CN113298510B - Deduction instruction initiating method and device - Google Patents
Deduction instruction initiating method and device Download PDFInfo
- Publication number
- CN113298510B CN113298510B CN202110589601.0A CN202110589601A CN113298510B CN 113298510 B CN113298510 B CN 113298510B CN 202110589601 A CN202110589601 A CN 202110589601A CN 113298510 B CN113298510 B CN 113298510B
- Authority
- CN
- China
- Prior art keywords
- deduction
- order
- machine learning
- learning model
- information
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 89
- 230000000977 initiatory effect Effects 0.000 title claims abstract description 70
- 238000010801 machine learning Methods 0.000 claims abstract description 111
- 238000012549 training Methods 0.000 claims abstract description 26
- 238000012545 processing Methods 0.000 claims abstract description 12
- 239000013598 vector Substances 0.000 claims description 28
- 230000035945 sensitivity Effects 0.000 claims description 25
- 238000004458 analytical method Methods 0.000 claims description 16
- 238000004364 calculation method Methods 0.000 claims description 12
- 238000012360 testing method Methods 0.000 claims description 9
- 238000000546 chi-square test Methods 0.000 claims description 5
- 238000001604 Rao's score test Methods 0.000 claims description 4
- 238000013211 curve analysis Methods 0.000 claims description 2
- 238000004891 communication Methods 0.000 claims 1
- 238000007477 logistic regression Methods 0.000 description 63
- 230000006870 function Effects 0.000 description 12
- 238000010586 diagram Methods 0.000 description 9
- 238000004590 computer program Methods 0.000 description 8
- 230000000694 effects Effects 0.000 description 8
- 230000007547 defect Effects 0.000 description 2
- 238000007689 inspection Methods 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000000737 periodic effect Effects 0.000 description 1
- 230000001737 promoting effect Effects 0.000 description 1
- 238000012827 research and development Methods 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 238000012795 verification Methods 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
- G06Q20/00—Payment architectures, schemes or protocols
- G06Q20/08—Payment architectures
- G06Q20/085—Payment architectures involving remote charge determination or related payment systems
-
- 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
- G06Q20/00—Payment architectures, schemes or protocols
- G06Q20/38—Payment protocols; Details thereof
- G06Q20/42—Confirmation, e.g. check or permission by the legal debtor of payment
Landscapes
- Business, Economics & Management (AREA)
- Accounting & Taxation (AREA)
- Finance (AREA)
- Strategic Management (AREA)
- Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a deduction instruction initiating method and a training method of a machine learning model, wherein the deduction instruction initiating method comprises the following steps: obtaining order information of a deduction order; inputting the order information into a machine learning model for processing, and acquiring a predicted value of successful money deduction of a money deduction order; and if the predicted value is larger than or equal to the target threshold value, initiating a money deduction instruction of the money deduction order. Therefore, in the withholding service of the installment payment agreement signed between the merchant and the consumer, the probability that the withholding fails due to the problems that the balance of the withheld account of the consumer is insufficient, the limit of the withheld account is exceeded and the like and the situation that the bank still collects the commission fee can be reduced, thereby reducing the collection cost of the merchant.
Description
The patent application of the invention is the application date of 2018, 7 and 10 months and the application number of 201810750604.6
The invention relates to a divisional application of Chinese invention patent application called 'a deduction instruction initiating method and a deduction instruction initiating device'.
Technical Field
The invention relates to the technical field of computers, in particular to a deduction instruction initiating method and device.
Background
With the advanced innovation of the financial industry, the number of transaction services such as utility charging, collection of goods payment, and periodic repayment after consumption is rapidly increased, and collection of payment handling fees are charged for such services through collection of payment deducting channels such as banks, unions of bank, internet and the like, for example: the merchant collects the installment payment fee to the consumer through the bank, and the bank collects the fixed amount of the fee or collects a certain proportion of the fee according to the deduction amount once the deduction is carried out, however, in practical application, because the balance of the deducted account of the consumer is insufficient, the balance exceeds the limit of the deducted account, and other problems, the condition of the deduction failure often occurs, and the bank can not avoid collecting the fee because of the deduction failure, thereby promoting the collection cost of the merchant.
In the related art, a deduction strategy that the money amount is increased gradually is established for each order according to the return condition of a bank, the method is obviously not suitable for batch deduction business, because one batch of deduction comprises a large number of deduction orders, the deduction operation usually needs to be executed by consuming time from tens of minutes to hours or even across days, and the method in the related art increases the time of the deduction operation due to establishment of a large number of deduction strategies, so that the time cost of the deduction is greatly increased. Therefore, the deduction method in the related art has the defect of high cost.
Disclosure of Invention
The embodiment of the invention provides a deduction instruction initiating method and device, aiming at solving the defect of high cost of a money deduction substitute method in the related art.
In order to solve the technical problem, the invention is realized as follows:
in a first aspect, an embodiment of the present invention provides a deduction instruction initiating method, where the method includes:
acquiring order information of a deduction order;
after the order information is input into a machine learning model for processing, obtaining a predicted value of successful deduction of the deduction order;
and if the predicted value is greater than or equal to the target threshold value, initiating a deduction instruction of the deduction order.
In a second aspect, an embodiment of the present invention further provides a deduction instruction initiating device, where the device includes:
the first acquisition module is used for acquiring order information of a deduction order;
the first prediction module is used for inputting the order information into a machine learning model for processing to obtain a predicted value of successful deduction of the deduction order;
and the initiating module is used for initiating a deduction instruction of the deduction order if the predicted value is greater than or equal to a target threshold value.
In a third aspect, an embodiment of the present invention further provides a deduction instruction initiating device, including:
the deduction instruction initiating method comprises a memory, a processor and a computer program which is stored on the memory and can run on the processor, wherein the processor realizes the steps in the deduction instruction initiating method provided by the embodiment of the invention when executing the computer program.
In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium, where a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, the computer program implements the steps in the deduction instruction initiating method provided in the embodiment of the present invention.
In the embodiment of the invention, the order information of the deduction order is obtained; inputting the order information into a machine learning model for processing, and acquiring a predicted value of successful deduction of the deduction order; and if the predicted value is greater than or equal to the target threshold value, initiating a deduction instruction of the deduction order. Therefore, before the deduction instruction is initiated, whether the deduction order can be successfully deducted or not can be predicted through the machine learning model, and the deduction instruction is initiated under the condition that the predicted value is larger than the target threshold value, so that the probability of unsuccessful execution of the issued deduction instruction is reduced, and the effect of reducing the deduction cost is achieved.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the embodiments or the description of the prior art will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art that other drawings can be obtained according to the drawings without inventive labor.
