Customer win-back method and device based on ensemble learning, equipment and storage medium

Abstract

This disclosure provides a customer reactivation method, apparatus, device, and storage medium based on ensemble learning, which can be applied to the field of artificial intelligence technology. The method includes: scoring a sample set of loan applicants using a customer sample application scoring card model and a rejected customer sample application scoring card model; constructing a customer reactivation model based on the scores and a first good / bad sample identifier of the loan applicant sample set; scoring the sample set of loan applicants using an online application scoring card model and the customer reactivation model; binning the scores from the online application scoring card model and the customer reactivation model to obtain a lower limit value for each bin; re-scoring customers rejected by the online application scoring card model using the customer reactivation model; and reactivating customers whose re-scored scores are greater than the lower limit value of the corresponding bin.

Description

Technical Field

[0001] This disclosure relates to the field of artificial intelligence, and more particularly to a customer recapture method, apparatus, device, medium, and program product based on ensemble learning. Background Technology

[0002] Loan application scoring models are tools used to predict the probability of loan default and are widely used in the bank's loan application and approval process. Application scoring models can be divided into expert scoring models and data-driven scoring models. Expert scoring models are built using expert experience when data is insufficient or nonexistent. Data-driven scoring models, on the other hand, are built using big data, machine learning, and deep learning technologies, provided sufficient data has been accumulated. Both expert and data-driven application scoring models can, to varying degrees, misjudge applicants, leading to the rejection of applications from customers with good future repayment records, thus impacting the bank's overall revenue and profits.

[0003] Rejection deduction methods can be used to win back rejected customers and uncover potential customers. Common rejection deduction methods fall into three categories: First, assign a "bad customer" status to all rejected samples, then merge the successful and rejected samples to build an application scoring card model. This new model then replaces or supplements the online application scoring card model. Second, randomly assign "good" and "bad" status to rejected samples, then merge the successful and rejected samples to build an application scoring card model. This new model then replaces or supplements the online application scoring card model. Third, lower the application scoring threshold in the online model, classifying rejected samples above or equal to the threshold as "good" and those below as "bad." This process is repeated, merging the successful and rejected samples to build an application scoring card model, which is then used to replace or supplement the online application scoring card model.

[0004] While assigning a "bad customer" status to all rejected samples is simple, it only works when the initial approval rate is high, such as 90% of applications being approved, and when the accuracy of the initial scoring plan has a high confidence level. Random assignment, being completely random, will affect the model's final accuracy and stability. Lowering the online model's application scoring threshold is difficult to control; a poorly defined threshold will also impact the model's final accuracy and stability. Summary of the Invention

[0005] In view of the above problems, this disclosure provides customer recapture methods, apparatus, devices, media and program products based on ensemble learning.

[0006] According to a first aspect of this disclosure, a customer reactivation method based on ensemble learning is provided, comprising: scoring a sample set of loan applicants based on a customer sample application scoring card model and a rejected customer sample application scoring card model; constructing a customer reactivation model based on the scores and a first good / bad sample identifier of the loan applicant sample set; scoring the sample set of loan applicants based on a preset online application scoring card model and the customer reactivation model; binning the scores of the online application scoring card model and the customer reactivation model to obtain a lower limit value for each bin; re-scoring customers rejected by the online application scoring card model using the customer reactivation model; and reactivating customers whose re-scored scores are greater than the lower limit value of the corresponding bin.

[0007] According to embodiments of this disclosure, the step of scoring the loan application customer sample set based on the customer sample application scoring card model and the rejected customer sample application scoring card model, and constructing the customer recovery model based on the scores and the first good / bad sample identifier of the loan application customer sample set includes: scoring the loan application customer sample set based on the customer sample application scoring card model and the rejected customer sample application scoring card model; standardizing the scores of the loan application customer sample set based on the customer sample application scoring card model and the rejected customer sample application scoring card model; and constructing the customer recovery model by using the standardized scores based on the customer sample application scoring card model and the rejected customer sample application scoring card model as input and the first good / bad sample identifier as output.

[0008] According to embodiments of this disclosure, the method includes: extracting feature values ​​from multiple dimensions of the approved customer sample set as input, taking a second good / bad sample identifier from the approved customer sample set as output, and constructing an application scoring card model for approved customer samples, wherein the second good / bad sample identifier is determined based on the post-loan performance of approved customer samples; and extracting feature values ​​from multiple dimensions of the rejected customer sample set as input, taking a third good / bad sample identifier from the rejected customer sample set as output, and constructing an application scoring card model for rejected customer samples, wherein the third good / bad sample identifier is determined based on other performance characteristics of rejected customer samples.

[0009] According to embodiments of this disclosure, determining the third good / bad sample identifier includes: determining the third good / bad sample identifier based on the post-loan performance of other loan products of the rejected customer sample.

[0010] According to embodiments of this disclosure, determining the third good / bad sample identifier includes: vectorizing the feature values ​​of each of the approved customer samples and each of the rejected customer samples; calculating the first similarity between the feature vector of each rejected customer sample and the approved customer samples with good post-loan performance; calculating the second similarity between the feature vector of each rejected customer sample and the approved customer samples with poor post-loan performance; extracting and comparing the maximum value of the first similarity and the second similarity; and determining the third good / bad sample identifier based on the comparison result.

