CN115062074A - Loan collection method and device - Google Patents

Abstract

The application discloses a loan collection method and a loan collection device, wherein the method comprises the following steps: after receiving loan data of a client to be tested sent by a user, analyzing the loan data of the client to be tested to obtain the type, deposit and loan balance of the client to be tested; performing feature extraction on loan data of a customer to be tested to obtain target features; inputting the target characteristics into a scoring card model corresponding to the type of the customer to be tested to obtain a scoring result output by the scoring card model; inputting the score, deposit and loan balance of the client to be tested into a decision tree model corresponding to the score type to obtain an identification result output by the decision tree model; and displaying the collection level of the customer to be tested to the user through a preset interface. The method determines the grade of the customer to be tested based on the grade card model, determines the collection level of the customer to be tested based on the decision tree model, and can help the user to obtain the collection level of the customer to be tested so that the user can prepare collection in advance according to the collection level.

Description

Loan collection method and device

Technical Field

The application relates to the field of finance, in particular to a loan collection method and device.

Background

With the rapid development and integration of the internet and big data technologies, internet products have come and started the internet finance surge. The internet finance is developed in the years, the problems of overdue loan, bad loan and the like can occur along with the expiration of internet loans of banks, particularly for the payment of one-time payment for consumption due, due to the fact that the payment is not carried out during the period of the payment, the banks have to carry out collection management on overdue clients once the overdue banks are due.

At present, most of existing collection methods carry out risk classification on overdue customers by means of big data analysis technology, so that collection of overdue customers with higher risk is tracked. However, the existing collection method only focuses on collection after the loan is overdue, and lacks management for collection before the loan is due, so that a large amount of loans cannot be effectively recovered, and the collection pressure is increased.

Disclosure of Invention

The application provides a loan hastening method and a loan hastening device, which aim to assist a user to obtain the hastening level of the client so that the user can make preparation for hastening the receipt in advance according to the hastening level of the client (for example, the more time is spent on urging the client to repay a loan), thereby ensuring that the loan of the client can be effectively recovered.

In order to achieve the above object, the present application provides the following technical solutions:

a loan collection method, comprising:

after receiving loan data of a client to be tested sent by a user, analyzing the loan data of the client to be tested to obtain the type, deposit and loan balance of the client to be tested;

performing feature extraction on the loan data of the customer to be detected to obtain target features;

inputting the target characteristics into a grading card model corresponding to the type of the customer to be tested to obtain a grading result output by the grading card model; the scoring card model is obtained by training a preset logistic regression model by using a pre-obtained sample set; the scoring result comprises the score of the customer to be tested and the type of the score;

inputting the score, deposit and loan balance of the client to be detected into a decision tree model corresponding to the type of the score to obtain an identification result output by the decision tree model; the decision tree model is constructed by utilizing a pre-acquired training data set; the identification result comprises the collection level of the customer to be tested;

and displaying the collection level of the customer to be tested to the user through a preset interface.

Optionally, the scoring card model comprises a pre-collection scoring card model; the pre-collection scoring card model is obtained by training a preset logistic regression model by utilizing a first sample set acquired in advance;

the step of inputting the target characteristics into a scoring card model corresponding to the type of the customer to be tested to obtain a scoring result output by the scoring card model comprises the following steps:

under the condition that the type of the customer to be tested is a non-overdue customer, inputting the target characteristics into the pre-collection scoring card model to obtain a first scoring result output by the pre-collection scoring card model; the first scoring result comprises pre-urging receiving scoring of the client to be tested.

Optionally, the process of training a preset logistic regression model by using a first sample set obtained in advance to obtain the pre-collection scoring card model includes:

obtaining loan data of a plurality of sample customers from a loan service system in advance; the sample clients comprise clients handling loan service within a preset observation period;

performing feature extraction on the loan data of each sample client to obtain a feature variable set; the set of characteristic variables comprises a plurality of characteristic variables;

performing variable screening on each characteristic variable in the characteristic variable set to obtain each sample characteristic;

selecting a sample client transacting the loan service at a first preset observation point from the sample clients as a first sample client; the first preset observation point comprises a time period which takes a time point which is earlier than a preset loan expiration time by a first preset time as an initial point and takes the preset loan expiration time as an end point;

classifying the first sample client according to a first preset classification rule to obtain the type of the first sample client, and setting a type label corresponding to the type of the first sample client for the first sample client; the first preset classification rule is as follows: identifying a first sample client repaying in a first preset presentation period as a good sample, and identifying a first sample client not repaying in the first preset presentation period as a bad sample; the first preset presentation period comprises a time period which takes the ending time of the first preset observation point as the beginning and is delayed by a second preset time;

constructing the first sample set based on sample characteristics of the first sample client;

and training a preset logistic regression model by using the first sample set to obtain the pre-collection scoring card model.

Optionally, the decision tree model comprises a first decision tree model; the first decision tree model is obtained by utilizing a first training data set obtained in advance;

inputting the score, deposit and loan balance of the client to be detected into a decision tree model corresponding to the type of the score to obtain an identification result output by the decision tree model, wherein the identification result comprises:

inputting the pre-catalytic receiving score, the deposit and the loan balance of the client to be detected into the first decision tree model to obtain a first identification result output by the first decision tree model; the first identification result comprises the pre-urging grade of the customer to be detected.