Fig. 1 is a flowchart of an initiating method of a deduction instruction according to an embodiment of the present invention;
fig. 2 is a flowchart of another method for initiating a deduction instruction according to an embodiment of the present invention;
fig. 3 is a schematic diagram of another method for initiating a deduction instruction according to an embodiment of the present invention;
FIG. 4 is a diagram illustrating a method for determining parameters of a logistic regression model according to an embodiment of the present invention;
FIG. 5 is a schematic diagram of an ROC curve of a logistic regression model according to an embodiment of the present invention;
fig. 6 is a structural diagram of an initiating device of a deduction instruction according to an embodiment of the present invention;
fig. 7 is a structural diagram of another deduction instruction initiating device according to an embodiment of the present invention;
fig. 8 is a structural diagram of another deduction instruction initiating device according to an embodiment of the present invention;
fig. 9 is a structural diagram of another deduction instruction initiating device according to an embodiment of the present invention;
fig. 10 is a block diagram of another deduction instruction initiating 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 some, not all, embodiments of the present invention. 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, fig. 1 is a flowchart of a deduction instruction initiating method according to an embodiment of the present invention. As shown in fig. 1, the deduction instruction initiating method includes the following steps:
The deduction order may also be referred to as a withholding order, for example: an installment agreement is signed between the trade company and the consumer, the trade company can initiate a deduction instruction to the bank on the specified repayment date, the bank deducts the corresponding repayment amount from the bank account of the consumer according to the deduction instruction, and transfers the repayment amount to the bank account of the trade company, and during the period, the bank can charge a certain procedure fee to the trade company.
In addition, the order information may include information related to the deduction order, such as a deduction amount, a deduction date, an order number, a deduction channel, and a last deduction time.
By obtaining the order information of the deduction order in this step, an information basis can be provided for predicting the predicted value of the deduction order, which can be successfully deducted, by machine learning in step 102.
And 102, inputting the order information into a machine learning model for processing, and acquiring a predicted value of successful deduction of the deduction order.
The machine learning model may be any one of machine learning models such as a logistic regression model and a neural network model, and the machine learning model may be obtained through training or configured, and is not particularly limited. The training may be obtained by training order information of a historical deduction order to obtain the machine learning model, and specifically may be obtained by training in a research and development process of the deduction instruction initiating method. In addition, the machine learning model can predict the predicted value of the successful deduction of the deduction order according to the order information.
In addition, the predicted value may be a success probability that the deduction order can be successfully deducted, and the greater the predicted value, the higher the success probability that the deduction order can be successfully deducted is.
Of course, the predicted value may also be other values or characters used for determining that the deduction order can be successfully deducted, for example: and if the machine learning model outputs 0, the deduction result of the predicted deduction order is deduction failure, and if the machine learning model outputs 1, the deduction result of the predicted deduction order is deduction success.
Optionally, the order information may include numerical value information and/or text information, where for text information, a numerical value may be assumed for each possible occurrence of the text order information, and the numerical value is used as an input corresponding to the order information one by one, so as to input the machine learning model, so as to obtain a predicted value. For example: the order information includes a deduction channel, and when the deduction channel is a bank, the deduction channel is assumed to be 1; when the deduction channel is the internet bank, the deduction channel is assumed to be 2, and the predicted value when the deduction channel is the bank or the internet bank can be predicted by inputting the value 1 or 2 into the machine learning model.
Through the step, the machine learning model can be adopted to predict the predicted value of whether the deduction order can be successfully deducted, and a basis is provided for determining whether to send the deduction instruction in the step 103.
And 103, if the predicted value is greater than or equal to a target threshold value, initiating a deduction instruction of the deduction order.
The target threshold may be preset or obtained by training according to the order information, and is used to judge whether the predicted value of the deduction order indicates that the deduction order can be successfully deducted, when the predicted value is greater than or equal to the target threshold, the deduction order indicates that the deduction order can be successfully deducted, and when the predicted value is less than the target threshold, the deduction order indicates that the deduction order cannot be successfully deducted. And only when the predicted value is greater than or equal to a target threshold value, initiating a deduction instruction.
In addition, the target threshold may be set as needed, for example: the target threshold may be increased when sensitivity requirements are high, or may be set according to a command ratio, for example: the target threshold may be decreased when the instructions are lower than required. Wherein the order ratio is a ratio of a quantity of the deduction orders from which the deduction order was initiated to a total quantity of the deduction orders.
It should be noted that the deduction instruction may include a deduction time, a deduction amount, a deduction channel, a deduction account, and the like of the deduction order, so that a third party deduction channel can execute a corresponding deduction operation according to the deduction instruction after acquiring the deduction instruction.
Of course, the deduction instruction may not include the deduction time, the deduction amount, the deduction channel, the deduction account, and the like, and when the deduction instruction is initiated, the third party deduction channel may obtain information of the deduction time, the deduction amount, the deduction channel, the deduction account, and the like in other manners, for example: checking order numbers, looking up order databases, etc.
Through the steps, the deduction instruction is initiated only when the predicted value is larger than or equal to the target threshold value, and the deduction instruction is not initiated when the predicted value is smaller than the target threshold value, so that the deduction instruction is initiated for the deduction order under the condition that the machine learning model predicts that the deduction order cannot be successfully deducted, further, the deduction commission charge is still collected when the deduction execution failure of a third party deduction channel is avoided, and the effect of reducing the deduction cost is achieved.
The deduction instruction initiating method can be applied to withholding services, and a third-party withholding device or channel can execute deduction operation according to the deduction instruction initiated by the deduction instruction initiating method. Of course, in the embodiment of the present invention, the deduction instruction executed by the third party deduction channel is not limited, and in some cases, the deduction instruction may also be executed by an apparatus executing the method, that is, in some cases, the method may further include: and executing the deduction instruction to deduct money.
It should be noted that the deduction instruction initiating method provided by the embodiment of the present invention may be applied to intelligent devices such as a mobile phone, a tablet computer, a server, and the like.
In the embodiment of the invention, the order information of the deduction order is obtained; inputting the order information into a machine learning model for processing, and acquiring a predicted value of successful deduction of the deduction order; and if the predicted value is greater than or equal to the target threshold value, initiating a deduction instruction of the deduction order. Therefore, before the deduction instruction is initiated, whether the deduction order can be successfully deducted or not can be predicted through the machine learning model, and the deduction instruction is initiated under the condition that the predicted value is larger than the target threshold value, so that the probability of unsuccessful execution of the issued deduction instruction is reduced, and the effect of reducing the deduction cost is achieved.
Referring to fig. 2, it is a flowchart of another deduction instruction initiating method provided in an embodiment of the present invention, and as shown in fig. 2, the method includes the following steps:
As an alternative implementation, as shown in fig. 3, the order information includes: n items of order sub-information, n being an integer greater than or equal to 1, before step 202, the method further comprises:
judging whether the deduction order is suitable for the machine learning model or not according to the n items of order sub-information;
and if the deduction order is suitable for the machine learning model, executing the step of obtaining a predicted value of successful deduction of the deduction order after inputting the order information into the machine learning model for processing.
Wherein, the deduction order is applicable to the machine learning model, that is, the model in fig. 3 is applicable to the order, and the n items of order sub-information may include: at least one of a withholding product code, a withholding amount, a number of overdue days, a withholding batch, a loan transaction, and a withholding channel. Preferably, the n items of order sub-information are the deduction amount and the number of overdue days.
In addition, as shown in fig. 3, if the deduction order is not applicable to the machine learning model, that is, the model in fig. 3 is not applicable to the order, a deduction instruction may be directly initiated, or parameters in the machine learning model may be adjusted to make the deduction order applicable to the machine learning model after the parameters are adjusted, so as to obtain an accurate predicted value.
According to the method and the device, the deduction instruction can be initiated under the condition that the deduction order is not suitable for the machine learning model, the phenomenon that the deduction order which can be successfully deducted actually is not initiated due to the fact that the machine learning model is used for obtaining an incorrect prediction result is avoided, and therefore the coverage rate of the deduction instruction initiating method on the deduction order which can be successfully deducted is improved.
