CN113450207A - Intelligent collection accelerating method, device, equipment and storage medium - Google Patents
Intelligent collection accelerating method, device, equipment and storage medium Download PDFInfo
- Publication number
- CN113450207A CN113450207A CN202110729495.1A CN202110729495A CN113450207A CN 113450207 A CN113450207 A CN 113450207A CN 202110729495 A CN202110729495 A CN 202110729495A CN 113450207 A CN113450207 A CN 113450207A
- Authority
- CN
- China
- Prior art keywords
- overdue
- clients
- client
- stage
- machine learning
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 56
- 238000003860 storage Methods 0.000 title claims abstract description 11
- 238000010801 machine learning Methods 0.000 claims abstract description 62
- 238000012549 training Methods 0.000 claims abstract description 38
- 230000006399 behavior Effects 0.000 claims abstract description 18
- 238000012360 testing method Methods 0.000 claims description 19
- 238000004458 analytical method Methods 0.000 claims description 18
- 238000012216 screening Methods 0.000 claims description 16
- 238000004422 calculation algorithm Methods 0.000 claims description 13
- 238000003306 harvesting Methods 0.000 claims description 12
- 238000004590 computer program Methods 0.000 claims description 10
- 238000010219 correlation analysis Methods 0.000 claims description 10
- 230000008569 process Effects 0.000 claims description 7
- 239000003054 catalyst Substances 0.000 claims description 5
- 238000007473 univariate analysis Methods 0.000 claims description 5
- 238000000638 solvent extraction Methods 0.000 claims description 2
- 230000003542 behavioural effect Effects 0.000 description 6
- 230000006870 function Effects 0.000 description 5
- 230000000694 effects Effects 0.000 description 4
- 238000005457 optimization Methods 0.000 description 4
- 230000003203 everyday effect Effects 0.000 description 3
- 238000012544 monitoring process Methods 0.000 description 3
- 238000012946 outsourcing Methods 0.000 description 3
- 238000006243 chemical reaction Methods 0.000 description 2
- 238000013277 forecasting method Methods 0.000 description 2
- 238000011835 investigation Methods 0.000 description 2
- 230000001960 triggered effect Effects 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000013145 classification model Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 230000000977 initiatory effect Effects 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000010202 multivariate logistic regression analysis Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
- G06Q40/03—Credit; Loans; Processing thereof
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
Abstract
The invention discloses an intelligent collection method, an intelligent collection device, intelligent collection equipment and a storage medium, wherein the method comprises the following steps: determining overdue clients aiming at a credit scene, and dividing different stages into overdue stages for each overdue client according to the current overdue days of each overdue client; predicting the probability of each overdue client continuing to overdue to the next stage according to the machine learning model corresponding to each overdue stage; grouping all overdue clients according to the basic information of each overdue client and the probability of continuing overdue to the next stage to obtain at least one guest group, and urging to collect the overdue clients in each guest group according to the corresponding urging mode of each guest group; the machine learning model corresponding to any overdue stage is obtained by training in advance by utilizing behavior information corresponding to the clients historically belonging to the any overdue stage during loan making. The customer overdue intelligent collection scheme designed aiming at the credit scene has the advantages of high intelligent level, low artificial dependence degree and capability of realizing efficient collection.
Description
Technical Field
The invention relates to the technical field of data processing, in particular to an intelligent collection urging method, device, equipment and storage medium.
Background
If the debt is not cleared beyond the specified period after the client loan, the client loan is called overdue. At present, for a client in an overdue state, staff usually monitors and pays corresponding collection after finding so as to prompt the client to pay as soon as possible. However, the mode has low intelligent level and high dependence on manpower, and cannot realize high-efficiency collection.
Disclosure of Invention
The invention aims to provide an intelligent collection accelerating method, device, equipment and storage medium, which have high intelligent level and low artificial dependence degree and can realize high-efficiency collection accelerating.
In order to achieve the above purpose, the invention provides the following technical scheme:
an intelligent harvesting method, comprising:
determining that the clients in the overdue state are overdue clients, and dividing different stages of the overdue clients into overdue stages according to the current number of overdue days of each overdue client;
predicting the probability of each overdue client continuing overdue to the next stage according to the machine learning model corresponding to each overdue stage; the machine learning model corresponding to any overdue stage is obtained by training in advance by utilizing behavior information corresponding to the clients historically belonging to the any overdue stage during loan making;
grouping all overdue customers according to the basic information of each overdue customer and the probability of continuing overdue to the next stage to obtain at least one corresponding customer group, and urging the overdue customers in each customer group according to an urging mode corresponding to each customer group.
Preferably, the training of the behavioral performance data corresponding to the past clients making loans in advance to obtain the machine learning model includes:
determining any overdue stage as a current stage, and obtaining corresponding behavior information which belongs to corresponding customers in the past when the plurality of customers in the current stage loan, wherein the behavior information belongs to samples of the corresponding customers;
and training to obtain a machine learning model corresponding to the current stage by utilizing a LightGBM machine learning algorithm based on the sample.