Optionally, the process of training the obtained first decision tree model by using a first training data set obtained in advance includes:

inputting the sample characteristics of the first sample client into the pre-collection grading card model to obtain a first sample grading result output by the pre-collection grading card model; the first sample scoring result comprises a pre-collection score of the first sample client;

analyzing the loan data of the first sample client to obtain the deposit and loan balance of the first sample client;

constructing the first training data set by utilizing the pre-urging receipt score, deposit and loan balance of the first sample client;

training the first decision tree model using the first training data set.

Optionally, the scoring card model comprises an overdue hasty scoring card model; the overdue collection scoring card model is obtained by training a preset logistic regression model by using a second sample set acquired in advance;

the step of inputting the target characteristics into a scoring card model corresponding to the type of the client to be tested to obtain a scoring result output by the scoring card model comprises the following steps:

under the condition that the type of the client to be detected is an overdue client, inputting the target characteristics into the overdue collection scoring card model to obtain a second scoring result output by the overdue collection scoring card model; and the second grading result comprises overdue collection grading of the client to be tested.

Optionally, the process of training a preset logistic regression model by using a second sample set obtained in advance to obtain the overdue collection scoring card model includes:

obtaining loan data of a plurality of sample customers from a loan service system in advance; the sample clients comprise clients handling loan service within a preset observation period;

performing feature extraction on the loan data of each sample client to obtain a feature variable set; the set of characteristic variables comprises a plurality of characteristic variables;

performing variable screening on each characteristic variable in the characteristic variable set to obtain each sample characteristic;

selecting sample clients transacting the loan service at a second preset observation point from the sample clients as second sample clients; the second preset observation point comprises a preset loan overdue time period;

classifying the second sample client according to a second preset classification rule to obtain the type of the second sample client, and setting a type label corresponding to the type of the second sample client for the second sample client; the second preset classification rule is as follows: identifying a second sample client paid in a second preset presentation period as a good sample, and identifying a second sample client not paid in the second preset presentation period as a bad sample; the second preset presentation period comprises a time period which takes the ending time of the second preset observation point as the beginning and is delayed by second preset time;

constructing the second sample set based on sample characteristics of the second sample client;

and training a preset logistic regression model by using the second sample set to obtain the overdue collection scoring card model.

Optionally, the decision tree model comprises a second decision tree model; the second decision tree model is obtained by utilizing a second training data set obtained in advance;

inputting the score, deposit and loan balance of the client to be detected into a decision tree model corresponding to the type of the score to obtain an identification result output by the decision tree model, wherein the identification result comprises:

inputting the overdue credit rating and the loan balance of the customer to be detected into the second decision tree model to obtain a second recognition result output by the second decision tree model; and the second identification result comprises the overdue hastening grade of the client to be detected.

Optionally, the process of training the obtained second decision tree model by using a second training data set obtained in advance includes:

inputting the sample characteristics of the second sample client into the overdue collection scoring card model to obtain a second sample scoring result output by the overdue collection scoring card model; the second sample scoring results comprise overdue incentive scores of the second sample clients;

analyzing the loan data of the second sample client to obtain the loan balance of the second sample client;

constructing the second training data set by utilizing the overdue collection score and the loan balance of the second sample client;

training the second decision tree model using the second training data set.

A loan payment acceleration device comprising:

the system comprises an analysis unit, a processing unit and a processing unit, wherein the analysis unit is used for analyzing loan data of a client to be tested after receiving the loan data of the client to be tested sent by a user to obtain the type, deposit and loan balance of the client to be tested;

the extracting unit is used for extracting the features of the loan data of the customer to be detected to obtain target features;

the scoring unit is used for inputting the target characteristics into a scoring card model corresponding to the type of the client to be tested to obtain a scoring result output by the scoring card model; the scoring card model is obtained by training a preset logistic regression model by using a pre-obtained sample set; the scoring result comprises the scoring of the customer to be tested and the type of the scoring;

the identification unit is used for inputting the score, the deposit and the loan balance of the client to be detected into a decision tree model corresponding to the type of the score to obtain an identification result output by the decision tree model; the decision tree model is constructed by utilizing a pre-acquired training data set; the identification result comprises the collection level of the customer to be tested;

and the display unit is used for displaying the collection level of the customer to be tested to the user through a preset interface.

According to the technical scheme, after the loan data of the client to be tested sent by the user is received, the loan data of the client to be tested is analyzed, and the type, deposit and loan balance of the client to be tested are obtained. Performing feature extraction on loan data of a customer to be tested to obtain target features; and inputting the target characteristics into a scoring card model corresponding to the type of the customer to be tested to obtain a scoring result output by the scoring card model. And inputting the score, the deposit and the loan balance of the client to be detected into a decision tree model corresponding to the score type to obtain an identification result output by the decision tree model. And displaying the collection level of the customer to be tested to the user through a preset interface. The method comprises the steps of determining the grade of a client to be tested based on a grade card model, the type of the client to be tested and target characteristics, determining the collection level of the client to be tested based on a decision tree model and the grade, deposit and loan balance of the client to be tested, and helping a user to obtain the collection level of the client to be tested, so that the user can make collection preparation in advance according to the collection level of the client to be tested (for example, the higher the collection level is, the more time is spent to promote the client to be tested to pay back), and the loan of the client to be tested can be effectively recovered.