Optionally, the step of determining whether the deduction order is applicable to the machine learning model according to the n items of order sub-information may be implemented by the following specific steps:
generating order vectors corresponding to the n items of order sub information;
calculating Euclidean distance values between the order vector and a central vector of the machine learning model;
if the Euclidean distance value is smaller than or equal to the extreme Euclidean distance value of the machine learning model, determining that the deduction order is suitable for the machine learning model;
and if the Euclidean distance value is greater than the Euclidean distance extreme value, determining that the deduction order is not suitable for the machine learning model.
The order vector may be an n-dimensional vector formed by arranging the n items of order sub-information according to a preset order, for example: the order sub-information includes: the overdue day, the deducted amount, the order number and the deducted date, the order vector may be a 4-dimensional vector: (number of overdue days, amount of money deducted, order number, date of money deducted).
In addition, the center vector of the machine learning model can be expressed by formula
Is determined wherein c1,c2,…,cnAre respectively the central vectors
The number of n coordinates of (a) is,
are respectively provided withThe order sub-information is an average value of n items of order sub-information in a first historical order set, and the first historical order set is a deduction order data set used for training to obtain the machine learning model.
The first historical order set may include a plurality of historical deduction orders in a historical time period and an actual deduction result of the plurality of historical deduction orders.
In addition, the above euclidean distance value can be represented by the formula:
is given by, wherein (x)1,x2,…,xn) Is the order vector for the deduction order.
For example: if the center vector C of the machine learning model is (1616.34465,47.6547), and the maximum value of the euclidean distance is | X-C |max69347.756, indicating that the Euclidean distance between the deduction order and the central vector is less than or equal to 69347.756, it can be determined that the deduction order is applicable to the machine learning model, wherein X represents the order vector of any one deduction order in the first historical order set.
In addition, the above-mentioned euclidean distance extreme value may be an order vector of any deduction order in the first historical order set and the above-mentioned central vector
The maximum value of the euclidean distance therebetween. In the deduction order and the central vector
The euclidean distance value of (a) is greater than the maximum value of the euclidean distance, which indicates that the deduction order has a larger difference from the deduction orders in the first historical order set, and the machine learning model is trained according to the deduction orders in the first historical order set, so that the deduction orders with larger differences are not suitable for the machine learning model.
In this embodiment, whether the deduction order is applicable to the machine learning model can be determined by calculating the euclidean distance value between the deduction order and the central vector, so that the machine learning model can be used for prediction when the deduction order is applicable to the machine learning model, thereby avoiding that a wrong prediction result is obtained by using the machine learning model when the deduction order is not applicable to the machine learning model, and improving the accuracy of the deduction instruction initiating method.
Of course, the deduction order is applicable to the machine learning model, and the type of the deduction order may be matched with the parameters in the machine learning model, or the type of the deduction order may be matched with the machine learning model.
For example: the machine learning model is only suitable for deduction orders of bank deduction channels, and is not suitable for deduction orders of internet bank deduction channels.
As an optional implementation manner, as shown in fig. 3, the n items of order sub-information include at least one of a deduction amount, a number of overdue days, and a time of last generation of a deduction instruction, and in a case where the deduction instruction is predicted not to be able to be successfully deducted, that is, in a case where the model shown in fig. 3 predicts "no", the deduction instruction initiating method may further include:
detecting whether a deduction order which predicts that the deduction instruction cannot successfully deduct money meets a bottom-binding strategy or not;
and initiating a money deduction instruction for the money deduction order of the bottom pocket, wherein the money deduction order of the bottom pocket comprises money deduction orders meeting the policy of the bottom pocket.
Wherein, the above-mentioned satisfying the bottom pocket strategy may include: the overdue days of the deduction order are larger than the preset overdue days, and/or the deduction amount is larger than the preset deduction amount, and/or the time of generating the deduction instruction last time exceeds the preset deduction period. And under the condition that the bottom-trapping strategy is met, initiating a money deducting instruction for a money deducting order of which the overdue day number is larger than the preset overdue day number and/or the money deducting amount is larger than the preset money deducting amount and/or the time for generating the money deducting instruction last time exceeds the preset money deducting period.
The preset overdue days can be any number of days such as 7 days, 15 days, 30 days and the like, the preset deduction amount can be any amount such as 1000 RMB, 10000 RMB and the like, and the amount can be other any currency units such as U.S. dollars, yen and the like.
The preset deduction period may be any time period such as 1 week and 1 month, and may be any time period having a period of any number of days or hours such as 10 days and 15 days.
The bottom-pocketing strategy can perform bottom-pocketing on the money-withholding orders which are long in overdue days, large in money-withholding amount and not initiating the money-withholding instruction for a long time, and can be executed before the step of predicting the predicted value of the successful money-withholding of the money-withholding orders through a machine learning model, so that the machine learning time can be reduced, and the money-withholding instruction can be directly initiated.
In this embodiment, when the deduction amount is greater than the preset deduction amount, it indicates that the amount related to the deduction order is large, and the preset deduction amount can be set according to the actual situation, so as to achieve the effects of timely refunding the large deduction order and the like; under the condition that the overdue days of the deduction order are larger than the preset overdue days, indicating that the deduction order is overdue for a long time, and initiating a deduction instruction so that a third party deduction channel can execute a deduction operation on the overdue deduction order; in addition, under the condition that the time for generating the deduction instruction last time exceeds the preset deduction period, the deduction order is shown to have a long time without executing the deduction operation, in order to avoid the order which is not suitable for the machine learning model or the order with multiple deduction failure prediction from not executing the deduction operation for a long time, the deduction instruction is initiated aiming at the deduction order which is overdue for a long time and has no deduction for a long time, and therefore the applicability of the deduction instruction initiating method is improved.
It should be noted that, as shown in fig. 3, in the case where the deduction instruction is predicted to be able to be successfully deducted, that is, in the case where the model prediction is "yes" as shown in fig. 3, the deduction instruction may be directly initiated.
Optionally, as shown in fig. 3, before initiating the deduction instruction, the deduction instruction initiating method may further include the following steps:
and making a deduction instruction.
The step of making the deduction instruction can include making deduction information of a deduction order needing deduction in the deduction instruction, and the third party deduction channel can execute deduction operation according to the deduction information in the deduction instruction.
In this embodiment, the money deduction instruction made by the money deduction instruction can be used for initiating the money deduction instruction to be sent to a third party money deduction channel so as to execute money deduction operation.
As an optional implementation manner, the machine learning model may be a logistic regression model, where the order information includes n items of order sub information, and the logistic regression model at least includes n model coefficients, where the n model coefficients respectively correspond to the n items of order sub information, and n is an integer greater than or equal to 1.
Wherein the logistic regression model may be
Wherein, p is the predicted value, when i is equal to 0, x isiIs constant, when i is not equal to 0, the xiFor the ith order sub-information, the logistic regression model comprises the n +1 model coefficients, the betaiIs xiThe model coefficients of (2).