[0011] According to embodiments of this disclosure, the step of scoring the loan application customer sample set based on the online application scoring card model and the customer return model, and binning the score sets of the online application scoring card model and the customer return model to obtain the lower limit value of the score for each bin includes: scoring the loan application customer sample set based on the online application scoring card model and the customer return model and performing standardization processing; binning the standardized online application scoring card model and the customer return model according to a preset step size, and calculating the proportion of bad customers in each bin and the cumulative proportion of bad customers in each bin; standardizing the bin number, using the standardized bin number as input and the proportion of bad customers in the bin as output to construct a univariate quadratic regression model; solving the univariate quadratic regression model to obtain the solution for the bin number; and calculating the lower limit value of the score for the bin based on the solution for the bin number and the mapping relationship between the bin number and the lower limit value of the bin.

[0012] According to embodiments of this disclosure, calculating the lower limit of the score for a bin based on the solution of the sequence number and the mapping relationship between the sequence number and the lower limit of the score for the bin includes: when the solution of the sequence number is greater than 0 and less than 1, the sequence number is reversed according to the standardization process, and the lower limit of the score for the bin is obtained according to the mapping relationship between the sequence number and the lower limit of the score for the bin; when the solution of the sequence number is less than 0 or greater than 1, the bad debt rate of the online application scoring card model is compared with the cumulative proportion of bad customers in the corresponding sequence number bin; when the bad debt rate is less than the cumulative proportion of bad customers in the bin, and the bad debt rate is greater than the cumulative proportion of bad customers in the bin with a sequence number less than 1, the sequence number is reversed according to the standardization process, and the lower limit of the score for the bin is obtained according to the mapping relationship between the sequence number and the lower limit of the score for the bin.

[0013] According to embodiments of this disclosure, the method further includes: reversibly restoring the lower limit of the score to the original score using a standardization process.

[0014] A second aspect of this disclosure provides a customer reactivation device based on ensemble learning, comprising: a reactivation model construction module, used to score a sample set of loan applicants based on a customer sample application scoring card model and a rejected customer sample application scoring card model, and to construct a customer reactivation model based on the scores and a first good / bad sample identifier of the loan applicant sample set; a binning processing module, used to score the sample set of loan applicants based on a preset online application scoring card model and the customer reactivation model, and to bin the score sets of the online application scoring card model and the customer reactivation model to obtain a lower limit value for each bin; a reactivation scoring module, used to re-score customers rejected by the online application scoring card model through the customer reactivation model; and a reactivation approval module, used to reactivate customers whose re-scores are greater than the lower limit value of the corresponding bin.

[0015] A third aspect of this disclosure provides an electronic device comprising: one or more processors; and a memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors perform the methods described above.

[0016] A fourth aspect of this disclosure also provides a computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, cause the processor to perform the methods described above.

[0017] The fifth aspect of this disclosure also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0018] According to the customer reactivation method, apparatus, equipment, medium and program products based on ensemble learning provided in this disclosure, the customer reactivation model is combined with the online application scoring card model, and customers rejected by the online application scoring card model are reactivated through the customer reactivation model, thereby reducing misjudgment of potential customers. Attached Figure Description

[0019] The foregoing contents, as well as other objects, features, and advantages of this disclosure, will become clearer from the following description of embodiments with reference to the accompanying drawings, in which:

[0020] Figure 1 The illustrations depict application scenarios of customer reactivation methods, apparatuses, devices, media, and program products based on ensemble learning according to embodiments of the present disclosure.

[0021] Figure 2 A flowchart illustrating a customer reactivation method based on ensemble learning according to an embodiment of the present disclosure is shown schematically.

[0022] Figure 3A schematic diagram illustrating the structure of a customer reactivation device based on ensemble learning according to an embodiment of the present disclosure is shown; and

[0023] Figure 4 A block diagram schematically illustrates an electronic device suitable for implementing a customer reactivation method based on ensemble learning, according to an embodiment of the present disclosure. Detailed Implementation

[0024] The embodiments of the present disclosure will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the disclosure. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of the present disclosure for ease of explanation. However, it will be apparent that one or more embodiments may be practiced without these specific details. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concepts of the present disclosure.

[0025] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit this disclosure. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0026] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0027] When using expressions such as "at least one of A, B, and C", they should generally be interpreted in accordance with the meaning that is commonly understood by a person skilled in the art (e.g., "a system having at least one of A, B, and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B, and C, etc.).

[0028] In the technical solutions disclosed herein, the collection, storage, use, processing, transmission, provision, disclosure, and application of data (including but not limited to user personal information) comply with the provisions of relevant laws and regulations, necessary confidentiality measures have been taken, and they do not violate public order and good morals.