Drawings

In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.

Fig. 1a is a schematic flow chart illustrating a loan collection method according to an embodiment of the application;

fig. 1b is a schematic flow chart illustrating a loan collection method according to an embodiment of the present disclosure;

fig. 1c is a schematic flow chart illustrating a loan collection method according to an embodiment of the present disclosure;

FIG. 2a is a table diagram according to an embodiment of the present application;

FIG. 2b is a schematic diagram of another table provided in the embodiments of the present application;

FIG. 2c is a schematic diagram of another table provided in the embodiments of the present application;

FIG. 2d is a schematic diagram of another table provided in the embodiments of the present application;

FIG. 3 is a schematic flow chart illustrating another loan collection method according to an embodiment of the disclosure;

fig. 4 is a schematic view illustrating a configuration of a loan collection device according to an embodiment of the present disclosure.

Detailed Description

The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.

As shown in fig. 1a, fig. 1b and fig. 1c, a flow chart of a loan collection method provided in the embodiment of the present application includes the following steps:

s101: loan data of a plurality of sample customers is acquired in advance from a loan transaction system.

Wherein the sample clients include clients transacting loan transactions within a preset observation period. So-called loan transactions include, but are not limited to, paying for payment once for a due.

It should be noted that the loan data includes, but is not limited to: client attribute information, financial asset information, held product information, transaction information, loan payment support information (including loan expiration time, loan overdue time, loan transaction time, and the like), deposit information, loan balance information, credit investigation information, and the like.

S102: and performing feature extraction on the loan data of each sample client to obtain a feature variable set.

Wherein the feature variable set comprises a plurality of feature variables.

It should be noted that the specific implementation process of feature extraction is common knowledge familiar to those skilled in the art, and is not described herein again.

S103: and carrying out variable screening on each characteristic variable in the characteristic variable set to obtain each sample characteristic.

The specific implementation manner of the variable screening is common knowledge familiar to those skilled in the art, and specifically includes, but is not limited to: and performing variable screening on each characteristic variable in the characteristic variable set by using a characteristic box-separating algorithm to obtain the characteristics of the sample.

S104: and selecting a sample client handling the loan service at a first preset observation point from all sample clients as a first sample client, and selecting a sample client handling the loan service at a second preset observation point as a second sample client.

The first preset observation point includes a time point which is earlier than the preset loan expiration time by a first preset time (for example, 2 months) as an initial time and takes the preset loan expiration time as an ending time, and the second preset observation point includes a preset loan overdue time.

It should be noted that the first preset observation point and the second preset observation point are both within the time range shown in the preset observation period. And the preset loan overdue time period occurs after the preset loan due time.

S105: classifying the first sample client according to a first preset classification rule to obtain the type of the first sample client, and setting a type label corresponding to the type of the first sample client.

Wherein, the first preset classification rule is as follows: and identifying the first sample client which is paid in the first preset presentation period as a good sample, and identifying the first sample client which is not paid in the first preset presentation period as a bad sample.

It should be noted that the first predetermined presentation period includes a time period starting from the end time of the first predetermined observation point and delayed by a second predetermined time (for example, one month).

S106: and classifying the second sample client according to a second preset classification rule to obtain the type of the second sample client, and setting a type label corresponding to the type of the second sample client for the second sample client.

Wherein the second preset classification rule is as follows: and identifying the second sample client which pays in the second preset presentation period as a good sample, and identifying the second sample client which does not pay in the second preset presentation period as a bad sample.

It should be noted that the second predetermined presentation period includes a time period starting from the end time of the second predetermined observation point and delayed by a third predetermined time (for example, 30 days).

S107: a first sample set is constructed based on sample characteristics of a first sample client, and a second sample set is constructed based on sample characteristics of a second sample client.

Optionally, for any one of the first sample set and the second sample set, the sample set may be further divided to obtain a training sample set, a testing sample set, and a verification sample set. The training sample set is used for training the scoring card model (i.e., the later-mentioned pre-collection scoring card model and overdue collection scoring model), the testing sample set is used for testing the generalization error of the scoring card model, and the verifying sample set is used for verifying the performance of the scoring card model (e.g., verifying whether the scoring card model is over-fitted, etc.).

Specifically, assuming that the sample clients included in the sample set are sample clients handling loan services in 1 month to 7 months in 2020, the sample set is divided, the obtained training sample set may include sample clients handling loan services in 1 month to 5 months in 2020, the test sample set may include sample clients handling loan services in 1 month to 5 months in 2020, and the verification sample set may include sample clients handling loan services in 6 months to 7 months in 2020.

It should be noted that the above specific implementation process is only for illustration. Optionally, 70% of the samples (i.e., the sample characteristics of the sample client) may be selected from the sample set as a training sample set, 20% of the samples may be selected from the sample set as a testing sample set, and 10% of the samples may be selected from the sample set as a verification sample set.