Wherein the n +1 model coefficients include n model coefficients respectively corresponding to the n pieces of order sub information, that is, β1To betanAnd when and i is equal to 0, said xiModel coefficients corresponding to constants, i.e. beta0。
Of course, the above logistic regression model may not be fixed as the formula:
for example: i can be defined as a positive integer, and a constant Y is added to the above formula to obtain a formula of the logistic regression model:
as an optional implementation manner, as shown in fig. 4, the logistic regression model is obtained by training, where the training process is as follows:
acquiring a first historical order set, wherein each historical order in the first historical order set comprises m order sub-information, and m is an integer greater than or equal to n;
gradually substituting m items of order sub-information of the first historical order set as covariates into an initial regression model, and performing significance analysis on the m items of order sub-information to obtain the correlation degree of each item of order sub-information and the initial regression model;
and selecting n items of order sub-information of which the correlation degree with the initial regression model meets preset conditions from the m items of order sub-information.
The first historical order set may be a set of a plurality of deduction orders which have executed deduction operations in a first historical time period.
In addition, the first historical order set may also be referred to as a "training data set" as shown in fig. 4, and the m order sub-information may also be referred to as an "initial covariate set composition" as shown in fig. 4.
It should be noted that the "applicability domain determination" in fig. 4 may be to determine an extreme euclidean distance value of the logistic regression model, and when the euclidean distance value between the order vector of the deduction order and the central vector of the logistic regression model is smaller than or equal to the extreme euclidean distance value, it may be determined that the deduction order is applicable to the logistic regression model.
The initial regression model may be configured before the training, for example: a user configured initial regression model. The initial regression model corresponds to the m items of order sub information, and the training is used for optimizing the initial regression model to obtain the logistic regression model corresponding to the n items of order sub information.
For example: as described aboveThe initial regression model may be the formula:
the method comprises the following steps of respectively and gradually bringing m items of order sub-information into the initial regression model, and performing significance analysis on the m items of order sub-information, thereby selecting n items of order sub-information with high correlation degree with the initial regression model to obtain a formula of the logistic regression model:
it should be noted that what is determined by the above manner is the type of the n items of order sub information, rather than the specific values of the n items of order sub information, for example: the m items of order sub-information include: the order number, the deduction channel, the deduction date, the deduction amount, the number of overdue days and the like, wherein the n items of order sub-information can be the deduction amount and the number of overdue days in the m items of order sub-information.
In addition, the types of the n order sub-information can be matched with xiIs in a one-to-one correspondence, where i is a positive integer.
The significance analysis method can be any one of a Lagrange multiplier test method, a chi-square test method, significance level value calculation, goodness-of-fit index test, a ratio calculation method and the like. In addition, the step of the significance analysis can also be understood as a step of "hypothesis testing" shown in fig. 4, and testing the correlation between each item of order sub-information and the logistic regression model.
In this embodiment, a part of order sub-information with higher correlation with the predicted value of the initial regression model in the order sub-information is selected as a covariate of the logistic regression function through the significance analysis, so that order sub-information with lower correlation or irrelevant order sub-information can be reduced, for example: the order number of the deduction order interferes with the predicted value of the logistic regression function, so that the accuracy of the logistic regression function can be improved.
Optionally, the step of gradually substituting m items of order sub information of the first historical order set as covariates into an initial regression model, and performing significance analysis on the m items of order sub information to obtain the correlation between each item of order sub information and the initial regression model includes:
taking m items of order sub information of the first historical order set as covariates, gradually substituting the covariates into the initial regression model, and adopting Lagrange multiplier test to obtain the correlation degree of each item of order sub information and the initial regression model; or
Gradually substituting m items of order sub-information of the first historical order set into the initial regression model as covariates, and obtaining the correlation degree of each item of order sub-information and the initial regression model by adopting chi-square test; or
Gradually substituting m items of order sub-information of the first historical order set into the initial regression model as covariates, and obtaining the correlation degree of each item of order sub-information and the initial regression model by adopting a significance level value calculation mode; or
Gradually substituting m items of order sub information of the first historical order set into the initial regression model as covariates, and adopting goodness-of-fit index inspection to obtain the correlation degree of each item of order sub information and the initial regression model; or
And taking m items of order sub information of the first historical order set as covariates to be gradually substituted into the initial regression model, and obtaining the correlation degree of each item of order sub information and the initial regression model by adopting a ratio calculation mode of a specific confidence interval.
Different significance analysis methods can be adopted to analyze the significance level of each order sub-information according to whether the value of the order sub-information belongs to a continuous variable or a discontinuous variable.
For example, for continuous variables such as the deduction amount, the overdue day number, and the order number with wide numerical value variation range, the significance analysis is performed by adopting a method of gradually introducing covariates, and specifically, the deduction amount can be substituted into the initial regression model as the covariates:
assuming that the confidence interval of the regression coefficient is 95% and the target threshold is equal to 0.5, gradually substituting the deduction amount of the plurality of deduction orders in the first historical order set into the logistic regression model, and checking the regression coefficient of the deduction amount by adopting a Lagrange multiplier-checking method so as to obtain the significance level of the deduction amount, and if the significance level of the deduction amount is less than or equal to 0.05, determining that the correlation degree of the deduction amount is high, so that the types of the n items of order sub-information include the deduction amount; if the significance level of the deduction amount is greater than or equal to 0.1, determining that the relevance of the deduction amount is low, and thus the type of the n items of order sub-information does not include the deduction amount.
The n items of order sub-information with the correlation degree meeting the preset condition may include order sub-information with the significance level less than or equal to 0.05.
For non-numerical variables or non-continuous variables with limited numerical variation ranges, the correlation degree of the non-continuous variables can be determined by means of chi-square test, significance level value calculation, goodness-of-fit index test, specific confidence interval ratio calculation and the like.
For example, the discontinuous variables: the deduction channel can be any one of bank and network payment, and a dummy variable analysis method can be adopted for the deduction channel to determine the correlation degree of the deduction channel. Specifically, when the deduction channel is a bank, the deduction channel takes a value of 1, when the deduction channel is a network payment, the deduction channel takes a value of 2, the deduction channels of the plurality of deduction orders in the first historical order set are respectively substituted into the initial regression model, and a card square checking method is adopted to obtain a card square value of the deduction channel, wherein the card square value is equal to a square value of a standard error of the deduction channel divided by a model coefficient of the deduction channel, and the card square value of the deduction channel is smaller than a preset value or much smaller than those of other discontinuous variables by comparison, for example: the credit card value of the debit number is 0.001, and the credit card value of the debit channel is 0.00001, it is determined that the correlation of the debit channel is small, and thus the n items of order sub-information of the logistic regression function do not include the debit channel.
The n items of order sub-information with the correlation degree meeting the preset condition may include order sub-information with a chi-square value smaller than a preset value or a chi-square value smaller than other discontinuous variables.
Alternatively, the sig value of each non-continuous variable may be determined by calculating the sig value of the significance level, and the non-continuous variables with the sig values greater than or equal to 0.1 are excluded from the logistic regression model, that is, the n pieces of order sub-information of the logistic regression function do not include the non-continuous variables with the sig values greater than or equal to 0.1.
The n items of order sub information whose correlation degree meets the preset condition may include the order sub information whose sig value is less than 0.1.
In addition, a goodness-of-fit index verification method can be adopted to determine the sig value of each discontinuous variable, wherein the n items of order sub-information of the logistic regression function only comprise the discontinuous variables of which the sig values are greater than or equal to 0.05.
The n items of order sub information with the correlation degree meeting the preset condition may include order sub information with a sig value greater than or equal to 0.05.