[0029] The embodiments of this disclosure provide a customer reactivation method based on ensemble learning. A customer reactivation model is constructed, representing the mapping relationship between the first good / bad sample identifiers and scores of loan applicants. A sample set of loan applicants is scored based on a preset online application scoring card model and the customer reactivation model, respectively. The score sets of the online application scoring card model and the customer reactivation model are binned to obtain a lower limit value for each bin. Customers rejected by the online application scoring card model are re-scored through the customer reactivation model. Customers whose re-scored scores exceed the lower limit value of the corresponding bin are reactivated.

[0030] Figure 1 The illustration shows an application scenario diagram of the customer reactivation method based on ensemble learning according to an embodiment of the present disclosure.

[0031] like Figure 1 As shown, application scenario 100 according to this embodiment may include a bank's loan business. Network 104 serves as a medium for providing a communication link between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. Network 104 may include various connection types, such as wired or wireless communication links or fiber optic cables, etc.

[0032] Users can interact with server 105 via network 104 using at least one of the first terminal device 101, second terminal device 102, and third terminal device 103 to receive or send messages, etc. Various communication client applications can be installed on the first terminal device 101, second terminal device 102, and third terminal device 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social media platform software, etc. (for example only).

[0033] The first terminal device 101, the second terminal device 102, and the third terminal device 103 can be various electronic devices with displays and support web browsing, including but not limited to smartphones, tablets, laptops, and desktop computers.

[0034] Server 105 can be a server that provides various services, such as a backend management server that supports websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103 (this is just an example). The backend management server can analyze and process data such as received user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal devices.

[0035] It should be noted that the customer reactivation method based on ensemble learning provided in this disclosure embodiment can generally be executed by server 105. Correspondingly, the customer reactivation device based on ensemble learning provided in this disclosure embodiment can generally be located in server 105. The customer reactivation method based on ensemble learning provided in this disclosure embodiment can also be executed by a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105. Correspondingly, the customer reactivation device based on ensemble learning provided in this disclosure embodiment can also be located in a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105.

[0036] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0037] The following will be based on Figure 1 The described scene, through Figure 2 The customer reactivation method based on ensemble learning, as described in the disclosed embodiments, will be described in detail.

[0038] Figure 2 A flowchart illustrating a customer reactivation method based on ensemble learning according to an embodiment of the present disclosure is shown.

[0039] like Figure 2 As shown, the customer reactivation method based on ensemble learning in this embodiment includes operations S210 to S240.

[0040] In operation S210, the customer application customer sample set is scored based on the customer sample application scoring card model and the rejected customer application scoring card model, respectively. Based on the scores and the first good and bad sample identifier of the loan application customer sample set, a customer recovery model is constructed.

[0041] The construction of the customer retention model includes:

[0042] Operation S211: The customer sample application scoring card model and the rejected customer sample application scoring card model are used to score the loan application customer sample set respectively. The customer sample application scoring card model represents the mapping relationship between the customer sample set and the second good and bad sample identifier, and the rejected customer sample application scoring card model represents the mapping relationship between the rejected customer sample set and the third good and bad sample identifier.

[0043] In this embodiment, feature values ​​from multiple dimensions of the customer sample set are extracted as input, and the second good / bad sample identifier of the customer sample set is used as output to construct a customer sample application scoring card model. The second good / bad sample identifier is determined based on the post-loan performance of the approved customer samples. As shown in Table 1, Table 1 schematically illustrates the feature values ​​and second good / bad sample identifiers of the customer sample set {PS} whose loan applications have been approved.

[0044] Table 1

[0045]

[0046] Among them, the feature variable is... j This indicates that the customer sample was obtained through ps i The corresponding feature dimension, ps_x i，j This indicates that the customer sample was obtained through ps i Corresponding feature variable j eigenvalues, ps_y j To label the second good / bad sample based on post-loan performance, for example, by using post-loan status such as whether the loan is overdue or in default, if the customer is overdue or in default, then ps_y j =1, otherwise ps_y j =0.

[0047] Select the "Feature Variable" from Table 1 j "Good or bad sample label" is used as the candidate independent variable and "good or bad sample label" is used as the dependent variable. A customer sample application scorecard model ps_model is constructed using common supervised classification algorithms (such as logistic regression, random forest, XGBOOST, etc.).

[0048] In this embodiment, feature values ​​from multiple dimensions of the rejected customer sample set are extracted as input, and the third good / bad sample identifier of the rejected customer sample set is used as output to construct a rejection customer sample application scoring card model. The third good / bad sample identifier is determined based on other performance characteristics of the rejected customer samples. As shown in Table 2, Table 2 schematically illustrates the feature values ​​and the second good / bad sample identifier of the rejected customer sample set {RS}.

[0049] Table 2

[0050]

[0051]

[0052] Among them, the feature variable is... j Indicates rejection of customer sample rs i The corresponding feature dimension, rs_x i，j Indicates rejection of customer sample rsi Corresponding feature variable j eigenvalues, rs_y j The third good / bad sample identifier is rs_y, because rejected customers are those whose loan applications were rejected, and therefore there is no post-loan performance data. j This can be determined based on the post-loan performance of other loan products from the sample of rejected customers.