S108: and training a preset logistic regression model by using the first sample set to obtain a pre-collection scoring card model.

The KS value of the pre-collection score card model obtained through training of the first sample set is generally 0.63 (the KS value represents the distinguishing capability of the pre-collection score card model, and if the KS value is within the interval of [0.5, 0.7], the distinguishing capability of the pre-collection score card model is high), and the PSI value is generally 0.026(PSI represents the stability of the pre-collection score card model, and if the PSI value is within the interval of [0, 0.1], the stability of the pre-collection score card model is high).

Generally speaking, a training sample set, a testing sample set and a verification sample set obtained by dividing a sample set are utilized to train a logistic regression model, and a concrete implementation principle of a scoring card model is obtained, which belongs to the common knowledge familiar to those skilled in the art, and specifically, the theoretical logic of the scoring card implementation is to map Odds (ratio of bad sample probability P to bad sample probability 1-P) predicted by the logistic regression model to a score of a sample client (the score is used for indicating the risk level of the sample client, and the score is in inverse proportion to the risk level, that is, the lower the score is, the higher the risk level is, the lower the possibility of repayment of the sample client is.

S109: and training the preset logistic regression model by using the second sample set to obtain an overdue collection scoring card model.

The overdue collection scoring card model trained by the second sample set also needs to have strong distinguishing capability and good stability.

S110: and inputting the sample characteristics of the first sample client into the pre-collection grading card model to obtain a first sample grading result output by the pre-collection grading card model.

Wherein the first sample scoring result comprises a pre-catalyst receiving score of the first sample client. Specifically, the pre-collection of the first sample scoring result output by the scoring card model may be as shown in fig. 2a, where in fig. 2a, the development sample represents a first sample set, and the cross-time verification sample represents a verification sample set divided from the first sample set.

S111: and inputting the sample characteristics of the second sample client into the overdue collection grading card model to obtain a second sample grading result output by the overdue collection grading card model.

And the second sample scoring result comprises overdue incentive scores of the second sample clients. Specifically, the second sample scoring result output by the overdue hastening scoring card model can be seen from fig. 2b, in which in fig. 2b, the development sample represents a second sample set, and the cross-time verification sample represents a verification sample set divided from the second sample set.

S112: and analyzing the loan data of the first sample client to obtain the deposit and loan balance of the first sample client, and analyzing the loan data of the second sample client to obtain the loan balance of the second sample client.

S113: and constructing a first training data set by utilizing the pre-collection score, the deposit and the loan balance of the first sample client, and constructing a second training data set by utilizing the overdue collection score and the loan balance of the second sample client.

S114: a first decision tree model is trained using a first training data set.

Specifically, pre-catalytic receiving scores, deposit and loan balances of a first sample client are respectively identified as a class of attributes, for each class of attributes, an attribute duty ratio (a ratio of a bad sample probability P to a bad sample probability 1-P in the first training data set) corresponding to the attributes is utilized to calculate an information entropy of the attributes, then the information entropy and a preset conditional entropy are utilized to calculate an information gain of the attributes, training data (namely the attributes of the first sample client) in the first training data set are sorted according to the sequence of the information gains from large to small to generate a decision tree corresponding to the attributes, and the first decision tree model is constructed on the basis of the decision tree corresponding to each class of attributes.

It should be noted that the decision tree corresponding to the pre-catalyst receipt score of the first sample client may be used to identify the risk level of the first sample client, the decision tree corresponding to the deposit of the first sample client may be used to identify the property level of the first sample client, and the decision tree corresponding to the loan balance of the first sample client may be used to identify the loan balance level of the first sample client.

In particular, the types of risk classes include, but are not limited to, high risk, medium risk, and low risk, the types of property classes include, but are not limited to, high property, medium property, and low property, and the types of loan balance classes include, but are not limited to, large amount and small amount.

Optionally, if the pre-hastening receiving score, the deposit, and the loan balance of the first sample client are input into the first decision tree model, a first sample identification result output by the first decision tree model is obtained, where the first sample identification result includes the pre-hastening receiving level of the first sample client. Specifically, as shown in fig. 2c, the first sample recognition result output by the first decision tree model may be represented by a development sample representing the first training data set, and the cross-time verification sample representing the verification data set divided from the first training data set in fig. 2 c.

In the embodiment of the present application, the types of the pre-inducement level include, but are not limited to: high risk low asset low amount queue, high risk low asset large amount queue, high risk medium asset small amount queue, high risk medium asset large amount queue, high risk high asset queue, medium risk low asset small amount queue, medium risk low asset large amount queue, medium risk high asset small amount queue, medium risk high asset large amount queue, low risk small amount queue, low risk large amount queue, and extremely low risk.

S115: a second decision tree model is trained using a second training data set.

Specifically, the pre-collection score and the loan balance of the second sample client are respectively identified as a class of attributes, for each class of attributes, an attribute duty ratio (a ratio of a bad sample probability P to a bad sample probability 1-P in the second training data set) corresponding to the attributes is used to calculate an information entropy of the attributes, the information entropy and a preset conditional entropy are used to calculate an information gain of the attributes, the training data (namely, the attributes of the second sample client) in the second training data set are sorted in a descending order of the information gains, a decision tree corresponding to the attributes is generated, and the second decision tree model is constructed based on the decision tree corresponding to each class of attributes.