In addition, a ratio analysis method of a specific confidence interval may also be used to determine the n items of order sub-information of the logistic regression function, where the specific confidence interval may be 95%, and of course, the specific confidence interval may also be other intervals such as 85%, 90%, 99%, and the like. Specifically, the n order sub-information items of the logistic regression function do not include discontinuous variables with smaller chi-squared values and ratio ratios equal to about 1.
The n items of order sub-information with the correlation degree meeting the preset condition may include order sub-information with a chi-square value larger than a preset value and/or a ratio smaller than 1.
In addition, the model coefficient in the logistic regression function, which corresponds to each piece of order sub information one to one, may be a constant obtained in advance, or may be a numerical value determined according to the sensitivity, specificity, accuracy, and the like of the logistic regression function after determining the types of the n pieces of order sub information and performing iterative optimization.
For example, if the logistic regression model includes 2 items of order sub-information, respectively: the number of overdue days and the amount of deductions, the logistic regression model can be expressed as:
specifically, the parameters of the logistic regression model are shown in table 1 below:
TABLE 1
It should be noted that, in the embodiment of the present invention, the determination of the n items of order sub-information by performing significance analysis is not limited, and for example: the order sub-information may be the order sub-information of the n items preset by the user.
As an optional implementation, the training process further includes:
analyzing the predicted value of each order of the first historical order and the actual deduction result of each order by using a Receiver Operating Characteristic (ROC) Curve to obtain an ROC Curve with the ordinate as sensitivity and the abscissa as 1 minus specificity;
and searching for specific sensitivity and specific specificity through the ROC curve, and taking the average value of the predicted values corresponding to the specific sensitivity and the specific specificity as the target threshold, wherein the difference between the specific sensitivity and the specific specificity in the ROC curve is the largest.
Wherein, the above step in the training process may be "ROC analysis" as shown in fig. 4. From the parameters of the logistic regression model, a ROC curve can be obtained as shown in FIG. 5, wherein the abscissa of the ROC curve is 1-specificity and the ordinate is sensitivity.
The sensitivity may be defined as (the number of the deduction orders with the predicted value greater than or equal to the preset threshold value/the number of the deduction orders with the actual deduction success) × 100%, which represents the probability of the deduction success of the correctly predicted deduction order.
In addition, the specificity may be defined as (the number of the deduction orders with the predicted value smaller than the preset threshold value/the number of the deduction orders with the deduction failure) × 100%, which represents the probability of correctly predicting the deduction failure of the deduction order.
In the "AUC and Cut _ Off optimization" step shown in fig. 4, the coverage of the ROC curve may be obtained from the ROC curve, and the coverage may also be referred to as AUC (area under curve) and represents the area under the ROC curve. As shown in table 2 below, the AUC parameters of the logistic regression model:
TABLE 2
Wherein the test variables of table 2 are predicted values of the logistic regression model.
For example: the calculation process of the target threshold value can be represented as follows: the target threshold value is Avg (P value corresponding to Max (sensitivity-specificity)).
Where Avg represents the calculated number average, and Max represents the maximum.
By searching a coordinate list of an ROC curve, the values of the specific sensitivity and the specific specificity can be determined when the value of the sensitivity-specificity is the maximum, so that a plurality of predicted values output by the logistic regression function when the value of the specific sensitivity and the specific specificity are matched with each other are determined, and the target threshold value is obtained by arithmetically averaging the plurality of predicted values.
As shown in Table 3, are examples of the logistic regression model target threshold, sensitivity, and specificity:
TABLE 3
Table 3 shows that the maximum difference between the sensitivity and the specificity of the logistic regression model is equal to 0.869-0.267, and the predicted values corresponding to the sensitivity and the specificity are respectively: 0.3359636, 0.3359773, …, 0.3360504, the target threshold value is Avg (Max (sensitivity-specificity) corresponding P value) is 0.336.
In this embodiment, by obtaining the target threshold value when the difference between the sensitivity and the specificity of the logistic regression model is the maximum, the prediction result can be in the best sensitivity when the logistic regression model is used to predict whether the deduction result of the deduction order is successful, so that the accuracy of the logistic regression model is improved, and the effect of improving the accuracy of the deduction instruction initiating method is achieved.
Optionally, the training process further includes:
acquiring order information and an actual money deduction result of each order in the second historical order set;
substituting the order information of each order in the second historical order set into the logistic regression model to obtain a predicted value of each order in the second historical order set;
comparing the predicted value of each order in the second historical order set with the target threshold value to obtain a predicted deduction result of each order in the second historical order set;
and comparing the predicted deduction result of each order in the second historical order set with the actual deduction result of each order in the second historical order set to obtain the coverage rate of successful deduction of the logistic regression model.
The second historical order set comprises a plurality of deduction orders and actual deduction results of the deduction orders.
In addition, the training process can be used to verify the accuracy of the logistic regression model obtained by the training step, and the "model test" as shown in fig. 4, and the second historical order set can also be referred to as the "test data set" as shown in fig. 4.
It should be noted that, the first historical order set and the second historical order set may be a plurality of deduction orders for which deduction operations have been performed within the same time period, and a ratio of deduction orders with deduction success and deduction failure in the first historical order set is substantially the same as a ratio of deduction orders with deduction success and deduction failure in the second historical order set, for example: the order in the first historical order set with the deduction success accounts for 20%, and the order in the second historical order set with the deduction success can be between 18% and 22%.
In addition, the number of deduction orders included in the first history order set and the second history order set can be distributed in a ratio of 2:1, for example: if the first historical order set comprises 200 deduction orders, the second historical order set comprises 100 deduction orders.
As shown in table 4, the test results of the logistic regression model using the second historical order set are:
TABLE 4
| Target threshold | Coverage rate | Predicted quantitative ratio |
| 0.336 | 88.010％ | 41.370％ |
The sensitivity of the logic back model is equal to 87%, the coverage rate is the number of the deduction orders with the predicted value larger than the target threshold value/the actual deduction result is the number of the deduction orders with the successful deduction, and the predicted quantity ratio is the number of the deduction orders with the predicted value larger than the target threshold value/the total number of the deduction orders.
As can be seen from the results shown in table 4, after the logistic regression model predicts that the deduction instruction is issued to the deduction order which is predicted to be able to be successfully deducted, only about 41% of the original deduction instruction needs to be issued, so that 88% of the deduction instruction which is able to be successfully deducted can be covered, for example: originally, the deduction instruction is issued for 100 deduction orders, and only 25 deduction orders can be successfully deducted, but in the embodiment, after prediction is carried out through a logistic regression model, only 41 deduction instructions are issued, so that 22 deduction orders can be successfully deducted.
Of course, the sensitivity of the above logic loop model can also be set to other values as required, for example: 95%, 99%, etc., so that the higher the sensitivity, the more deduction instructions initiated by the logic loop model, the higher its coverage.
For example, as shown in table 5:
TABLE 5
| Amount of money deducted | Days of expiry | Probability of success P of model prediction | Predicted results | Actual deduction result |
| 850.6 | 75 | 0.0000005403 | Failure of | Failure of |
Taking a logistic regression model with the sensitivity equal to 99% as an example, the target threshold of the logistic regression model is equal to 0.008, and the deduction amount and the overdue days of the deduction order shown in table 5 are substituted into the logistic regression model for prediction, so that the obtained prediction value is smaller than the target threshold of 0.008. Therefore, the logistic regression model will lose money, so that the deduction command is not initiated any more, and the deduction result of the deduction order is actually observed to be a deduction failure. Therefore, the deduction cost of the next deduction can be saved after the logistic regression model is used for prediction.