[0053] Specifically, the third good / bad sample identifier can be determined in the following way:

[0054] First, determine whether the rejected customer has other loan products within the bank. If so, label the customer as a good or bad sample based on the post-loan performance of the other loan products.

[0055] Secondly, if the rejected customer does not have other loan products within the bank, then credit data is used to determine whether the customer has loan products at other banks. If so, the customer is labeled as a good or bad sample based on the post-loan performance of the loan products at other banks.

[0056] If the rejected customer has neither other loan products within the bank nor loan products from other banks, the processing logic is as follows.

[0057] Vectorize the feature values ​​of each passed customer sample and each rejected customer sample. Utilize the feature variable. j The corresponding values ​​are represented in vector form as follows:

[0058] vec-ps i =<ps_x i，1 ps_x i，2 ,.....,ps_x i，j ,.....,ps_x i，m >

[0059] vec_rs i =<rs_x i，1 rs_x i，2 , ......, rs_x i，j， ..., rs_x i，m >

[0060] Among them, vec_ps i To pass customer sample ps i The corresponding vector, vec_rs i To reject customer samples rs i The corresponding vector.

[0061] Calculate the first similarity between the feature vector of each rejected customer sample and the approved customer sample with good post-loan performance; calculate the second similarity between the feature vector of each rejected customer sample and the approved customer sample with poor post-loan performance. The first and second similarities for each rejected customer sample are as follows: Figure 3 As shown.

[0062] Table 3

[0063]

[0064]

[0065] Where {GPS} represents the set of good customers whose loan applications have been approved, and {BPS} represents the set of bad customers whose loan applications have been approved, and {PS} = {GPS} ∪ {BPS}. Calculate the rejection samples rs using the sample vectorization results i The cosine distance between the sample and all good customer samples in {GPS} is used as the similarity between the samples, and the maximum value of the cosine distance is taken as max_gps. i Similarly, the rejection samples rs are calculated using the sample vectorization results. i The cosine distance between the sample and all samples passing through bad customer samples in {BPS} is used as the similarity between the samples, and the maximum value of the cosine distance is taken as max_bps. i .

[0066] Extract and compare the first similarity max_gps i Second similarity max_bps i The maximum value of the first similarity (max_gps) is used to determine the third good / bad sample identifier based on the comparison results. Specifically, when the first similarity (max_gps) is reached... i The maximum value is greater than the second similarity max_bps i When the value is at its maximum, the third good / bad sample is marked as "good"; otherwise, the third good / bad sample is marked as "bad".

[0067] The feature variable in Table 2 j As candidate independent variables, the third good and bad sample label is used as the dependent variable. A scorecard model for rejecting customer sample applications, rs_model, is constructed using commonly used supervised classification algorithms (such as logistic regression, random forest, XGBOOST, etc.).

[0068] Operation S212 constructs a customer retention model based on the first good and bad sample identifiers of the customer score and loan application sample set.

[0069] Specifically, operation S212 includes S2121 to S2123.

[0070] Operation S2121: Score the loan application customer sample set based on the customer sample application scoring card model and the rejected customer sample application scoring card model, respectively.

[0071] Operation S2122 standardizes the scores of the loan application customer sample set based on the customer sample application scoring card model and the rejected customer sample application scoring card model.

[0072] Operation S2123 takes the standardized scores from the customer sample application scoring card model and the rejected customer sample application scoring card model as input, and the first good / bad sample identifier as output to construct a customer return model.

[0073] The customer sample application scoring card model ps_model and the rejected customer sample application scoring card model rs_model were used to score the loan application customer sample set {AS}, and the results are shown in Table 4.

[0074] Table 4

[0075]

[0076] Among them, as i For samples in set {AS}, ps_score i For sample as i The score calculated using the sub-model ps_mode, rs_score i For sample as i The score calculated using the sub-model rs_model, as_y i Let {AS} be the first good / bad sample identifier for the customer sample, {PS} be the set of customer samples whose loan applications have been approved, and {RS} be the set of customer samples whose loan applications have been rejected. Let {AS} = {PS} ∪ {RS} and {PS} ∪ {RS} ∪ ...

[0077] For ps_score respectively i and rs_score i，1 Standardization is performed, where ps_score i The standardization processing logic is as follows:

[0078] Calculate the maximum and minimum values ​​of ps_score, and denote them as max_ps_score and mm_ps_score, respectively;

[0079] Standardized ps_score i Let it be std_ps_score i The calculation formula is as follows:

[0080] std_ps_score i =(ps_score i -min_ps_score) / (max_ps_score-min_ps_score).

[0081] Similarly, rs_score i The standardization processing logic is as follows:

[0082] Calculate the maximum and minimum values ​​of rs_score, and denote them as max_rs_score and min_rs_score, respectively;

[0083] Standardized rs_score i Let it be std_rs_score i The calculation formula is as follows:

[0084] std_rs_score i =(rs_score i -min_rs_score) / (max_rs_score-min_rs_score).