Preferably, after the sample is obtained, the method further comprises:
and screening out variables with unqualified prediction capability from the variables contained in the sample to delete the variables based on univariate analysis, correlation analysis and feature importance analysis.
Preferably, the training to obtain the machine learning model corresponding to the current stage by using the LightGBM machine learning algorithm based on the sample includes:
dividing the sample into a training sample and a test sample;
for the training sample, learning sample characteristics by using a LightGBM machine learning algorithm to obtain a current model;
and testing the current model by using the test sample to obtain the prediction accuracy of the current model, determining the current model as a final machine learning model if the prediction accuracy of the current model meets the requirement, and otherwise, returning to the step of performing single variable analysis, correlation analysis and feature importance analysis-based screening of the corresponding variables.
Preferably, the method further comprises the following steps:
the step of obtaining the sample is performed every lapse of a preset time interval.
Preferably, the urging collection of overdue customers in each customer group according to the urging collection mode corresponding to each customer group includes:
based on the total amount of clients of overdue clients in an overdue state at present and the number of clients of overdue clients under different client groups, dividing the client groups into an automatic collection queue and a manual collection queue respectively;
all overdue customers under the automatic collection queue are automatically collected in batches through a collection system, and all overdue customers under the manual collection queue are collected in sequence through manual collection.
Preferably, the method further comprises the following steps:
in the process of urging the overdue client to receive, carrying out voice recognition on voice contained in the urged dialog to obtain corresponding text information;
and performing semantic analysis on the text information to obtain the repayment willingness and repayment capacity of the corresponding overdue client, and recording.
An intelligent harvesting device comprising:
a partitioning module to: determining that the clients in the overdue state are overdue clients, and dividing different stages of the overdue clients into overdue stages according to the current number of overdue days of each overdue client;
a prediction module to: predicting the probability of each overdue client continuing overdue to the next stage according to the machine learning model corresponding to each overdue stage; the machine learning model corresponding to any overdue stage is obtained by training in advance by utilizing behavior information corresponding to the clients historically belonging to the any overdue stage during loan making;
a catalyst module for: grouping all overdue customers according to the basic information of each overdue customer and the probability of continuing overdue to the next stage to obtain at least one corresponding customer group, and urging the overdue customers in each customer group according to an urging mode corresponding to each customer group.
An intelligent harvesting apparatus comprising:
a memory for storing a computer program;
a processor for implementing the steps of the intelligent collection method when executing the computer program.
A computer-readable storage medium having stored thereon a computer program which, when executed by a processor, carries out the steps of the smart catalyst method as claimed in any one of the preceding claims.
The invention provides an intelligent collection method, an intelligent collection device, intelligent collection equipment and a storage medium, wherein the method comprises the following steps: determining that the clients in the overdue state are overdue clients, and dividing different stages of the overdue clients into overdue stages according to the current number of overdue days of each overdue client; predicting the probability of each overdue client continuing overdue to the next stage according to the machine learning model corresponding to each overdue stage; grouping all overdue customers according to the basic information of each overdue customer and the probability of continuing overdue to the next stage to obtain at least one corresponding customer group, and urging the overdue customers in each customer group according to an urging mode corresponding to each customer group; the machine learning model corresponding to any overdue stage is obtained by training in advance by utilizing behavior information corresponding to the clients historically belonging to the any overdue stage during loan making. According to the method, for the clients in the overdue state in batch at present, the overdue stage to which the clients belong is determined based on the current overdue days of the clients, the probability that the clients continue to overdue to the next stage is predicted according to a machine learning model corresponding to the overdue stage to which the clients belong, then the clients are divided into different client groups based on the probability that the clients continue to overdue to the next stage and basic information, and then clients in the client groups are prompted according to corresponding client group prompting modes. Therefore, the client overdue forecasting method and the client terminal have the advantages that whether the client continues overdue forecasting is achieved by the aid of the corresponding models according to the overdue days of the client, and after passenger groups are divided based on forecasting results and basic information of the client, collection prompting is achieved by the aid of corresponding collection prompting modes of the corresponding passenger groups, so that the client can automatically collect prompting without manual participation, the client continues overdue forecasting and the collection prompting modes used by the client for collection are all in accordance with the overdue days, the basic information and the like of the client, effectiveness of collection prompting is improved, intelligence level is high, manual dependence degree is low, and efficient collection prompting can be achieved.
Drawings
In order to more clearly illustrate the embodiments of the present invention 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 embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the provided drawings without creative efforts.
FIG. 1 is a first flowchart of an intelligent harvesting method according to an embodiment of the present invention;
FIG. 2 is a second flowchart of an intelligent harvesting method according to an embodiment of the present invention;
fig. 3 is a schematic structural diagram of an intelligent harvesting device according to an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, 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 invention.