It should be noted that the decision tree corresponding to the pre-catalyst receipt score of the second sample client may be used to identify the risk level of the second sample client, and the decision tree corresponding to the loan balance of the second sample client may be used to identify the loan balance level of the second sample client.

Optionally, if the overdue collection score and the loan balance of the second sample client are input into the second decision tree model, a second sample identification result output by the second decision tree model is obtained, where the second sample identification result includes the overdue collection level of the second sample client. Specifically, as shown in fig. 2d, the second sample recognition result output by the second decision tree model may be represented by a development sample representing the second training data set, and the cross-time verification sample representing the verification data set divided from the second training data set in fig. 2 d.

In the present embodiment, the types of overdue hastening ratings include, but are not limited to: a high risk high rate queue, a high risk low rate queue, a medium risk high rate queue, a medium risk low rate queue, a low risk high rate queue, a low risk low rate queue.

S116: after receiving the loan data of the client to be tested sent by the user, analyzing the loan data of the client to be tested to obtain the type, deposit and loan balance of the client to be tested.

S117: and performing feature extraction on the loan data of the customer to be detected to obtain target features.

S118: and under the condition that the type of the client to be detected is a non-overdue client, inputting the target characteristics into the pre-collection grading card model to obtain a first grading result output by the pre-collection grading card model.

After execution of S118, execution of S119 is continued.

The first scoring result comprises pre-urging receiving scoring of the client to be tested.

S119: and inputting the pre-urging receiving score, the deposit and the loan balance of the client to be detected into the first decision tree model to obtain a first identification result output by the first decision tree model.

After execution of S119, execution continues with S120.

The first identification result comprises the pre-urging grade of the customer to be detected.

S120: and displaying the pre-collection level of the customer to be tested to the user through a preset interface.

S121: and under the condition that the type of the client to be detected is an overdue client, inputting the target characteristics into the overdue hastening scoring card model to obtain a second scoring result output by the overdue hastening scoring card model.

After execution of S121, execution continues with S122.

And the second grading result comprises overdue gathering grading of the client to be tested.

S122: and inputting the overdue credit rating and the loan balance of the customer to be detected into the second decision tree model to obtain a second recognition result output by the second decision tree model.

After execution of S122, execution continues with S123.

And the second identification result comprises the overdue hastening grade of the client to be detected.

S123: and displaying the overdue hastening grade of the client to be tested to the user through a preset interface.

Based on the above-mentioned flow shown in S101-S123, the present embodiment can achieve the following beneficial effects:

1. extending the collection-hastening products from the loan products of the existing period and the due period to the loan products of the due one-time payment and consumption to the loan products of the due one-time payment;

2. from the time of the overdue stage of the client, expanding the time to the time of two stages of before expiration and after expiration, establishing a rating card model for the two stages respectively, and carrying out risk classification on the client to realize the forward movement of the time-lapse gateway and the full coverage of the time-lapse process, thereby better reducing the time-lapse pressure;

3. on the basis of the collection prompting grading card model, a decision tree model is further established according to the deposit and loan conditions of the customers, the customers are divided to form a collection prompting queue, the fine demands of collection prompting can be better met, and the optimization of the collection prompting resource allocation is realized.

In summary, the score of the client to be tested is determined based on the scoring card model, the type and the target characteristics of the client to be tested, the collection level of the client to be tested is determined based on the decision tree model and the score, deposit and loan balance of the client to be tested, and the user can be helped to learn the collection level of the client to be tested, so that the user can make collection preparation in advance according to the collection level of the client to be tested (for example, the higher the collection level is, the more time is spent to promote the client to pay back), and the loan of the client to be tested can be effectively recovered.

It should be noted that, in the above embodiment, the aforementioned S101 is an alternative implementation of the loan collection method described in this application. In addition, the above embodiment mentioned S120 is also an alternative implementation of the loan collection method in the present application. For this reason, the flow mentioned in the above embodiment can be summarized as the method shown in fig. 3.

As shown in fig. 3, a schematic flow chart of another loan collection method provided in the embodiment of the present application includes the following steps:

s301: after receiving the loan data of the client to be tested sent by the user, analyzing the loan data of the client to be tested to obtain the type, deposit and loan balance of the client to be tested.

S302: and performing feature extraction on the loan data of the customer to be detected to obtain target features.

S303: and inputting the target characteristics into a scoring card model corresponding to the type of the client to be tested to obtain a scoring result output by the scoring card model.

The scoring card model is obtained by training a preset logistic regression model by using a pre-obtained sample set; the scoring result comprises the scoring of the customer to be tested and the type of the scoring.

S304: and inputting the score, the deposit and the loan balance of the client to be detected into a decision tree model corresponding to the score type to obtain an identification result output by the decision tree model.

The decision tree model is constructed by utilizing a pre-acquired training data set; the identification result comprises the urging grade of the customer to be detected.

S305: and displaying the collection level of the customer to be tested to the user through a preset interface.