[0085] Replacing the original scores with the standardized scores yields Table 5, as shown below:

[0086] Table 5

[0087]

[0088] Based on Table 5 above, using the standardized scores std_ps_score and std_rs_score calculated by the sub-model ps_model as independent variables, and the good / bad sample identifier as_y as the dependent scalar, a customer recovery model lh_model is constructed using the logistic regression algorithm, as shown in the following formula:

[0089] lh_model=w1*ps_model+w2*rs_model

[0090] Where w1 and w2 are the weight coefficients of the sub-models trained using the logistic regression algorithm.

[0091] Furthermore, the customer reactivation model is combined with the pre-set online application scoring card model, and customers rejected by the online application scoring card model are reactivated through the customer reactivation model.

[0092] In operation S220, the loan application customer sample set is scored based on the online application scoring card model and the customer return model respectively. The scoring sets of the online application scoring card model and the customer return model are binned to obtain the lower limit value of the score for each bin.

[0093] Operation S220 includes S221 to S225.

[0094] In operation S221, the loan application customer sample set is scored and standardized based on the online application scoring card model and the customer return model, as shown in Table 6.

[0095] Table 6

[0096]

[0097]

[0098] Among them, std_online_score i =(online_score i -min_online_score) / (max_online_score-min_online_score)

[0099] `max_online_score` represents the maximum value of the original score `online_score` of the online application scorecard model, and `min_online_score` represents the minimum value of the original score `online_score` of the online application scorecard model, where 0 <= `std_online_score`. i <=1.

[0100] std_lh_score i =(lh_score i -min_lh_score) / (max_lh_score-min_lh_score)

[0101] `max_lh_score` represents the maximum value of the original score `lh_score` of the customer retrieval model `lh_model`, and `min_lh_score` represents the minimum value of the original score `lh_score` of the customer retrieval model `lh_model`, where 0 <= `std_lh_score`. i <=1.

[0102] In operation S222, the standardized online application scoring card model and customer return model are binned according to a preset step size, and the percentage of bad customers in each bin and the cumulative percentage of bad customers in each bin are calculated.

[0103] Assuming the bins are divided into k bins (k is an adjustable parameter), the step size is 1 / k. An example of binning is shown below:

[0104] Table 7

[0105]

[0106]

[0107] Among them, bad_per i = (Number of bad customers in bin number i of the customer recovery model + Number of bad customers in bin number i of the online application scoring card model) / (Total number of customers in bin number i of the customer recovery model + Total number of customers in bin number i of the online application scoring card model).

[0108]

[0109] In operation S223, the serial number of the bin is standardized, and the standardized serial number is used as input, while the proportion of bad customers in the bin is used as output to construct a univariate quadratic regression model.

[0110] Standardize the bin sequence number i, and denote the standardized sequence number as j. The standardization logic is as follows:

[0111] j = (i-1) / (k-1), where k is the number of bins, 0 = <j<=1;

[0112] Using j as the independent variable, the percentage of bad customers in the sorting table (bad_per) is shown in Table 7. i As the dependent variable, a univariate quadratic regression model is constructed as follows:

[0113] bad_per=a*j 2 +b*j+c, where a, b, and c are the weight coefficients of the trained model.

[0114] In operation S224, the univariate quadratic regression model is solved to obtain the solution for the bin number.

[0115] for bad_per = a*j 2 Taking the derivative of +b*j+c, bad_per'=2*a*j+b. Let bad_per'=0, then 2*a*j+b=0, and we can get the solution j=-b / (2*a) for the index.

[0116] In operation S225, the lower limit of the bin score is calculated based on the solution of the sequence number and the mapping relationship between the sequence number and the lower limit of the bin.

[0117] If the solution for the sequence number satisfies -b / (2*a)>=0 and -b / (2*a)<=1, since the sequence number j is the standardized bin number, j=(i-1) / (k-1), by reversing the standardization method to restore the original bin number i, we can get i=j*(k-1)+1=(-b / (2*a))*(k-1) / +1.

[0118] Referring to the correspondence between the sorting box number and the lower limit of the sorting box interval in Table 7, the lower limit of the sorting box interval corresponding to sorting box number i (ni) / k = (n-(-b / (2*a))*(k-1) / -1) / k is taken as the passing score threshold of the customer retrieval model.

[0119] Since the binning intervals in Table 7 are standardized scores, it is also necessary to reverse the standardization process to restore the original scores to their lower limits, according to the standardization formula:

[0120] std_lh_score i =(lh_score i -min_lh_score) / (max_lh_score-min_lh_score)

[0121] available:

[0122] lh_score i =std_lh_score i *(max_lh_score-min_lh_score)+min_lh_score

[0123] At this time, std_lh_score i The value is (n-(-b / (2*a))*(k-1)-1) / k.

[0124] Customers rejected by the online application scoring model (online_model) are retrieved using the customer retrieval model (lh_model). If a customer is rejected by the online_model but scores calculated by the lh_model are greater than or equal to the aforementioned lh_score, then the retrieval process is initiated. i If so, the customer can get the approval back.