Referring to fig. 1, a flowchart of an intelligent collection method according to an embodiment of the present invention is shown, which specifically includes:
s11: and determining that the clients in the overdue state are overdue clients, and dividing different stages of the clients to be overdue stages for the clients according to the current overdue days of each overdue client.
The embodiment of the application is specifically realized for a credit scene, and for the clients in the overdue state in the existing batch, corresponding stages can be divided according to the current overdue days of the clients. Specifically, the classification of the stages according to the number of overdue days may be a preset rule, in which the number of the overdue days and the corresponding stage are specified, and further, according to the rule, the stage corresponding to the number of the overdue days of any client is determined to be the overdue stage to which the any client belongs. If the overdue days are 1-15 days in mild overdue period, 15-30 days in secondary moderate overdue period, 30-60 days in moderate overdue period, 60-90 days in severe overdue period, and more than 90 days in bad clearance period; and if the current overdue days of any client are within the range, the client is the overdue stage to which the corresponding range belongs.
The longer the overdue days of any client are, the more urgent the operation of corresponding charging of any client is, so that in the embodiment of the application, the overdue stages to which the client belongs are divided according to the overdue days of the client, the charging operation is further realized based on the overdue stages to which the client belongs, the charging operation can be made to correspond to the overdue conditions of the client, and the accuracy of the charging operation is improved.
S12: predicting the probability of each overdue client continuing to overdue to the next stage according to the machine learning model corresponding to each overdue stage; the machine learning model corresponding to any overdue stage is obtained by training in advance by utilizing behavior information corresponding to the clients historically belonging to the any overdue stage during loan making.
Different machine learning models can be created in advance for different overdue stages, namely the overdue stages and the machine learning models can be in one-to-one correspondence; on this basis, after determining the overdue stage to which any client belongs, the embodiment of the application can predict the probability that any client rolls to the next stage without paying back in the overdue stage by using the machine learning model corresponding to the overdue stage to which the client belongs. Specifically, for a machine learning model corresponding to any overdue stage, behavior performance information of a plurality of clients historically belonging to the any overdue stage during loan is acquired in advance, the machine learning model corresponding to the any overdue stage is obtained through training by using the behavior performance information, and finally the behavior performance information of the overdue clients belonging to the any overdue stage is processed by using the machine learning model corresponding to the any overdue stage, so that the probability that the overdue clients roll to the next stage without paying back in the any overdue stage is predicted. The behavior data can be selected according to actual needs, such as loan application, loan payment, repayment information, history overdue, quota utilization rate data, client credit investigation data and the like, so that corresponding probability prediction is realized by fully utilizing data information before and after client loan, and accuracy of probability prediction is improved. In addition, any overdue client continues to overdue to the next stage, namely, the any overdue client cannot pay back debts in the overdue stage and continues to be in an overdue state.
S13: grouping all overdue clients according to the basic information of each overdue client and the probability of continuing overdue to the next stage to obtain at least one corresponding client group, and urging the overdue clients in each client group according to an urging mode corresponding to each client group.
After the probability that each overdue client continues overdue to the next stage is obtained, client portrait can be drawn by combining the basic information of the overdue clients, and corresponding client group labels are generated; specifically, the basic information may include client work industries (such as buildings, communications, and the like), client work unit properties (such as national enterprises, private enterprises, special units, and the like), and further, the client groups may be divided according to different client work industries, client work unit properties, and the probability that the client continues to overdue to the next stage, so as to obtain at least one client group, and a corresponding client group tag may be set for each client group, so as to distinguish different client groups. In addition, the grouping can be realized by using a clustering algorithm, or can be realized by indicating related personnel to use business meanings and experiences, and of course, other settings can be performed according to actual needs, which are within the protection scope of the invention. The reason for grouping the overdue clients is to know the overall structure of the overdue clients and the similarity of the groups, so that the similar overdue clients are divided into the same group and have the same group label, and the overdue clients can be conveniently and subsequently called in the group according to the calling mode corresponding to the group. In addition, because the overdue clients in each customer group are similar, the different overdue clients in the same customer group can be urged to receive according to a uniform urging mode, on the basis, the corresponding urging mode is set for different customer groups, and then urging is realized for the overdue clients in any customer group according to the urging mode corresponding to any customer group, so that urging of the overdue clients is consistent with relevant information of the overdue clients, and urging has high accuracy.
