein. In this way, the program for realizing the present invention can be stored on a computer-readable medium, and the files or documents generated by the program can be counted, and can generate data reports and cpk reports, etc., and can perform batch testing and statistics on power amplifiers. It should be noted that the above-mentioned embodiments illustrate rather than limit the invention, and that those skilled in the art will be able to design alternative embodiments without departing from the scope of the appended claims. In the claims, any reference signs placed between parentheses shall not be construed as limiting the claim. The word "comprising" does not exclude the presence of elements or steps not listed in a claim. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. The invention can be implemented by means of hardware comprising several distinct elements, and by means of a suitably programmed computer. In a unit claim enumerating several means, several of these means can be embodied by one and the same item of hardware. The use of the words first, second, and third, etc. does not indicate any order. These words can be interpreted as names.

Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: it can still be Modifications are made to the technical solutions described in the foregoing embodiments, or equivalent replacements are made to some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims (10)

1. a kind of personal credit methods of marking of data based on instruction, which is characterized in that including：

Sample account group is obtained, and selects positive sample and negative sample from sample account group according to preset rules；

Branch mailbox processing is carried out to the first pre-set level group, and each branch mailbox is obtained according to the accounting of negative sample in each branch mailbox and is corresponded to WOE value；

The estimator of the parameter of prebuild Logic Regression Models is obtained according to the corresponding WOE value of each branch mailbox；

Penalty term is configured to the estimator of the Logic Regression Models parameter, the penalty term is for configuration signal data and other Contribution of the non-signaling data to the Logic Regression Models；

It is scored according to personal credit of the Logic Regression Models to user, obtains the personal credit scoring of user.

2. the method according to claim 1, wherein described select from sample account group according to preset rules Positive sample and negative sample include：

The dispersion degree of the second pre-set level group of each sample account is judged using Information Entropy；

Chosen from the sample account group according to the dispersion degree of the second pre-set level group of each sample account positive sample and Negative sample.

3. the method according to claim 1, wherein the estimator to the Logic Regression Models parameter is matched Setting penalty term includes：

The first pre-set level group is analyzed, the first index group relevant to signaling data and and signaling data are obtained The second unrelated index group；

Construct the penalty term of the second index group middle finger target coefficient Yu the first index group middle finger target coefficient；

The penalty term is configured to the estimator of the Logic Regression Models parameter.

4. according to the method described in claim 3, it is characterized in that, the penalty term is

Wherein, ψ₁For penalty coefficient, β_jFor the coefficient of j-th of index in the second index group,For in the first index group Kth_nThe coefficient of a index.

5. method according to claim 1-4, which is characterized in that it is described according to the Logic Regression Models to The personal credit at family carries out scoring：

Under default constraint condition, Rating Model is converted by the Logic Regression Models；

Using the corresponding data of the second pre-set level group of user as the input of Rating Model, the second pre-set level group is obtained In each index correspond to the score value of each branch mailbox；

The personal credit scoring of user is obtained according to the score value that each index corresponds to each branch mailbox.

6. a kind of personal credit scoring apparatus of data based on instruction, which is characterized in that including：

Module is obtained, for obtaining sample account group, and positive sample is selected from sample account group according to preset rules and bears Sample；

Branch mailbox module for carrying out branch mailbox processing to the first pre-set level group, and is obtained according to the accounting of negative sample in each branch mailbox Take the corresponding WOE value of each branch mailbox；

Modeling module, the estimator of the parameter for obtaining prebuild Logic Regression Models according to the corresponding WOE value of each branch mailbox；

Configuration module configures penalty term for the estimator to the Logic Regression Models parameter, and the penalty term is for configuring Signaling data and other contributions of non-signaling data to the Logic Regression Models；

Grading module obtains the individual of user for scoring according to personal credit of the Logic Regression Models to user Credit scoring.

7. device according to claim 6, which is characterized in that the acquisition module, for each using Information Entropy judgement The dispersion degree of second pre-set level group of sample account；According to the dispersion degree of the second pre-set level group of each sample account Positive sample and negative sample are chosen from the sample account group.

8. device according to claim 6, which is characterized in that the configuration module, for first pre-set level Group is analyzed, and the first index group relevant to signaling data and the second index group unrelated with signaling data are obtained；Building The penalty term of the second index group middle finger target coefficient and the first index group middle finger target coefficient；The penalty term is matched It sets to the estimator of the Logic Regression Models parameter.

9. device according to claim 8, which is characterized in that the penalty term is

Wherein, Ψ₁For penalty coefficient, β_jFor the coefficient of j-th of index in the second index group,For the first index group Middle kth_nThe coefficient of a index.

10. according to the described in any item devices of claim 6-9, which is characterized in that institute's scoring module, in default constraint Under the conditions of, Rating Model is converted by the Logic Regression Models；Using the corresponding data of the second pre-set level group of user as The input of Rating Model obtains the score value that each index in the second pre-set level group corresponds to each branch mailbox；According to each finger The score value of the corresponding each branch mailbox of mark obtains the personal credit scoring of user.