In summary, the score of the client to be tested is determined based on the scoring card model, the type and the target characteristics of the client to be tested, the collection level of the client to be tested is determined based on the decision tree model and the score, deposit and loan balance of the client to be tested, and the user can be helped to learn the collection level of the client to be tested, so that the user can make collection preparation in advance according to the collection level of the client to be tested (for example, the higher the collection level is, the more time is spent to promote the client to pay back), and the loan of the client to be tested can be effectively recovered.

Corresponding to the loan payment prompting method provided by the embodiment of the application, the embodiment of the application also provides a loan payment prompting device.

As shown in fig. 4, a schematic view of a loan collection apparatus provided in this embodiment of the present application includes:

the analysis unit 100 is configured to, after receiving the loan data of the customer to be tested sent by the user, analyze the loan data of the customer to be tested to obtain the type, deposit and loan balance of the customer to be tested.

And the extraction unit 200 is used for performing feature extraction on the loan data of the customer to be tested to obtain target features.

The scoring unit 300 is used for inputting the target characteristics into a scoring card model corresponding to the type of the customer to be tested to obtain a scoring result output by the scoring card model; the scoring card model is obtained by training a preset logistic regression model by using a pre-obtained sample set; the scoring result comprises the scoring of the customer to be tested and the type of the scoring.

Optionally, the scoring card model includes a pre-collection scoring card model; the pre-collection scoring card model is obtained by training a preset logistic regression model by using a first sample set acquired in advance.

The scoring unit 300 is specifically configured to: under the condition that the type of the client to be detected is a non-overdue client, inputting the target characteristics into a pre-collection grading card model to obtain a first grading result output by the pre-collection grading card model; the first scoring result comprises pre-urging receiving scores of the clients to be tested.

The scoring unit 300 is specifically configured to: obtaining loan data of a plurality of sample customers from a loan service system in advance; the sample clients comprise clients handling loan service within a preset observation period; performing feature extraction on loan data of each sample client to obtain a feature variable set; the characteristic variable set comprises a plurality of characteristic variables; performing variable screening on each characteristic variable in the characteristic variable set to obtain each sample characteristic; selecting a sample client transacting loan service at a first preset observation point from all sample clients as a first sample client; the first preset observation point comprises a time period which takes a time point which is earlier than the preset loan expiration time by a first preset time as an initial point and takes the preset loan expiration time as an end point; classifying the first sample client according to a first preset classification rule to obtain the type of the first sample client, and setting a type label corresponding to the type of the first sample client for the first sample client; the first preset classification rule is as follows: identifying a first sample client which pays in a first preset presentation period as a good sample, and identifying a first sample client which does not pay in the first preset presentation period as a bad sample; the first preset presentation period comprises a time period which takes the ending time of the first preset observation point as the beginning and is delayed by second preset time; constructing a first sample set based on sample characteristics of a first sample client; and training a preset logistic regression model by using the first sample set to obtain a pre-collection scoring card model.

Optionally, the scoring card model comprises an overdue collection scoring card model; the overdue collection scoring card model is obtained by training a preset logistic regression model by using a second sample set acquired in advance.

The scoring unit 300 is specifically configured to: under the condition that the type of the client to be detected is an overdue client, inputting the target characteristics into an overdue urging receiving scoring card model to obtain a second scoring result output by the overdue urging receiving scoring card model; the second scoring result comprises overdue gathering scoring of the client to be tested.

The scoring unit 300 is specifically configured to: obtaining loan data of a plurality of sample customers from a loan service system in advance; the sample clients comprise clients handling loan service within a preset observation period; performing feature extraction on loan data of each sample customer to obtain a feature variable set; the characteristic variable set comprises a plurality of characteristic variables; performing variable screening on each characteristic variable in the characteristic variable set to obtain each sample characteristic; selecting sample clients transacting loan service at a second preset observation point from all the sample clients as second sample clients; the second preset observation point comprises a preset loan overdue time period; classifying the second sample client according to a second preset classification rule to obtain the type of the second sample client, and setting a type label corresponding to the type of the second sample client for the second sample client; the second preset classification rule is as follows: identifying a second sample client which pays in a second preset presentation period as a good sample, and identifying a second sample client which does not pay in the second preset presentation period as a bad sample; the second preset presentation period comprises a time period which takes the ending time of the second preset observation point as the beginning and is delayed by a second preset time; constructing a second sample set based on sample characteristics of a second sample client; and training the preset logistic regression model by using the second sample set to obtain an overdue collection scoring card model.

The identification unit 400 is used for inputting the scores, the deposits and the loan balances of the clients to be detected into the decision tree model corresponding to the scores to obtain the identification result output by the decision tree model; the decision tree model is constructed by utilizing a pre-acquired training data set; the identification result comprises the urging grade of the customer to be detected.

Optionally, the decision tree model comprises a first decision tree model; the first decision tree model is obtained by utilizing a first training data set acquired in advance.

The identifying unit 400 is specifically configured to: inputting the pre-urging acceptance score, the deposit and the loan balance of the client to be detected into a first decision tree model to obtain a first identification result output by the first decision tree model; the first recognition result comprises the pre-urging receiving grade of the client to be detected.