According to the method, for the clients in the overdue state in batch at present, the overdue stage to which the clients belong is determined based on the current overdue days of the clients, the probability that the clients continue to overdue to the next stage is predicted according to a machine learning model corresponding to the overdue stage to which the clients belong, then the clients are divided into different client groups based on the probability that the clients continue to overdue to the next stage and basic information, and then clients in the client groups are prompted according to corresponding client group prompting modes. Therefore, the client overdue forecasting method and the client terminal have the advantages that whether the client continues overdue forecasting is achieved by the aid of the corresponding models according to the overdue days of the client, and after passenger groups are divided based on forecasting results and basic information of the client, collection prompting is achieved by the aid of corresponding collection prompting modes of the corresponding passenger groups, so that the client can automatically collect prompting without manual participation, the client continues overdue forecasting and the collection prompting modes used by the client for collection are all in accordance with the overdue days, the basic information and the like of the client, effectiveness of collection prompting is improved, intelligence level is high, manual dependence degree is low, and efficient collection prompting can be achieved.
The intelligent collection method provided by the embodiment of the invention is characterized in that a machine learning model is obtained by training corresponding behavioral data of a historical customer during loan, and the method comprises the following steps:
determining any overdue stage as a current stage, and obtaining corresponding behavior information which belongs to corresponding customers in the past when the plurality of customers in the current stage loan, wherein the behavior information belongs to samples of the corresponding customers;
and training to obtain a machine learning model corresponding to the current stage by utilizing a LightGBM machine learning algorithm based on the sample.
According to the embodiment of the application, the LightGBM machine learning algorithm can be adopted to construct the two-classification machine learning model, so that the obtained machine learning model is ensured to have higher accuracy. Specifically, the aim of constructing the machine learning model in the embodiment of the application is to construct a binary model for determining whether debt is paid back within M-N days by using historical customer samples; and the step of modeling may include:
a) preparing data:
sample acquisition: selecting a sample which is recently put in loan and has complete performance after the loan according to the product attributes and the customer quantity of the loan products, and analyzing the sample quantity of the month/quarter, the overdue and repayment performance of the customer;
the target variable defines: if the customer pays within the overdue M-N days, the good customer Y is 0, otherwise the bad customer Y is 1; such as: machine learning model of overdue 1-15 days, target variables are defined for overdue customers, if the customer repays within 1-15 days of overdue, then it is good customer (Y ═ 0), otherwise it is bad customer (Y ═ 1).
b) The characteristic structure is as follows:
and establishing behavioral expression variables of the customer dimension as behavioral expression data, wherein the behavioral expression variables can comprise loan application, loan payment, repayment information, historical overdue, quota utilization rate data, customer credit investigation data and the like, so that a perfect characteristic system of the customer dimension is constructed by fully utilizing data information before and after the customer loan for model prediction.
c) And (3) feature screening:
and deleting the variables contained in the sample which do not meet the requirements.
d) And in the model training link, carrying out optimal model hyperparametric search by using a Bayesian optimization parameter adjusting method.
Through the steps, the machine learning model with strong prediction capability is obtained through training, so that the repayment probability of the client in M-N days, namely the probability of continuing overdue to the next stage, is predicted.
The intelligent collection method provided by the embodiment of the invention can further comprise the following steps of after the sample is obtained:
and screening out variables with unqualified prediction capability from the variables contained in the sample to delete the variables based on univariate analysis, correlation analysis and feature importance analysis.
Specifically, the sample screening may be to screen out variables with strong prediction ability as variables meeting requirements by using univariate analysis, correlation analysis and feature importance analysis, and retain the variables as model-in variables, while deleting other variables. Specifically, the condition of each variable can be analyzed through single-variable analysis (loss rate, stability, variable binning and the like), and the variables with high loss degree, instability, unobvious distinction and poor business interpretability are eliminated; then analyzing pairwise correlation (namely correlation analysis) among the variables through multi-variable analysis, and removing the variables with weak distinctiveness so as to simplify the model-entering variables; and finally, calculating the feature importance of the variables by using a LightGBM algorithm, and removing the variables with lower feature importance. In the single variable analysis, aiming at the loss degree of the variable, the variable with the loss degree higher than 90% can be removed, the occupation ratio of a single value of the variable is analyzed, the variable with the occupation ratio higher than 95% of the single value is removed, and the information degree of the variable is calculated; in the correlation analysis, variables with high correlation (higher than 0.8) and low information degree can be eliminated. Therefore, the variables which can play an effective classification role in the model training process are screened out through screening the variables, and the number of the variables is reduced, so that the model training speed is accelerated while the model is effectively trained.
The intelligent collection method provided by the embodiment of the invention is characterized in that a LightGBM machine learning algorithm is utilized based on a sample to train and obtain a machine learning model corresponding to the current stage, and the method comprises the following steps:
dividing a sample into a training sample and a test sample;
aiming at the training sample, learning sample characteristics by using a LightGBM machine learning algorithm to obtain a current model;
and testing the current model by using the test sample to obtain the prediction accuracy of the current model, if the prediction accuracy of the current model meets the requirement, determining the current model as a final machine learning model, and if not, returning to the step of performing single variable analysis, correlation analysis and feature importance analysis-based screening of the corresponding variables.