[0125] If the solution b / (2*a) < 0 or -b / (2*a) > 1 for the serial number, then take the bad debt rate obp corresponding to the current online application scoring card model online_model. If the cumulative percentage of bad customers in the bin of serial number i satisfies the condition obp <= sum_bad_per i And obp > sum_bad_per i-1Then, the lower limit (ni) / k of the bin interval corresponding to bin number i is taken as the pass score threshold of the customer retrieval model.

[0126] Since the binning intervals in Table 7 are standardized scores, it is necessary to restore them to the original scores, according to the standardization formula:

[0127] std_lh_score i =(lh_score i -min_lh_score) / (max_lh_score-min_lh_score)

[0128] available:

[0129] lh_score i =std_lh_score i *(max_lh_score-min_lh_score)+min_lh_score

[0130] At this time, std_lh_score i The value is (ni) / k.

[0131] In operation S230, customers who were rejected by the online application scorecard model (online_model) will be re-scored by the customer recall model (lh_model).

[0132] In operation S240, the re-scoring is compared with the corresponding bin lower limit value. That is, the re-scoring is compared with the bin lower limit value of the bin to which the online application scorecard model (online_model) belongs. When the customer return model (lh_model) scores the customer higher than the corresponding bin lower limit value (lh_score), the re-scoring is considered successful. i If the score is higher than the corresponding lower limit of the sub-box, the customer will be retrieved and approved again.

[0133] The customer reactivation method based on ensemble learning provided in this disclosure is applicable not only to situations with a high initial approval rate but also to situations with a low approval rate; the model has high accuracy and stability; and the model accuracy is not affected by expert subjective experience or threshold interference.

[0134] Based on the above-described customer reactivation method based on ensemble learning, this disclosure also provides a customer reactivation device based on ensemble learning. The following will be combined with... Figure 3 The device is described in detail.

[0135] Figure 3 A schematic block diagram of a customer reactivation device based on ensemble learning according to an embodiment of the present disclosure is shown.

[0136] like Figure 3 As shown, the customer retrieval device 300 based on ensemble learning in this embodiment includes a retrieval model construction module 310, a sorting and processing module 320, a retrieval scoring module 330, and a retrieval approval module 340.

[0137] The customer return model construction module 310 is used to construct a customer return model, which represents the mapping relationship between the first good / bad sample identifiers and scores of loan applicants. In one embodiment, the return model construction module 310 can be used to perform the operation S210 described above, which will not be repeated here.

[0138] The construction of the customer retention model includes: scoring the loan application customer sample set based on the customer sample application scoring card model and the rejected customer sample application scoring card model respectively; the customer sample application scoring card model represents the mapping relationship between the customer sample set and the second good and bad sample identifiers; the rejected customer sample application scoring card model represents the mapping relationship between the rejected customer sample set and the third good and bad sample identifiers; and the customer retention model is constructed based on the scores and the first good and bad sample identifiers of the loan application customer sample set.

[0139] Binning module 320 is used to score the loan application customer sample set based on a preset online application scoring card model and a customer return model, and to bin the scoring sets of the online application scoring card model and the customer return model to obtain the lower limit value of the score for each bin. In one embodiment, binning module 320 can be used to perform the operation S220 described above, which will not be repeated here.

[0140] The customer recall scoring module 330 is used to re-score customers rejected by the online application scoring card model through the customer recall model. In one embodiment, the customer recall scoring module 330 can be used to perform the operation S230 described above, which will not be repeated here.

[0141] The customer retrieval approval module 340 is used to approve customer retrieval when the customer retrieval model's score for the customer is greater than the lower limit of the score for the corresponding sub-bin. In one embodiment, the customer retrieval approval module 340 can be used to execute the operation S240 described above, which will not be repeated here.

[0142] According to embodiments of this disclosure, any plurality of modules among the repatriation model construction module 310, bin sorting processing module 320, repatriation scoring module 330, and repatriation approval module 340 can be combined into one module, or any one of these modules can be split into multiple modules. Alternatively, at least part of the functionality of one or more of these modules can be combined with at least part of the functionality of other modules and implemented in one module. According to embodiments of this disclosure, at least one of the repatriation model construction module 310, bin sorting processing module 320, repatriation scoring module 330, and repatriation approval module 340 can be at least partially implemented as hardware circuitry, such as a field-programmable gate array (FPGA), a programmable logic array (PLA), a system-on-a-chip, a system-on-a-substrate, a system-on-package, an application-specific integrated circuit (ASIC), or implemented in hardware or firmware by any other reasonable means of integrating or packaging the circuitry, or implemented in software, hardware, or firmware, or in any suitable combination of any of these three implementation methods. Alternatively, at least one of the repatriation model construction module 310, the bin sorting processing module 320, the repatriation scoring module 330, and the repatriation approval module 340 can be at least partially implemented as a computer program module, which can perform corresponding functions when the computer program module is run.

[0143] Figure 4 A block diagram schematically illustrates an electronic device suitable for implementing a customer reactivation method based on ensemble learning, according to an embodiment of the present disclosure.

[0144] like Figure 4 As shown, an electronic device 400 according to an embodiment of the present disclosure includes a processor 401, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 402 or a program loaded from a storage portion 408 into a random access memory (RAM) 403. The processor 401 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 401 may also include onboard memory for caching purposes. The processor 401 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present disclosure.

[0145] RAM 403 stores various programs and data required for the operation of electronic device 400. Processor 401, ROM 402, and RAM 403 are interconnected via bus 404. Processor 401 performs various operations of the method flow according to embodiments of the present disclosure by executing programs in ROM 402 and / or RAM 403. It should be noted that the programs may also be stored in one or more memories other than ROM 402 and RAM 403. Processor 401 may also perform various operations of the method flow according to embodiments of the present disclosure by executing programs stored in said one or more memories.

[0146] According to embodiments of this disclosure, the electronic device 400 may further include an input / output (I / O) interface 405, which is also connected to a bus 404. The electronic device 400 may also include one or more of the following components connected to the I / O interface 405: an input section 406 including a keyboard, mouse, etc.; an output section 407 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 408 including a hard disk, etc.; and a communication section 409 including a network interface card such as a LAN card, modem, etc. The communication section 409 performs communication processing via a network such as the Internet. A drive 410 is also connected to the I / O interface 405 as needed. A removable medium 411, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 410 as needed so that computer programs read from it can be installed into the storage section 408 as needed.

[0147] This disclosure also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs that, when executed, implement the method according to the embodiments of this disclosure.

[0148] According to embodiments of this disclosure, the computer-readable storage medium may be a non-volatile computer-readable storage medium, such as, but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this disclosure, the computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to embodiments of this disclosure, the computer-readable storage medium may include ROM 402 and / or RAM 403 and / or one or more memories other than ROM 402 and RAM 403 described above.

[0149] Embodiments of this disclosure also include a computer program product comprising a computer program containing program code for performing the methods shown in the flowchart. When the computer program product is run on a computer system, the program code enables the computer system to implement the customer reactivation method based on ensemble learning provided in embodiments of this disclosure.

[0150] When the computer program is executed by the processor 401, it performs the functions defined in the system / apparatus of this disclosure embodiments. According to embodiments of this disclosure, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0151] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and downloaded and installed via communication section 409, and / or installed from removable medium 411. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0152] In such an embodiment, the computer program can be downloaded and installed from a network via communication section 409, and / or installed from removable medium 411. When the computer program is executed by processor 401, it performs the functions defined in the system of this disclosure embodiment. According to embodiments of this disclosure, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0153] According to embodiments of this disclosure, program code for executing the computer programs provided in embodiments of this disclosure can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C", or similar programming languages. The program code can execute entirely on the user's computing device, partially on the user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0154] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0155] Those skilled in the art will understand that the features described in the various embodiments and / or claims of this disclosure can be combined or combined in various ways, even if such combinations or combinations are not explicitly described in this disclosure. In particular, the features described in the various embodiments and / or claims of this disclosure can be combined or combined in various ways without departing from the spirit and teachings of this disclosure. All such combinations and / or combinations fall within the scope of this disclosure.

[0156] The embodiments of this disclosure have been described above. However, these embodiments are for illustrative purposes only and are not intended to limit the scope of this disclosure. Although various embodiments have been described above, this does not mean that the measures in the various embodiments cannot be used advantageously in combination. The scope of this disclosure is defined by the appended claims and their equivalents. Various substitutions and modifications can be made by those skilled in the art without departing from the scope of this disclosure, and all such substitutions and modifications should fall within the scope of this disclosure.

Claims

1. A customer reactivation method based on ensemble learning, characterized in that, include: A customer recovery model is constructed based on the customer application scoring card model and the rejected customer application scoring card model, which score the loan application customer sample set respectively. The loan application customer sample set is scored based on the preset online application scoring card model and the customer return model respectively. The scoring sets of the online application scoring card model and the customer return model are binned to obtain the lower limit value of the score for each bin. Customers rejected by the online application scoring card model will be re-scored using the customer reactivation model. Customers whose re-rating is higher than the lower limit of the corresponding sub-box will be recalled; The online application scoring card model and the customer return model, respectively, score the sample set of loan applicants. The score sets from both models are binned to obtain the lower limit value for each bin, including: The loan application customer sample set was scored and standardized based on the online application scoring card model and the customer return model respectively; The standardized online application scoring card model and the customer return model are divided into bins according to a preset step size, and the percentage of bad customers in each bin and the cumulative percentage of bad customers in each bin are calculated. The serial numbers of the bins are standardized, and the standardized serial numbers are used as inputs, while the proportion of bad customers in each bin is used as outputs to construct a univariate quadratic regression model. The solution for the bin number is obtained by differentiating the univariate quadratic regression model. The lower limit of the score for the corresponding bin is calculated based on the solution of the sequence number and the mapping relationship between the sequence number and the lower limit of the score for the corresponding bin, including: When the solution of the sequence number is greater than 0 and less than 1, the sequence number is reversed according to the standardization process, and the lower limit of the score of the corresponding bin is obtained according to the mapping relationship between the sequence number and the lower limit of the score of the corresponding bin. When the solution of the sequence number is less than 0 or greater than 1, compare the bad debt rate of the online application scoring card model with the cumulative proportion of bad customers in the corresponding sequence number bin. When the bad debt rate is less than the cumulative percentage of bad customers in the sub-box, and the bad debt rate is greater than the cumulative percentage of bad customers in the sub-box with a serial number less than 1, the serial number is reversed according to the standardization process, and the lower limit of the rating of the corresponding sub-box is obtained according to the mapping relationship between the serial number and the lower limit of the rating of the corresponding sub-box.

2. The method according to claim 1, characterized in that, The step of scoring the loan application customer sample set using a customer sample application scoring card model and a rejected customer sample application scoring card model, and constructing the customer recovery model based on the scores and the first good / bad sample identifier of the loan application customer sample set includes: The loan application customer sample set is scored based on the customer sample application scoring card model and the rejected customer application scoring card model, respectively; the scores of the loan application customer sample set based on the customer sample application scoring card model and the rejected customer application scoring card model are then standardized. The customer retention model is constructed by taking the standardized scores based on the customer sample application scorecard model and the rejected customer sample application scorecard model as input and the first good and bad sample identifier as output.

3. The method according to claim 1, characterized in that, include: The feature values ​​of multiple dimensions of the customer sample set are extracted as input, and the second good and bad sample identifier of the customer sample set is used as output to construct a customer sample application scoring card model. The second good and bad sample identifier is determined based on the post-loan performance of the customer sample. The feature values ​​of multiple dimensions of the rejected customer sample set are extracted as input, and the third good and bad sample identifier of the rejected customer sample set is used as output to construct a rejected customer sample application scoring card model. The third good and bad sample identifier is determined based on other performance of the rejected customer samples.

4. The method according to claim 3, characterized in that, Determining the third good / bad sample identifier includes: The third good / bad sample identifier is determined based on the post-loan performance of other loan products from the rejected customer sample.

5. The method according to claim 3, characterized in that, Determining the third good / bad sample identifier includes: Vectorize the feature values ​​of each passed customer sample and each rejected customer sample, calculate the first similarity between the feature vector of each rejected customer sample and the passed customer sample with good post-loan performance, and calculate the second similarity between the feature vector of each rejected customer sample and the passed customer sample with poor post-loan performance; Extract and compare the maximum value of the first similarity and the second similarity, and determine the third good / bad sample identifier based on the comparison result.

6. The method according to claim 1, characterized in that, The method further includes: The lower limit of the score is reversed and restored to the original score using a standardization process.

7. A customer reactivation device based on ensemble learning, characterized in that, include: The customer return model construction module is used to score the loan application customer sample set based on the customer sample application scoring card model and the rejected customer sample application scoring card model, and to construct a customer return model based on the scores and the first good and bad sample identifier of the loan application customer sample set. The binning module is used to score the sample set of loan applicants based on a preset online application scoring card model and a customer return model, and to bin the scoring sets of the online application scoring card model and the customer return model to obtain the lower limit value of the score for each bin. The customer recall scoring module is used to re-score customers who have been rejected by the online application scoring card model through the customer recall model. The recall approval module is used to recall customers whose re-score is greater than the lower limit of the corresponding sub-box. The online application scoring card model and the customer return model, respectively, score the sample set of loan applicants. The score sets from both models are binned to obtain the lower limit value for each bin, including: The loan application customer sample set was scored and standardized based on the online application scoring card model and the customer return model respectively; The standardized online application scoring card model and the customer return model are divided into bins according to a preset step size, and the percentage of bad customers in each bin and the cumulative percentage of bad customers in each bin are calculated. The serial numbers of the bins are standardized, and the standardized serial numbers are used as inputs, while the proportion of bad customers in each bin is used as outputs to construct a univariate quadratic regression model. The solution for the bin number is obtained by differentiating the univariate quadratic regression model. The lower limit of the score for the corresponding bin is calculated based on the solution of the sequence number and the mapping relationship between the sequence number and the lower limit of the score for the corresponding bin, including: When the solution of the sequence number is greater than 0 and less than 1, the sequence number is reversed according to the standardization process, and the lower limit of the score of the corresponding bin is obtained according to the mapping relationship between the sequence number and the lower limit of the score of the corresponding bin. When the solution of the sequence number is less than 0 or greater than 1, compare the bad debt rate of the online application scoring card model with the cumulative proportion of bad customers in the corresponding sequence number bin. When the bad debt rate is less than the cumulative percentage of bad customers in the sub-box, and the bad debt rate is greater than the cumulative percentage of bad customers in the sub-box with a serial number less than 1, the serial number is reversed according to the standardization process, and the lower limit of the rating of the corresponding sub-box is obtained according to the mapping relationship between the serial number and the lower limit of the rating of the corresponding sub-box.

8. An electronic device, characterized in that, include: One or more processors; Storage device for storing one or more programs. Wherein, when the one or more programs are executed by the one or more processors, the one or more processors perform the method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, It stores executable instructions that, when executed by a processor, cause the processor to perform the method according to any one of claims 1 to 6.

10. A computer program product, characterized in that, Includes a computer program, which, when executed by a processor, implements the method according to any one of claims 1 to 6.

Images