The identifying unit 400 is specifically configured to: inputting the sample characteristics of the first sample client into a pre-collection grading card model to obtain a first sample grading result output by the pre-collection grading card model; the first sample scoring result comprises a pre-collection score of the first sample client; analyzing the loan data of the first sample client to obtain the deposit and loan balance of the first sample client; constructing a first training data set by utilizing the pre-urging receiving score, the deposit and the loan balance of a first sample client; a first decision tree model is trained using a first training data set.

Optionally, the decision tree model comprises a second decision tree model; the second decision tree model is obtained by utilizing a second training data set obtained in advance.

The identifying unit 400 is specifically configured to: inputting overdue credit rating and loan balance of the customer to be detected into a second decision tree model to obtain a second identification result output by the second decision tree model; the second recognition result comprises the overdue collection level of the client to be detected.

The identifying unit 400 is specifically configured to: inputting the sample characteristics of the second sample client into the overdue collection grading card model to obtain a second sample grading result output by the overdue collection grading card model; the second sample scoring result comprises overdue incentive scores of the second sample clients; analyzing the loan data of the second sample client to obtain the loan balance of the second sample client; constructing a second training data set by utilizing overdue credit rating and loan balance of a second sample client; a second decision tree model is trained using a second training data set.

And the display unit 500 is used for displaying the collection level of the customer to be tested to the user through a preset interface.

In summary, the score of the client to be tested is determined based on the scoring card model, the type and the target characteristics of the client to be tested, the collection level of the client to be tested is determined based on the decision tree model and the score, deposit and loan balance of the client to be tested, and the user can be helped to learn the collection level of the client to be tested, so that the user can make collection preparation in advance according to the collection level of the client to be tested (for example, the higher the collection level is, the more time is spent to promote the client to pay back), and the loan of the client to be tested can be effectively recovered.

The functions described in the method of the embodiment of the present application, if implemented in the form of software functional units and sold or used as independent products, may be stored in a storage medium readable by a computing device. Based on such understanding, part of the contribution to the prior art of the embodiments of the present application or part of the technical solution may be embodied in the form of a software product stored in a storage medium and including several instructions for causing a computing device (which may be a personal computer, a server, a mobile computing device or a network device) to execute all or part of the steps of the method described in the embodiments of the present application. And the aforementioned storage medium includes: u disk, removable hard disk, read only memory, random access memory, magnetic or optical disk, etc. for storing program codes.

The embodiments are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same or similar parts among the embodiments are referred to each other.

The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the application. Thus, the present application is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims (10)

1. A loan collection method, comprising:

after receiving loan data of a client to be tested sent by a user, analyzing the loan data of the client to be tested to obtain the type, deposit and loan balance of the client to be tested;

performing feature extraction on the loan data of the customer to be detected to obtain target features;

inputting the target characteristics into a grading card model corresponding to the type of the customer to be tested to obtain a grading result output by the grading card model; the scoring card model is obtained by training a preset logistic regression model by using a pre-obtained sample set; the scoring result comprises the scoring of the customer to be tested and the type of the scoring;

inputting the score, deposit and loan balance of the client to be detected into a decision tree model corresponding to the type of the score to obtain an identification result output by the decision tree model; the decision tree model is constructed by utilizing a pre-acquired training data set; the identification result comprises the collection level of the customer to be tested;

and displaying the collection level of the customer to be tested to the user through a preset interface.

2. The method of claim 1, wherein the scoring card model comprises a pre-collection scoring card model; the pre-collection scoring card model is obtained by training a preset logistic regression model by utilizing a first sample set acquired in advance;

the step of inputting the target characteristics into a scoring card model corresponding to the type of the customer to be tested to obtain a scoring result output by the scoring card model comprises the following steps:

under the condition that the type of the customer to be tested is a non-overdue customer, inputting the target characteristics into the pre-collection scoring card model to obtain a first scoring result output by the pre-collection scoring card model; the first scoring result comprises pre-urging receiving scoring of the client to be tested.

3. The method according to claim 2, wherein the training of the pre-set logistic regression model with the pre-obtained first sample set to obtain the pre-collection score card model comprises:

obtaining loan data of a plurality of sample customers from a loan service system in advance; the sample clients comprise clients handling loan service within a preset observation period;

performing feature extraction on the loan data of each sample client to obtain a feature variable set; the set of characteristic variables comprises a plurality of characteristic variables;

performing variable screening on each characteristic variable in the characteristic variable set to obtain each sample characteristic;

selecting sample clients transacting the loan service at a first preset observation point from the sample clients as first sample clients; the first preset observation point comprises a time period which takes a time point which is earlier than a preset loan expiration time by a first preset time as an initial point and takes the preset loan expiration time as an end point;

classifying the first sample client according to a first preset classification rule to obtain the type of the first sample client, and setting a type label corresponding to the type of the first sample client for the first sample client; the first preset classification rule is as follows: identifying a first sample client which pays in a first preset presentation period as a good sample, and identifying a first sample client which does not pay in the first preset presentation period as a bad sample; the first preset presentation period comprises a time period which takes the ending time of the first preset observation point as the beginning and is delayed by a second preset time;

constructing the first sample set based on sample characteristics of the first sample client;

and training a preset logistic regression model by using the first sample set to obtain the pre-collection scoring card model.