It should be noted that, in the embodiment of the present application, a sample can be divided into a training sample and a testing sample, so that model training is implemented by using the training sample, model testing is implemented by using the testing sample, and a final machine learning model is determined after the test is passed, thereby ensuring accuracy of probability prediction implemented by using the machine learning model. When the test sample is used for realizing the model test, the AUC and KS indexes can be used for evaluating the orderliness and the distinguishing capability of the model, so that when the values of the AUC and KS of the model respectively meet corresponding requirements, the model is determined to be a model which can be subsequently used for realizing probability prediction, and otherwise, the steps of sample division, variable screening, model training, model test and the like are realized again until the model test is passed.
The intelligent harvesting method provided by the embodiment of the invention can further comprise the following steps:
the step of obtaining the sample is performed every time a preset time interval elapses.
The preset time interval can be set according to actual needs, such as months or seasons; the model iteration module can automatically perform the steps of sample screening, model training, model testing and the like according to a set model iteration period (namely a preset time interval) until a final machine learning model corresponding to the currently screened sample is obtained, so that the model is stably suitable for different hastening scenes and changed overdue customer groups, and a better probability prediction effect is ensured. It should be noted that, each time a sample is obtained, data information within a period of time closest to the current time may be obtained, so as to ensure that the machine learning model obtained based on the currently obtained sample adapts to changes of the customer group.
In addition, the prediction accuracy of the machine learning model can be automatically analyzed and calculated according to the set monitoring frequency, and then when the prediction accuracy does not meet the corresponding requirement, the model iteration module is triggered to carry out steps of sample screening, model training, model testing and the like again.
When the model is retrained, the sample screening, the target variable definition and the like can be realized according to the above mode to obtain a new sample, the running number of the in-mode variables and the model training function are automatically completed, the model file corresponding to the newly trained optimal model is stored in a set system path, and then the latest model is automatically called to carry out probability prediction. Therefore, the method realizes the autonomous monitoring and the automatic iterative optimization of the model so as to adapt to the rapid development and change of the business.
The intelligent collection method provided by the embodiment of the invention is used for collecting overdue customers in each passenger group according to the collection method corresponding to each passenger group, and comprises the following steps:
based on the total amount of clients of overdue clients in an overdue state and the number of clients of overdue clients under different client groups, dividing the client groups into an automatic collection urging queue and a manual collection urging queue;
all overdue customers under the automatic collection queue are automatically collected in batches through the collection system, and all overdue customers under the manual collection queue are collected in sequence through manual collection.
It should be noted that, in the embodiment of the present application, the customer groups can be adaptively classified into the automatic hastening and collecting queue and the manual hastening and collecting queue according to the number of customers to be hastened on the same day and the number distribution of each customer group, so that the hastening and collecting of the corresponding customers can be automatically realized through manual or hastening and collecting systems, and it is ensured that customers to be hastened on different orders of magnitude can be effectively hastened and collected on the same day. Specifically, the amount of the clients to be promoted refers to the total amount of the clients (M) in an overdue state every day, the number of the clients under labels corresponding to different client groups, and the amount of the clients needing to enter a charge queue on the day; the collection queue is divided into two categories of manual collection and automatic collection of the system, the manual collection is limited by the number of fixed workers, so that the quantity of customers which can collect every day is limited, and the automatic collection of the system is not limited by manpower and can be initiated in a large scale; therefore, when the amount of overdue customers suddenly increases on a certain day, the proportion of customers automatically urged to be received by the system needs to be increased, the system refers to the amount of manpower for urging, the proportion of overdue customers which need to be allocated to the system for automatic urging is automatically calculated, and the rest of the overdue customers are allocated to manual urging. Such as: the manual collection queue has 15 workers, and the maximum collection amount per day is 1000 clients (Limit is 1000); the system sets that min (M R1, Limit) overdue clients can be automatically placed into the manual collection queue every day, and when the total M of the overdue clients suddenly rises to cause that M R1 is larger than Limit, only the Limit overdue clients are placed into the manual collection queue, so that the efficiency and load balance of the manual collection queue are guaranteed. In addition, the automatic collection queue and the manual collection queue can be subdivided into a short message collection queue, an intelligent voice collection queue, a manual collection queue and an outsourcing collection queue, wherein the short message collection queue is used for automatically sending short messages to corresponding overdue customers by the collection system to realize collection, the intelligent voice collection queue is used for automatically sending voices to corresponding overdue customers by the collection system to realize collection, the manual collection queue is used for realizing collection by calling or sending information to the corresponding overdue customers by workers, the outsourcing collection queue is used for entrusting the corresponding personnel to realize collection by calling or sending information to the corresponding overdue customers, and the automatic collection queue can be used for batch collection so as to further improve collection flexibility.
The intelligent harvesting method provided by the embodiment of the invention can further comprise the following steps:
in the process of urging the overdue client to receive, carrying out voice recognition on the voice contained in the urge-to-receive conversation to obtain corresponding text information;
and performing semantic analysis on the text information to obtain the repayment willingness and repayment capacity of the corresponding overdue client, and recording.