In this embodiment, the deduction orders in the second historical order set are taken in, and the obtained prediction result is compared with the deduction result of the deduction order, so as to determine whether the prediction result of the logistic regression model is accurate, and thus, if it is determined that the accuracy of the prediction result of the logistic regression model is not high, the parameters of the logistic regression model can be further optimized, so that the accuracy of the logistic regression function is improved.
Of course, the machine learning model may be any one of a neural network, a support vector machine, and the like, and the method for determining the parameters in the machine learning model is not limited to the iteration, chi-square check, and the like listed above.
In the step, the machine learning model is adopted to predict the deduction order suitable for the machine learning model so as to obtain the prediction result that the deduction operation can be successfully executed, so that a deduction instruction is initiated on the deduction order under the condition that the deduction operation can be successfully predicted, the accuracy of the machine learning model is improved, and the effect of improving the accuracy of the deduction instruction initiating method is achieved.
And 203, if the predicted value is greater than or equal to the target threshold value, initiating a deduction instruction of the deduction order.
And 204, if the predicted value is smaller than the target threshold value, predicting that the money deduction result of the money deduction order is money deduction failure.
When the deduction result of the deduction order is predicted to be deduction failure, supplementary deduction can be carried out according to the overdue day number, the deduction amount and the time far away from the last deduction of the deduction order, and a deduction instruction is initiated on the order with the overdue day number, the deduction amount being large or the order without deduction for a long time, so that supplementary deduction operation is carried out on the order.
Of course, the deduction instruction is not initiated for the order with the prediction result of the deduction failure.
In this step, the deduction order with the deduction failure can be predicted, so that the deduction instruction is not initiated for the order, and the deduction cost generated by executing the deduction operation on the deduction order which cannot be successfully deducted can be saved, thereby reducing the deduction cost of the deduction instruction initiating method.
In the embodiment of the invention, before the machine learning model is used for predicting the predicted value of the successful deduction of the deduction order, whether the deduction order is suitable for the machine learning model or not is detected, so that the deduction order which is not suitable for the machine learning model can be prevented from learning by using the machine learning model to obtain a wrong prediction result, the accuracy of the deduction instruction initiating method can be improved, and the deduction cost of the deduction instruction initiating method can be further reduced.
Referring to fig. 6, an embodiment of the present invention further provides an initiating device 600 for a deduction instruction, where the initiating device 600 for a deduction instruction includes:
a first obtaining module 601, configured to obtain order information of a deduction order;
the first prediction module 602 is configured to input the order information into a machine learning model for processing, and then obtain a predicted value of successful deduction of the deduction order;
the initiating module 603 is configured to initiate a deduction instruction of the deduction order if the predicted value is greater than or equal to a target threshold.
Optionally, as shown in fig. 7, the apparatus 600 further includes:
a second prediction module 604, configured to predict that the money deduction result of the money deduction order is money deduction failure if the prediction value is smaller than the target threshold.
Optionally, the order information includes: n items of order sub information, where n is an integer greater than or equal to 1, as shown in fig. 8, the apparatus 600 further includes:
a judging module 605, configured to judge whether the deduction order is applicable to the machine learning model according to the n items of order sub-information;
an executing module 606, configured to execute the step of obtaining a predicted value of a successful deduction of the deduction order after inputting the order information into the machine learning model for processing if the deduction order is applicable to the machine learning model.
Optionally, as shown in fig. 9, the determining module 605 includes:
a generating unit 6051, configured to generate an order vector corresponding to the n items of order sub information;
a calculation unit 6052 configured to calculate a euclidean distance value between the order vector and a center vector of the machine learning model;
a first determining unit 6053, configured to determine that the deduction order is applicable to the machine learning model if the euclidean distance value is less than or equal to the euclidean distance extremum of the machine learning model;
a second determining unit 6054, configured to determine that the deduction order is not applicable to the machine learning model if the euclidean distance value is greater than the euclidean distance extremum.
Optionally, the machine learning model includes a logistic regression model, the order information includes n items of order sub information, the logistic regression model at least includes n model coefficients, where the n model coefficients respectively correspond to the n items of order sub information, and n is an integer greater than or equal to 1.
Optionally, the logistic regression model is:
wherein, p is the predicted value, and when i is equal to 0, x isiIs constant, when i is not equal to 0, the xiFor the ith order sub-information, the logistic regression model comprises the n +1 model coefficients, the betaiIs xiThe model coefficients of (2).
Optionally, the logistic regression model is obtained by training, wherein the training process is as follows:
acquiring a first historical order set, wherein each historical order in the first historical order set comprises m order sub-information, and m is an integer greater than or equal to n;
gradually substituting m items of order sub-information of the first historical order set as covariates into an initial regression model, and performing significance analysis on the m items of order sub-information to obtain the correlation degree of each item of order sub-information and the initial regression model;
and selecting n pieces of order sub information of which the correlation degree with the initial regression model meets preset conditions from the m pieces of order sub information, and optimizing the initial regression model according to the n pieces of order sub information to obtain the logistic regression model.
Optionally, the step of gradually substituting m pieces of order sub information of the first historical order set as covariates into an initial regression model, and performing significance analysis on the m pieces of order sub information to obtain a correlation between each piece of order sub information and the initial regression model includes:
taking m items of order sub information of the first historical order set as covariates, gradually substituting the covariates into the initial regression model, and adopting Lagrange multiplier test to obtain the correlation degree of each item of order sub information and the initial regression model; or
Gradually substituting m items of order sub-information of the first historical order set into the initial regression model as covariates, and obtaining the correlation degree of each item of order sub-information and the initial regression model by adopting chi-square test; or
Gradually substituting m items of order sub-information of the first historical order set into the initial regression model as covariates, and obtaining the correlation degree of each item of order sub-information and the initial regression model by adopting a significance level value calculation mode; or
Gradually substituting m items of order sub information of the first historical order set into the initial regression model as covariates, and adopting goodness-of-fit index inspection to obtain the correlation degree of each item of order sub information and the initial regression model; or
And taking m items of order sub information of the first historical order set as covariates to be gradually substituted into the initial regression model, and obtaining the correlation degree of each item of order sub information and the initial regression model by adopting a ratio calculation mode of a specific confidence interval.
Optionally, the training process further includes:
carrying out Receiver Operating Characteristic (ROC) curve analysis on the predicted value of each order in the first historical order set and the actual deduction result of each order to obtain an ROC curve with the ordinate of sensitivity and the abscissa of 1 minus specificity;
and searching for specific sensitivity and specific specificity through the ROC curve, and taking the average value of the predicted values corresponding to the specific sensitivity and the specific specificity as the target threshold value, wherein the difference value between the specific sensitivity and the specific specificity in the ROC curve is the largest.
Optionally, the training process further includes:
acquiring order information and an actual money deduction result of each order in the second historical order set;
substituting the order information of each order in the second historical order set into the logistic regression model to obtain a predicted value of each order in the second historical order set;
comparing the predicted value of each order in the second historical order set with the target threshold value to obtain a predicted deduction result of each order in the second historical order set;
and comparing the predicted deduction result of each order in the second historical order set with the actual deduction result of each order in the second historical order set to obtain the coverage rate of successful deduction of the logistic regression model.