4. The method of claim 3, wherein the decision tree model comprises a first decision tree model; the first decision tree model is obtained by utilizing a first training data set obtained in advance;

inputting the score, deposit and loan balance of the client to be tested into a decision tree model corresponding to the type of the score to obtain an identification result output by the decision tree model, wherein the identification result comprises:

inputting the pre-catalytic receiving score, the deposit and the loan balance of the client to be detected into the first decision tree model to obtain a first identification result output by the first decision tree model; the first identification result comprises the pre-urging grade of the customer to be detected.

5. The method according to claim 4, wherein the training the obtained first decision tree model by using the pre-acquired first training data set comprises:

inputting the sample characteristics of the first sample client into the pre-collection grading card model to obtain a first sample grading result output by the pre-collection grading card model; the first sample scoring results include a pre-catalytic acceptance score for the first sample customer;

analyzing the loan data of the first sample client to obtain the deposit and loan balance of the first sample client;

constructing the first training data set by utilizing the pre-urging receiving score, the deposit and the loan balance of the first sample client;

training the first decision tree model using the first training data set.

6. The method of claim 1, wherein the scoring card model comprises an overdue claim scoring card model; the overdue collection scoring card model is obtained by training a preset logistic regression model by using a second sample set acquired in advance;

the step of inputting the target characteristics into a scoring card model corresponding to the type of the customer to be tested to obtain a scoring result output by the scoring card model comprises the following steps:

under the condition that the type of the client to be detected is an overdue client, inputting the target characteristics into the overdue collection scoring card model to obtain a second scoring result output by the overdue collection scoring card model; and the second grading result comprises overdue collection grading of the client to be tested.

7. The method of claim 6, wherein the training of the pre-defined logistic regression model with the pre-obtained second sample set to obtain the overdue claim card model comprises:

obtaining loan data of a plurality of sample clients from a loan service system in advance; the sample clients comprise clients handling loan service within a preset observation period;

performing feature extraction on the loan data of each sample client to obtain a feature variable set; the set of characteristic variables comprises a plurality of characteristic variables;

performing variable screening on each characteristic variable in the characteristic variable set to obtain each sample characteristic;

selecting a sample client transacting the loan service at a second preset observation point from the sample clients as a second sample client; the second preset observation point comprises a preset loan overdue time period;

classifying the second sample client according to a second preset classification rule to obtain the type of the second sample client, and setting a type label corresponding to the type of the second sample client for the second sample client; the second preset classification rule is as follows: identifying a second sample client paid in a second preset presentation period as a good sample, and identifying a second sample client not paid in the second preset presentation period as a bad sample; the second preset presentation period comprises a time period which takes the ending time of the second preset observation point as the beginning and is delayed by second preset time;

constructing the second sample set based on sample characteristics of the second sample client;

and training a preset logistic regression model by using the second sample set to obtain the overdue collection scoring card model.

8. The method of claim 7, wherein the decision tree model comprises a second decision tree model; the second decision tree model is obtained by utilizing a second training data set obtained in advance;

inputting the score, deposit and loan balance of the client to be detected into a decision tree model corresponding to the type of the score to obtain an identification result output by the decision tree model, wherein the identification result comprises:

inputting the overdue credit rating and the loan balance of the customer to be detected into the second decision tree model to obtain a second recognition result output by the second decision tree model; and the second identification result comprises the overdue hastening grade of the client to be detected.

9. The method according to claim 8, wherein the training of the obtained second decision tree model using a second training data set obtained in advance comprises:

inputting the sample characteristics of the second sample client into the overdue collection scoring card model to obtain a second sample scoring result output by the overdue collection scoring card model; the second sample scoring results comprise overdue incentive scores of the second sample clients;

analyzing the loan data of the second sample client to obtain the loan balance of the second sample client;

constructing the second training data set by using the overdue collection score and the loan balance of the second sample client;

training the second decision tree model using the second training data set.

10. A loan collection apparatus, comprising:

the system comprises an analysis unit, a processing unit and a processing unit, wherein the analysis unit is used for analyzing loan data of a client to be tested after receiving the loan data of the client to be tested sent by a user to obtain the type, deposit and loan balance of the client to be tested;

the extracting unit is used for extracting the features of the loan data of the customer to be detected to obtain target features;

the scoring unit is used for inputting the target characteristics into a scoring card model corresponding to the type of the client to be tested to obtain a scoring result output by the scoring card model; the scoring card model is obtained by training a preset logistic regression model by using a pre-obtained sample set; the scoring result comprises the scoring of the customer to be tested and the type of the scoring;

the identification unit is used for inputting the score, the deposit and the loan balance of the client to be detected into a decision tree model corresponding to the type of the score to obtain an identification result output by the decision tree model; the decision tree model is constructed by utilizing a pre-acquired training data set; the identification result comprises the collection level of the customer to be tested;

and the display unit is used for displaying the collection level of the customer to be tested to the user through a preset interface.

Images