It should be noted that, the embodiment of the application can effectively record the collection process for subsequent query and other operations; the method comprises the steps that a speech recognition module is used for carrying out character conversion on speech of a collection prompting conversation based on a deep learning algorithm to obtain corresponding text information, and a Chinese character string of one sentence is formed; extracting keywords in the text information through semantic analysis, understanding and analyzing the repayment willingness and repayment capacity of the client by combining the keywords and the context, and further automatically generating an acceptance result label; specifically, the payment willingness and payment capability of the customer can be judged through the set specific questions and keywords in the customer answers, for example:
asking you can return the arrears in time? Cannot be relaxed for several days? -a keyword: the payment system has the advantages of wide limit, moderate payment willingness and moderate payment capability;
asking you can return the arrears in time? -? -a keyword: good, high repayment willingness and high repayment capacity;
asking you can return the arrears in time? -on-hook or no money in answer? -a keyword: hang up/lack of money, low willingness to repay, low repayment capacity.
In a specific implementation manner, an intelligent collection method provided by an embodiment of the present invention is shown in fig. 2, and may include:
1) the overdue stages of the clients are divided according to the current overdue days of the overdue clients aiming at the overdue clients in batches;
2) predicting the probability of the overdue client continuing to roll to the next stage by using machine learning models of different overdue stages;
3) and setting separate machine learning models in different overdue stages, wherein the modeling samples and model variables of each machine learning model are different. And aiming at the overdue client which just enters the overdue client, predicting whether the overdue client pays the overdue debt in the overdue client, and if the overdue client does not pay the loan in the overdue client, indicating that the overdue client can continue to overdue and roll to the next stage. Thus, the machine learning model predicts the probability that a overdue customer will roll to the next overdue stage without having cleared the overdue debt within the overdue stage.
The modeling method comprises the following steps: method for constructing two classification models by adopting machine learning method (LightGBM)
Model object: and constructing a binary model of whether the payment is paid within M-N days by using overdue samples of historical customers.
Modeling:
a) preparing data: sample analysis and screening, and target variable definition.
b) The characteristic structure is as follows: and establishing behavioral variables of the customer dimension.
c) And screening out a characteristic subset with strong prediction capability by utilizing univariate analysis, correlation analysis and characteristic importance analysis, and taking the characteristic subset as a model input variable.
d) And (4) carrying out optimal model hyperparametric search by using a Bayesian optimization parameter adjusting method in a model training link.
e) In the evaluation link of the model, the AUC and KS indexes are used for evaluating the orderliness and the distinguishing capability of the model.
4) Meanwhile, basic information of the customer (credit information can also be combined at the same time) is combined to perform customer imaging, and a customer group label is generated.
5) The system adaptively divides the customers into a short message collection queue, an intelligent voice collection queue, a manual collection queue and an outsourcing collection queue according to the number of the customers to be promoted on the current day and the number distribution of each customer group, and ensures that the customers to be promoted on different orders of magnitude can be effectively collected on the current day.
And initiating the urging and receiving outbound in batch by using basic software and hardware equipment such as short messages and intelligent voice.
6) The collection prompting process is effectively recorded, the voice of the collection prompting conversation is subjected to character conversion through the voice recognition module, the repayment willingness and the repayment capacity of the customer are analyzed through semantic understanding, and then a collection prompting result label is automatically generated.
7) According to the set model iteration period (month or quarter), the model iteration optimization functions of sample screening, feature calculation and automatic training parameter adjustment are automatically carried out, so that the model is ensured to be stably suitable for overdue guests with different hastening scenes and changes
a) The machine learning model is retrained and optimized automatically through a model iteration function of the system, and then a model obtained by training a new sample is obtained, so that a better prediction effect is ensured;
b) according to the set monitoring frequency, a design program automatically analyzes and calculates the model effect (AUC/KS), and if the reduction amplitude of the model effect reaches a certain threshold value, the model iteration function is triggered;
c) and generating a new modeling sample according to a set sample selection rule and a target variable definition method, autonomously completing the functions of the run-in variable and the model training, storing a model file corresponding to the newly trained optimal model under a set system path, and automatically calling the latest model for scoring in the follow-up process.
The method is suitable for overdue customer collection and management in credit scenes, has the advantages of strong self-adaption capability and high intelligent level, and can be fully suitable for various collection promotion business scenes of banks, consumption financial companies and internet loan companies.
An embodiment of the present invention further provides an intelligent harvesting device, as shown in fig. 3, which may include:
a dividing module 11 configured to: determining that the clients in the overdue state are overdue clients, and dividing different stages of the clients to be overdue stages for the clients according to the current overdue days of each overdue client;
a prediction module 12 for: predicting the probability of each overdue client continuing to overdue to the next stage according to the machine learning model corresponding to each overdue stage; the machine learning model corresponding to any overdue stage is obtained by training in advance by utilizing behavior information corresponding to the clients historically belonging to the any overdue stage during loan making;
a catalyst module 13 for: grouping all overdue clients according to the basic information of each overdue client and the probability of continuing overdue to the next stage to obtain at least one corresponding client group, and urging the overdue clients in each client group according to an urging mode corresponding to each client group.