In the implementation of the present invention, each step in the method embodiments shown in fig. 1 to fig. 2 can be implemented, and the same beneficial effects can be obtained, and in order to avoid repetition, the details are not described herein again.
Fig. 10 is a structural diagram of an initiating device of a deduction instruction according to an embodiment of the present invention, and as shown in fig. 10, the initiating device includes: memory 1001, processor 1002, transceiver 1003 and a computer program stored on said memory 1001 and executable on said processor 1002, wherein:
the processor 1002 is configured to read a program in the memory 1001, may execute each step in the method for initiating a debit instruction according to the embodiment of the present invention, and may achieve the same technical effect, and is not described herein again to avoid repetition.
An embodiment of the present invention further provides a computer storage medium, where a computer program is stored on the computer readable storage medium, and when executed by a processor, the computer program implements the steps in the deduction instruction initiating method, and can achieve the same technical effect, and in order to avoid repetition, details are not repeated here.
It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.
Through the above description of the embodiments, those skilled in the art will clearly understand that the method of the above embodiments can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware, but in many cases, the former is a better implementation manner. Based on such understanding, the technical solution of the present invention or the portions contributing to the prior art may be embodied in the form of a software product, where the computer software product is stored in a storage medium (such as ROM/RAM, magnetic disk, optical disk), and includes several instructions for enabling an initiating device (which may be a mobile phone, a computer, a server, an air conditioner, or a network device) of a deduction instruction to execute the method according to the embodiments of the present invention.
While the present invention has been described with reference to the embodiments shown in the drawings, the present invention is not limited to the embodiments, which are illustrative and not restrictive, and it will be apparent to those skilled in the art that various changes and modifications can be made therein without departing from the spirit and scope of the invention as defined in the appended claims.
Claims (11)
1. A deduction instruction initiating method is characterized by comprising the following steps:
acquiring order information of a deduction order;
judging whether the deduction order is suitable for a machine learning model according to the order information, wherein the machine learning model is obtained by training the order information of the historical deduction order;
if the deduction order is suitable for the machine learning model, inputting the order information into the machine learning model for processing, and obtaining a predicted value of successful deduction of the deduction order;
if the predicted value is larger than or equal to a target threshold value, initiating a money deduction instruction of the money deduction order;
wherein, said judging whether the deduction order is suitable for a machine learning model according to the order information comprises:
the order information is applicable to the machine learning model when the category of the deduction order information matches a parameter in the machine learning model or when the type of the deduction order matches the machine learning model.
2. The method of claim 1, wherein the order information comprises: n items of order sub-information, wherein n is an integer greater than or equal to 1, and the step of judging whether the deduction order is suitable for a machine learning model according to the order information comprises the following steps:
generating order vectors corresponding to the n items of order sub information;
calculating Euclidean distance values between the order vector and a central vector of the machine learning model;
and if the Euclidean distance value is smaller than or equal to the extreme Euclidean distance value of the machine learning model, determining that the deduction order is suitable for the machine learning model.
3. The method of claim 2, wherein the Euclidean distance extremum is a difference between an order vector of any one of a set of historical orders from which the machine learning model was trained and a center vector of the machine learning model;
the center vector is an average of the sub-information of each order in the historical order set for which the machine learning model was trained.
4. The method of claim 1, wherein the order information comprises: the machine learning model at least comprises n model coefficients, wherein the n model coefficients respectively correspond to the n order sub information, and n is an integer greater than or equal to 1.
5. The method of claim 1, wherein after entering the order information into a machine learning model process and before obtaining a predictive value of a successful debit for the debit order, the method further comprises:
detecting whether the deduction order meets a bottom-pocket strategy, wherein the bottom-pocket strategy comprises the following steps: the overdue days of the deduction order are more than the preset overdue days, the deduction amount is more than the preset deduction amount, and the time from the last generation of the deduction instruction exceeds one or more of the preset deduction period;
and if the deduction order meets the bottom-holding strategy, initiating a deduction instruction of the deduction order.
6. A method of training a machine learning model, the method comprising:
acquiring a first historical order set, wherein the first historical order set comprises a plurality of historical deduction orders;
gradually substituting the plurality of historical deduction orders into an initial machine learning model for significance analysis to obtain the correlation degree of each historical deduction order and the initial machine learning model;
and selecting a historical deduction order with the correlation degree meeting a preset condition to optimize the initial machine learning model to obtain the machine learning model.
7. The method of claim 6, wherein said progressively substituting said plurality of historical deduction orders into an initial machine learning model for significance analysis to obtain a degree of correlation of each historical deduction order with said initial machine learning model comprises:
when the numerical values in the plurality of historical withholding orders belong to continuous variables, taking the plurality of historical withholding orders as covariates to be gradually substituted into the initial machine learning model, and obtaining the correlation degree of each historical withholding order and the initial machine learning model by adopting Lagrange multiplier test; or alternatively
When non-numerical variables or discontinuous variables with limited numerical variation ranges exist in the plurality of historical deduction orders, the plurality of historical deduction orders are used as covariates to be gradually substituted into the initial machine learning model, and the correlation degree of each historical deduction order and the initial machine learning model is obtained by adopting any one of chi-square test, significance level value calculation, goodness-of-fit index test and specific confidence interval ratio calculation modes.
8. The method of claim 6, wherein after the obtaining the machine learning model, the method further comprises:
carrying out Receiver Operating Characteristic (ROC) curve analysis on the predicted value of each order in the first historical order set and the actual deduction result of each order to obtain an ROC curve with the ordinate of sensitivity and the abscissa of 1 minus specificity;
and taking the average value of the predicted values corresponding to the sensitivity and the specificity with the maximum difference in the ROC curve as a target threshold value, wherein the target threshold value is used for judging whether the deduction order can be successfully deducted.
9. A deduction instruction initiating device, the device comprising:
the first acquisition module is used for acquiring order information of a deduction order;
the judging module is used for judging whether the deduction order is suitable for a machine learning model according to the order information, and the machine learning model is obtained by training the order information of the historical deduction order;
the first prediction module is used for inputting the order information into the machine learning model for processing if the deduction order is suitable for the machine learning model, and obtaining a prediction value of successful deduction of the deduction order;
the initiating module is used for initiating a deduction instruction of the deduction order if the predicted value is larger than or equal to a target threshold value;
the judging module is used for enabling the order information to be suitable for the machine learning model when the category of the deduction order information is matched with the parameters in the machine learning model or when the type of the deduction order information is matched with the machine learning model.
10. An electronic device, comprising: a processor, a memory, a bus;
the processor and the memory complete mutual communication through the bus; the processor is configured to invoke program instructions in the memory to perform the deduction instruction initiating method of any one of claims 1 to 5 or the machine learning model training method of any one of claims 6 to 8.