The embodiment of the invention also provides an intelligent collection device, which comprises:
a memory for storing a computer program;
and the processor is used for realizing the steps of the intelligent collection method when executing the computer program.
The embodiment of the invention also provides a computer-readable storage medium, wherein a computer program is stored on the computer-readable storage medium, and when being executed by a processor, the computer program realizes the steps of the intelligent collection method.
It should be noted that for the description of the relevant parts of the intelligent collection device, the equipment and the storage medium provided in the embodiments of the present invention, reference is made to the detailed description of the corresponding parts of the intelligent collection method provided in the embodiments of the present invention, and no further description is given here. In addition, parts of the technical solutions provided in the embodiments of the present invention that are consistent with the implementation principles of the corresponding technical solutions in the prior art are not described in detail, so as to avoid redundant description.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. 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 invention. Thus, the present invention 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. An intelligent harvesting method is characterized by comprising the following steps:
determining that the clients in the overdue state are overdue clients, and dividing different stages of the overdue clients into overdue stages according to the current number of overdue days of each overdue client;
predicting the probability of each overdue client continuing overdue to the next stage according to the machine learning model corresponding to each overdue stage; the machine learning model corresponding to any overdue stage is obtained by training in advance by utilizing behavior information corresponding to the clients historically belonging to the any overdue stage during loan making;
grouping all overdue customers according to the basic information of each overdue customer and the probability of continuing overdue to the next stage to obtain at least one corresponding customer group, and urging the overdue customers in each customer group according to an urging mode corresponding to each customer group.
2. The method of claim 1, wherein the pre-training of the machine learning model with historical performance data corresponding to the loan made by the customer comprises:
determining any overdue stage as a current stage, and obtaining corresponding behavior information which belongs to corresponding customers in the past when the plurality of customers in the current stage loan, wherein the behavior information belongs to samples of the corresponding customers;
and training to obtain a machine learning model corresponding to the current stage by utilizing a LightGBM machine learning algorithm based on the sample.
3. The method of claim 2, wherein after obtaining the sample, further comprising:
and screening out variables with unqualified prediction capability from the variables contained in the sample to delete the variables based on univariate analysis, correlation analysis and feature importance analysis.
4. The method of claim 3, wherein training a machine learning model corresponding to the current stage based on the samples by using a LightGBM machine learning algorithm comprises:
dividing the sample into a training sample and a test sample;
for the training sample, learning sample characteristics by using a LightGBM machine learning algorithm to obtain a current model;
and testing the current model by using the test sample to obtain the prediction accuracy of the current model, determining the current model as a final machine learning model if the prediction accuracy of the current model meets the requirement, and otherwise, returning to the step of performing single variable analysis, correlation analysis and feature importance analysis-based screening of the corresponding variables.
5. The method of claim 4, further comprising:
the step of obtaining the sample is performed every lapse of a preset time interval.
6. The method of claim 5, wherein collecting overdue customers within each of the groups in a collection manner corresponding to each of the groups comprises:
based on the total amount of clients of overdue clients in an overdue state at present and the number of clients of overdue clients under different client groups, dividing the client groups into an automatic collection queue and a manual collection queue respectively;
all overdue customers under the automatic collection queue are automatically collected in batches through a collection system, and all overdue customers under the manual collection queue are collected in sequence through manual collection.
7. The method of claim 6, further comprising:
in the process of urging the overdue client to receive, carrying out voice recognition on voice contained in the urged dialog to obtain corresponding text information;
and performing semantic analysis on the text information to obtain the repayment willingness and repayment capacity of the corresponding overdue client, and recording.
8. An intelligent harvesting device, comprising:
a partitioning module to: determining that the clients in the overdue state are overdue clients, and dividing different stages of the overdue clients into overdue stages according to the current number of overdue days of each overdue client;
a prediction module to: predicting the probability of each overdue client continuing overdue to the next stage according to the machine learning model corresponding to each overdue stage; the machine learning model corresponding to any overdue stage is obtained by training in advance by utilizing behavior information corresponding to the clients historically belonging to the any overdue stage during loan making;
a catalyst module for: grouping all overdue customers according to the basic information of each overdue customer and the probability of continuing overdue to the next stage to obtain at least one corresponding customer group, and urging the overdue customers in each customer group according to an urging mode corresponding to each customer group.
9. An intelligent harvesting device, comprising:
a memory for storing a computer program;
a processor for implementing the steps of the smart catalyst method according to any one of claims 1 to 7 when executing the computer program.
10. A computer-readable storage medium, having a computer program stored thereon, which, when being executed by a processor, carries out the steps of the smart collection method according to any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110729495.1A CN113450207A (en) | 2021-06-29 | 2021-06-29 | Intelligent collection accelerating method, device, equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110729495.1A CN113450207A (en) | 2021-06-29 | 2021-06-29 | Intelligent collection accelerating method, device, equipment and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113450207A true CN113450207A (en) | 2021-09-28 |
Family
ID=77814119
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110729495.1A Pending CN113450207A (en) | 2021-06-29 | 2021-06-29 | Intelligent collection accelerating method, device, equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113450207A (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109559221A (en) * | 2018-11-20 | 2019-04-02 | 中国银行股份有限公司 | Collection method, apparatus and storage medium based on user data |
| CN110533520A (en) * | 2019-06-06 | 2019-12-03 | 上海凯京信达科技集团有限公司 | A kind of ranking method of the individual customer overdue loan grade based on multi-model |
| CN111062518A (en) * | 2019-11-22 | 2020-04-24 | 成都铂锡金融信息技术有限公司 | Method, device and storage medium for processing hastening service based on artificial intelligence |
| CN111192136A (en) * | 2019-12-24 | 2020-05-22 | 中信百信银行股份有限公司 | Credit service collection method and device, electronic equipment and storage medium |
| CN111882431A (en) * | 2020-08-04 | 2020-11-03 | 武汉众邦银行股份有限公司 | Intelligent message pushing method based on NLP deep learning |
| CN112150271A (en) * | 2020-09-23 | 2020-12-29 | 上海维信荟智金融科技有限公司 | Customer grouping method and system |
-
2021
- 2021-06-29 CN CN202110729495.1A patent/CN113450207A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109559221A (en) * | 2018-11-20 | 2019-04-02 | 中国银行股份有限公司 | Collection method, apparatus and storage medium based on user data |
| CN110533520A (en) * | 2019-06-06 | 2019-12-03 | 上海凯京信达科技集团有限公司 | A kind of ranking method of the individual customer overdue loan grade based on multi-model |
| CN111062518A (en) * | 2019-11-22 | 2020-04-24 | 成都铂锡金融信息技术有限公司 | Method, device and storage medium for processing hastening service based on artificial intelligence |
| CN111192136A (en) * | 2019-12-24 | 2020-05-22 | 中信百信银行股份有限公司 | Credit service collection method and device, electronic equipment and storage medium |
| CN111882431A (en) * | 2020-08-04 | 2020-11-03 | 武汉众邦银行股份有限公司 | Intelligent message pushing method based on NLP deep learning |
| CN112150271A (en) * | 2020-09-23 | 2020-12-29 | 上海维信荟智金融科技有限公司 | Customer grouping method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111291816B (en) | Method and device for carrying out feature processing aiming at user classification model | |
| CN107766929B (en) | Model analysis method and device | |
| CN110837963A (en) | Risk control platform construction method based on data, model and strategy | |
| CN111062518B (en) | Method, device and storage medium for processing collect-promoting business based on artificial intelligence | |
| CN109636482B (en) | Data processing method and system based on similarity model | |
| CN110807699B (en) | Overdue event payment collection method and device and computer readable storage medium | |
| CN111325248A (en) | Method and system for reducing pre-loan business risk | |
| CN110796485A (en) | Method and device for improving prediction precision of prediction model | |
| CN111383093A (en) | Intelligent overdue bill collection method and system | |
| CN111652661B (en) | Mobile phone client user loss early warning processing method | |
| CN113205403A (en) | Method and device for calculating enterprise credit level, storage medium and terminal | |
| CN113435627A (en) | Work order track information-based electric power customer complaint prediction method and device | |
| CN111061948B (en) | User tag recommendation method and device, computer equipment and storage medium | |
| CN116109139A (en) | Wind control strategy generation method, decision method, server and storage medium | |
| CN115545886A (en) | Overdue risk identification method, overdue risk identification device, overdue risk identification equipment and storage medium | |
| CN113112347A (en) | Determination method of hasty collection decision, related device and computer storage medium | |
| CN111967521B (en) | Cross-border active user identification method and device | |
| CN112884569A (en) | Credit assessment model training method, device and equipment | |
| CN113450207A (en) | Intelligent collection accelerating method, device, equipment and storage medium | |
| CN116342255A (en) | Internet consumption credit anti-fraud risk identification method and system | |
| CN111091460A (en) | Data processing method and device | |
| CN116257758A (en) | Model training method, crowd expanding method, medium, device and computing equipment | |
| CN115391539A (en) | Corpus data processing method and device and electronic equipment | |
| Kong et al. | The risk prediction of mobile user tricking account overdraft limit based on fusion model of logistic and GBDT | |
| CN113762415A (en) | Neural network-based intelligent matching method and system for automobile financial products |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210928 |
|
| RJ01 | Rejection of invention patent application after publication |