11. A computer-readable storage medium, comprising: a stored program;
when the program runs, the device of the storage medium is controlled to execute the deduction instruction initiating method of any one of claims 1 to 5 or the machine learning model training method of any one of claims 6 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110589601.0A CN113298510B (en) | 2018-07-10 | 2018-07-10 | Deduction instruction initiating method and device |
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810750604.6A CN110705975B (en) | 2018-07-10 | 2018-07-10 | Deduction instruction initiating method and device |
| CN202110589601.0A CN113298510B (en) | 2018-07-10 | 2018-07-10 | Deduction instruction initiating method and device |
Related Parent Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810750604.6A Division CN110705975B (en) | 2018-07-10 | 2018-07-10 | Deduction instruction initiating method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113298510A CN113298510A (en) | 2021-08-24 |
| CN113298510B true CN113298510B (en) | 2022-06-17 |
Family
ID=69192315
Family Applications (2)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810750604.6A Active CN110705975B (en) | 2018-07-10 | 2018-07-10 | Deduction instruction initiating method and device |
| CN202110589601.0A Active CN113298510B (en) | 2018-07-10 | 2018-07-10 | Deduction instruction initiating method and device |
Family Applications Before (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810750604.6A Active CN110705975B (en) | 2018-07-10 | 2018-07-10 | Deduction instruction initiating method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (2) | CN110705975B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116308215B (en) * | 2023-05-17 | 2023-07-21 | 云账户技术(天津)有限公司 | Method and device for generating batch outgoing information and related equipment |
| CN116882986A (en) * | 2023-07-10 | 2023-10-13 | 支付宝(杭州)信息技术有限公司 | Payment processing method and device based on medical insurance certificate |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105426956A (en) * | 2015-11-06 | 2016-03-23 | 国家电网公司 | Ultra-short-period photovoltaic prediction method |
| CN106021364A (en) * | 2016-05-10 | 2016-10-12 | 百度在线网络技术（北京）有限公司 | Method and device for establishing picture search correlation prediction model, and picture search method and device |
| CN106844152A (en) * | 2017-01-17 | 2017-06-13 | 清华大学 | Bank's background task runs the correlation analysis and device of batch time |
| CN106920099A (en) * | 2017-03-09 | 2017-07-04 | 携程旅游信息技术（上海）有限公司 | The payment routing intelligence monitoring method and system of payment system |
| CN107590689A (en) * | 2017-08-25 | 2018-01-16 | 北京奇艺世纪科技有限公司 | A kind of ad data recommends method and system |
Family Cites Families (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102829967B (en) * | 2012-08-27 | 2015-12-16 | 中国舰船研究设计中心 | A kind of time domain fault recognition method based on regression model index variation |
| AU2014302603A1 (en) * | 2013-06-24 | 2016-01-07 | Cylance Inc. | Automated system for generative multimodel multiclass classification and similarity analysis using machine learning |
| US20160004798A1 (en) * | 2014-07-07 | 2016-01-07 | Ingersoll-Rand Company | Determining the performance of an environmental system using a predictive model |
| CN107025384A (en) * | 2015-10-15 | 2017-08-08 | 赵乐平 | A kind of construction method of complex data forecast model |
| US10565599B2 (en) * | 2015-12-16 | 2020-02-18 | Alegeus Technologies, Llc | Systems and methods for managing information technology infrastructure to generate a dynamic interface |
| US20180053188A1 (en) * | 2016-08-17 | 2018-02-22 | Fair Isaac Corporation | Customer transaction behavioral archetype analytics for cnp merchant transaction fraud detection |
| CN107784312B (en) * | 2016-08-24 | 2020-12-22 | 腾讯征信有限公司 | Machine learning model training method and device |
| US10489719B2 (en) * | 2016-09-09 | 2019-11-26 | Facebook, Inc. | Shared per content provider prediction models |
| CN106897918A (en) * | 2017-02-24 | 2017-06-27 | 上海易贷网金融信息服务有限公司 | A kind of hybrid machine learning credit scoring model construction method |
| CN107424070A (en) * | 2017-03-29 | 2017-12-01 | 广州汇融易互联网金融信息服务有限公司 | A kind of loan user credit ranking method and system based on machine learning |
| CN107346515A (en) * | 2017-06-01 | 2017-11-14 | 银联智策顾问（上海）有限公司 | A kind of credit card Forecasting Methodology and device by stages |
| CN108109063A (en) * | 2017-12-07 | 2018-06-01 | 上海点融信息科技有限责任公司 | For the method, apparatus and computer readable storage medium of prediction label predicted value |
| CN108256691A (en) * | 2018-02-08 | 2018-07-06 | 成都智宝大数据科技有限公司 | Refund Probabilistic Prediction Model construction method and device |
-
2018
- 2018-07-10 CN CN201810750604.6A patent/CN110705975B/en active Active
- 2018-07-10 CN CN202110589601.0A patent/CN113298510B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105426956A (en) * | 2015-11-06 | 2016-03-23 | 国家电网公司 | Ultra-short-period photovoltaic prediction method |
| CN106021364A (en) * | 2016-05-10 | 2016-10-12 | 百度在线网络技术（北京）有限公司 | Method and device for establishing picture search correlation prediction model, and picture search method and device |
| CN106844152A (en) * | 2017-01-17 | 2017-06-13 | 清华大学 | Bank's background task runs the correlation analysis and device of batch time |
| CN106920099A (en) * | 2017-03-09 | 2017-07-04 | 携程旅游信息技术（上海）有限公司 | The payment routing intelligence monitoring method and system of payment system |
| CN107590689A (en) * | 2017-08-25 | 2018-01-16 | 北京奇艺世纪科技有限公司 | A kind of ad data recommends method and system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113298510A (en) | 2021-08-24 |
| CN110705975B (en) | 2021-06-18 |
| CN110705975A (en) | 2020-01-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108846520B (en) | Loan overdue prediction method, loan overdue prediction device and computer-readable storage medium | |
| US8744899B2 (en) | Systems and methods for migrating customers to alternative financial products | |
| CN111080338A (en) | User data processing method and device, electronic equipment and storage medium | |
| CN111062518A (en) | Method, device and storage medium for processing hastening service based on artificial intelligence | |
| CN113298510B (en) | Deduction instruction initiating method and device | |
| US8762194B2 (en) | Systems and methods for evaluating alternative financial products | |
| CN112561691A (en) | Customer credit prediction method, device, equipment and storage medium | |
| CN115545886A (en) | Overdue risk identification method, overdue risk identification device, overdue risk identification equipment and storage medium | |
| KR20200068069A (en) | Apparatus for predicting loan defaults based on machine learning and method thereof | |
| CN113158124A (en) | Data processing method and device based on artificial intelligence and storage medium | |
| CN111179070A (en) | Loan risk timeliness prediction system and method based on LSTM | |
| CN114140023A (en) | Project risk determination method and device, electronic equipment and storage medium | |
| CN113487408A (en) | Information processing method and device | |
| CN114240599A (en) | Loan calculation method and device, computer equipment and storage medium | |
| US11276117B2 (en) | Generating payables and receivables netting proposals based on historical information | |
| US12148056B1 (en) | Intellectual property model parameter training and utilization | |
| CN109754324A (en) | A kind of stock price forward prediction method | |
| CN116703534B (en) | Intelligent management method for data of electronic commerce orders | |
| CN113837183B (en) | Multi-stage certificate intelligent generation method, system and medium based on real-time mining | |
| CN112232893A (en) | Credit assessment method, device and computer system | |
| CN117635310A (en) | Method, device, equipment and medium for determining overdue risk of loan | |
| CN114970932A (en) | Neural network-based transaction scale prediction method and device | |
| CN115631023A (en) | Credit auditing method and related device | |
| CN114092259A (en) | Processing method, device and equipment for exchange rate data and computer storage medium | |
| CN113591932A (en) | User abnormal behavior processing method and device based on support vector machine |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |