CN113313582A - Guest refusing and reflashing model training method and device and electronic equipment - Google Patents
Guest refusing and reflashing model training method and device and electronic equipment Download PDFInfo
- Publication number
- CN113313582A CN113313582A CN202110713287.2A CN202110713287A CN113313582A CN 113313582 A CN113313582 A CN 113313582A CN 202110713287 A CN202110713287 A CN 202110713287A CN 113313582 A CN113313582 A CN 113313582A
- Authority
- CN
- China
- Prior art keywords
- sample
- model
- user data
- samples
- representing
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000012549 training Methods 0.000 title claims abstract description 109
- 238000000034 method Methods 0.000 title claims abstract description 41
- 238000012545 processing Methods 0.000 claims abstract description 22
- 238000011156 evaluation Methods 0.000 claims description 43
- 238000004364 calculation method Methods 0.000 claims description 9
- 238000012216 screening Methods 0.000 claims description 9
- 238000000605 extraction Methods 0.000 claims description 8
- 238000012163 sequencing technique Methods 0.000 claims description 2
- 230000006870 function Effects 0.000 description 100
- 230000000694 effects Effects 0.000 description 9
- 238000010586 diagram Methods 0.000 description 6
- 230000008569 process Effects 0.000 description 6
- 238000013210 evaluation model Methods 0.000 description 5
- 230000009471 action Effects 0.000 description 3
- 238000004590 computer program Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000006467 substitution reaction Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000000087 stabilizing effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
- G06Q40/03—Credit; Loans; Processing thereof
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Accounting & Taxation (AREA)
- Evolutionary Computation (AREA)
- Software Systems (AREA)
- General Engineering & Computer Science (AREA)
- Finance (AREA)
- Artificial Intelligence (AREA)
- Economics (AREA)
- Technology Law (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Development Economics (AREA)
- Evolutionary Biology (AREA)
- Marketing (AREA)
- Strategic Management (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Business, Economics & Management (AREA)
- Medical Informatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The application provides a method and a device for training a customer refusing and reflashing model and electronic equipment, wherein the method comprises the following steps: determining a model entering variable according to a user data training sample and inputting the model to be trained for processing to obtain an output result; adjusting model parameters of the model to be trained according to the sample label, the output result and a preset target function; the loss function of the model to be trained comprises a first function item and a second function item, the first function item represents the loss function value of the positive sample, the second function item represents the loss function value of the negative sample, the first function item comprises a first penalty coefficient representing that the positive sample is predicted to be the negative sample, the second function item comprises a second penalty coefficient representing that the negative sample is predicted to be the positive sample, and the first penalty coefficient is larger than the second penalty coefficient. Therefore, the attention degree for predicting the positive sample into the negative sample is higher than that for predicting the negative sample into the positive sample, so that the accuracy for predicting the positive sample by the model is improved, and the risk of fishing back customers in the actual use of the model is ensured to be in a controllable range.
Description
Technical Field
The application relates to the technical field of big data, in particular to a method and a device for training a customer refusing and salvage model and electronic equipment.
Background
The method is a wind control mode developed in recent years, and means that a financial lending institution constructs a new wind control model for secondary screening in a guest group which has been rejected by one wind control business, and salvages part of the customers as far as possible for lending on the basis of ensuring controllable risk. The customer refusal and salvage mode can reduce the customer obtaining cost of the financial lending institution, and further improve the passing rate on the basis of stabilizing the risk.
On modeling and thinking, the customer refusing and fishing model is obviously different from the credit evaluation model. Because the visitor-refusal and salvage model needs to ensure controllable risk firstly, and the attention to the overdue rate is higher than the passing rate, the attention to the fact that the positive sample is predicted as the negative sample is higher than the attention to the fact that the negative sample is predicted as the positive sample in the actual business scene, so that the traditional credit evaluation model is not suitable for the visitor-refusal and salvage scene.
Disclosure of Invention
In order to overcome the above defects in the prior art, the present application aims to provide a training method for a customer-refusing and salvage model, which is characterized in that the method includes:
the method comprises the steps of obtaining a plurality of user data training samples, wherein the user data training samples comprise positive samples and negative samples, sample labels of the positive samples represent users who do not need to be salvaged, and sample labels of the negative samples represent users who need to be salvaged;
carrying out feature extraction and feature screening on the user data training sample, and determining a mode entering variable;
inputting the user data training sample and the model entering variable into a model to be trained for processing to obtain an output result of the user data training sample;
adjusting the model parameters of the model to be trained according to the sample label of the user data training sample, the output result and a preset target function;
wherein the loss function of the model to be trained comprises a first function item representing the loss function values of all positive samples and a second function item representing the loss function values of all negative samples, the first function item comprises a first penalty coefficient representing when a positive sample is predicted as a negative sample, the second function item comprises a second penalty coefficient representing when a negative sample is predicted as a positive sample, and the first penalty coefficient is greater than the second penalty coefficient.
In one possible implementation, the formula of the objective function is as follows:
wherein, yiThe authentic label representing the user data training sample i,
representing the probability that the prediction label of the user data training sample i is a positive sample;
data item
Is said first function term, w0Representing the first penalty coefficient;
the modulation coefficients of the loss function values calculated by different positive samples are represented, r represents the weight adjustment proportion, and r is more than or equal to 0;
data item
Is said second function term, w1Representing the second penalty coefficient;
modulation system for calculating loss function value representing different negative samplesThe number r represents the weight adjustment ratio, and r is greater than or equal to 0.
In one possible implementation, the method further includes:
setting the first penalty coefficient, the second penalty coefficient and the weight adjustment proportion according to the proportion of positive and negative samples in the user data training samples; or
Performing a grid search using the user data training samples to determine the first penalty coefficient, second penalty coefficient, and weight adjustment scale.
In one possible implementation, the method further includes:
training the model to be trained for multiple times by using the user data training sample and the model entering variable to obtain different trained models; wherein, different coefficients used in each training are subjected to grid search;
calculating the evaluation index value of each trained model through a preset evaluation function;
and taking the trained model with the highest evaluation index value as a customer refusing and fishing model after training.
In one possible implementation, the evaluation index of the evaluation function includes an area under the curve AUC index or a Kolmogorov-Smirnov index.
In a possible implementation manner, the step of calculating an evaluation index value of each trained model by using a preset evaluation function includes:
and aiming at each trained model, sequencing the user data training samples according to the sequence of the output result of each user data training sample from small to large, and selecting a preset number of samples ranked at the top to obtain an evaluation index value through calculation of the evaluation function.
The application also provides a refuse return-fishing model training device, the device includes:
the data acquisition module is used for acquiring a plurality of user data training samples, wherein the user data training samples comprise positive samples and negative samples, the sample labels of the positive samples represent users who do not need to be salvaged, and the sample labels of the negative samples represent users who need to be salvaged;
the characteristic processing module is used for carrying out characteristic extraction and characteristic screening on the user data training sample and determining a mode entering variable;
the model processing module is used for inputting the user data training sample and the model entering variable into a model to be trained for processing to obtain an output result of the user data training sample;
the parameter adjusting module is used for adjusting the model parameters of the model to be trained according to the sample labels of the user data training samples, the output result and a preset target function;
wherein the loss function of the model to be trained comprises a first function item representing the loss function values of all positive samples and a second function item representing the loss function values of all negative samples, the first function item comprises a first penalty coefficient representing when a positive sample is predicted as a negative sample, the second function item comprises a second penalty coefficient representing when a negative sample is predicted as a positive sample, and the first penalty coefficient is greater than the second penalty coefficient.
In one possible implementation, the formula of the objective function is as follows:
wherein, yiThe authentic label representing the user data training sample i,
representing the probability that the prediction label of the user data training sample i is a positive sample;
data item
Is said first function term, w0Representing the first penalty coefficient;
representing calculated losses of different positive samplesThe modulation coefficient of the loss function value, r represents the weight adjustment proportion, and r is more than or equal to 0;
data item
Is said second function term, w1Representing the second penalty coefficient;
and the modulation coefficients representing the loss function values calculated by different negative samples, r representing the weight adjustment proportion, and r being more than or equal to 0.
In one possible implementation, the apparatus further includes:
the coefficient adjusting module is used for setting the first penalty coefficient, the second penalty coefficient and the weight adjusting proportion according to the proportion of positive and negative samples in the user data training samples; or performing a grid search using the user data training samples to determine the first penalty coefficient, the second penalty coefficient, and the weight adjustment scale.
The application also provides an electronic device which comprises a processor and a machine-readable storage medium, wherein the machine-readable storage medium stores machine-executable instructions, and when the machine-executable instructions are executed by the processor, the method for training the customer refusing and bailing model is realized.
Compared with the prior art, the method has the following beneficial effects:
according to the method, the device and the electronic equipment for training the customer refusing and bailing model, the loss function used in the model is set to comprise a first function item representing the loss function values of all positive samples and a second function item representing the loss function values of all negative samples, the first function item comprises a first penalty coefficient representing that the positive samples are predicted to be the negative samples, the second function item comprises a second penalty coefficient representing that the negative samples are predicted to be the positive samples, and the first penalty coefficient is larger than the second penalty coefficient. The accuracy of the model for predicting the positive sample is improved, the using effect of the model is improved, and the risk of the model for fishing back the client in the actual use is ensured to be within a controllable range. So, can be so that to "predict the negative sample with the positive sample" the attention degree be higher than "predict the negative sample as the positive sample" to improve the accuracy that the model was just sample prediction, promote the model result of use, ensure that the model returns the risk of dragging for the customer in the in-service use and in controllable range.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings that are required to be used in the embodiments will be briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present application and therefore should not be considered as limiting the scope, and for those skilled in the art, other related drawings can be obtained from the drawings without inventive effort.
Fig. 1 is a schematic flow chart illustrating steps of a customer refusal and salvage model training method provided in an embodiment of the present application;
fig. 2 is a schematic diagram of an electronic device provided in an embodiment of the present application;
fig. 3 is a functional module schematic diagram of a customer refusing and salvage model training device provided in the embodiment of the present application.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are some embodiments of the present application, but not all embodiments. The components of the embodiments of the present application, generally described and illustrated in the figures herein, can be arranged and designed in a wide variety of different configurations.
Thus, the following detailed description of the embodiments of the present application, presented in the accompanying drawings, is not intended to limit the scope of the claimed application, but is merely representative of selected embodiments of the application. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.
It should be noted that: like reference numbers and letters refer to like items in the following figures, and thus, once an item is defined in one figure, it need not be further defined and explained in subsequent figures.
In the description of the present application, it is noted that the terms "first", "second", "third", and the like are used merely for distinguishing between descriptions and are not intended to indicate or imply relative importance.
In the description of the present application, it is further noted that, unless expressly stated or limited otherwise, the terms "disposed," "mounted," "connected," and "connected" are to be construed broadly, e.g., as meaning either a fixed connection, a removable connection, or an integral connection; can be mechanically or electrically connected; they may be connected directly or indirectly through intervening media, or they may be interconnected between two elements. The specific meaning of the above terms in the present application can be understood in a specific case by those of ordinary skill in the art.
Only the inventor researches and discovers that the customer refusing and fishing model and the credit evaluation model have obvious difference on modeling and thinking. Because the visitor rejection and salvage model needs to ensure controllable risk firstly, and the attention to the overdue rate is higher than the passing rate, the attention to the fact that the positive sample is predicted as the negative sample is higher than the attention to the fact that the negative sample is predicted as the positive sample in the actual business scene.
In a traditional credit evaluation scenario, the following cross entropy function and evaluation index calculation methods are generally used as an objective function and an evaluation function, respectively:
1. the loss function is used for evaluating the difference degree between the predicted value of the model and the real label, and the smaller the loss function is in the training process of the actual model, the better the performance of the model is.
The cross-entropy function is the most common set of penalty functions in the classification problem. The credit evaluation model is generally a binary problem, so the loss function thereof is generally calculated as follows:
wherein: y isiA real label representing the sample i is set to be a positive sample in a credit evaluation scene, and the value of the real label is 1; the normal repayment sample is a negative sample, and the real label value is 0;
It can be seen from the above formula that the cross entropy function is calculated by directly summing the cross entropies of the training samples, and the weights of different samples are consistent.
2. The evaluation function is used for evaluating the distinguishing effect of the model, and generally, the accuracy, the precision, the recall rate, F1-Score, AUC, KS and the like can be used as evaluation indexes for the two-classification problem. Under a general credit evaluation scenario, the AUC and KS values of the model score on the whole modeling sample are often more concerned. The larger the AUC and KS values, the better the discrimination of the model over the whole sample.
Based on the characteristics, the traditional calculation method using the following cross entropy function and evaluation index as a target function and an evaluation function respectively has the following defects in the scene of refusing customers and fishing back:
1. as known from the conventional calculation of the cross entropy function, the weights of different samples are consistent in the cross entropy function calculation process. There is therefore no differentiated setting when calculating the loss for both the cases "predict positive samples as negative samples" and "predict negative samples as positive samples".
In the scene of refusing to return to fishing, the service requires that the attention degree of the positive sample to be predicted as the negative sample is higher than that of the negative sample to be predicted as the positive sample, because the positive sample brings higher overdue rate and loss under the condition that the passing rate of refusing to return to fishing is relatively low. Therefore, the credit evaluation model trained by using the traditional cross entropy function cannot pay more attention to the classification effect of the positive sample, and the use effect of the final salvage model can be influenced.
2. In the conventional evaluation index calculation, a specific evaluation index is usually calculated for all samples. In the customer refusing and fishing-back scene, the distinguishing effect of the models in the section with the front ranking after the models are ranked according to the results is usually more concerned, for example, the distinguishing effect of the models with the scores on the first 20% of the customer groups or the distinguishing effect of the models with the scores on the first 30% of the customer groups. Therefore, the traditional evaluation index calculation method is not suitable for a customer refusing and fishing model, and the effect of the trained model in actual use may not be optimal.
Based on this, the embodiment provides a training method and device for a customer-refusing and bailing model, and an electronic device, which have solved the above problems, and the scheme provided by the embodiment is described in detail below.
Referring to fig. 1, fig. 1 is a schematic flow chart illustrating steps of a training method for a customer refusal bailing model according to this embodiment, where the method may include the following steps.
Step S110, obtaining a plurality of user data training samples, wherein the user data training samples comprise positive samples and negative samples, the sample labels of the positive samples represent users who do not need to be salvaged, and the sample labels of the negative samples represent users who need to be salvaged.
In this embodiment, each of the user data training samples may include personal information data, credit records, credit investigation data, and the like of the user. In the training sample, users who do not need to return for fishing are used as positive samples, and users who need to return for fishing are used as negative samples.
And step S120, performing feature extraction and feature screening on the user data training sample, and determining a modulus entering variable.
In this embodiment, feature extraction and feature screening may be performed on the training sample using a feature extraction and feature screening method in a conventional credit evaluation scheme to obtain the modelled variable. Meanwhile, the upper limit of the estimated passing rate can be determined and recorded as m according to the actual service requirement of refusing customers and fishing back.
Step S130, inputting the user data training sample and the model entering variable into a model to be trained for processing, and obtaining an output result of the user data training sample.
In this embodiment, the user data training sample and the model entering variable may be input into a model to be trained, and a prediction probability value obtained by the model is obtained as an output result.
Step S140, adjusting the model parameters of the model to be trained according to the sample label of the user data training sample, the output result and a preset objective function.
Wherein the loss function of the model to be trained comprises a first function item representing the loss function values of all positive samples and a second function item representing the loss function values of all negative samples, the first function item comprises a first penalty coefficient representing when a positive sample is predicted as a negative sample, the second function item comprises a second penalty coefficient representing when a negative sample is predicted as a positive sample, and the first penalty coefficient is greater than the second penalty coefficient.
Specifically, in this embodiment, the formula of the objective function is as follows:
wherein, yiThe authentic label representing the user data training sample i,
representing the probability that the prediction label of the user data training sample i is a positive sample;
data item
Is said first function term, w0Representing the first penalty coefficient;
the modulation coefficients of the loss function values calculated by different positive samples are represented, r represents the weight adjustment proportion, and r is more than or equal to 0;
data item
Is said second function term, w1Representing the second penalty coefficient;
and the modulation coefficients representing the loss function values calculated by different negative samples, r representing the weight adjustment proportion, and r being more than or equal to 0.
In some possible implementation manners, the loss coefficient w needs to be set in the process of actually constructing the model0、w1The weight adjustment proportion r is set as a hyper-parameter of the objective function item, and the first penalty coefficient, the second penalty coefficient and the weight adjustment proportion can be set according to the proportion of positive and negative samples in the user data training sample; or performing a grid search using the user data training samples to automatically determine the first penalty factor, second penalty factor, and weight adjustment scale.
In some possible implementation manners, the user data training sample and the pair of input variables may also be used, and the model to be trained is trained for multiple times to obtain different trained models; wherein different coefficients used for each training are used for the grid search. And then calculating the evaluation index value of each trained model through a preset evaluation function, and taking the trained model with the highest evaluation index value as the customer refusing and fishing back model after training. Wherein the evaluation index of the evaluation function comprises an area under the curve AUC index or a Kolmogorov-Smirnov index.
Optionally, in this embodiment, each trained model may also be tested through a test sample set, and when a final model is determined, a model in which an evaluation function value of a training sample is as good as possible and a difference value between the evaluation function value of the training sample and the evaluation function value of the test sample is within a set range may be used as a final customer refusing and salvage model.
Specifically, when the evaluation index value of each trained model is calculated through a preset evaluation function, for each trained model, the user data training samples may be sorted in the order from small to large according to the output result of each user data training sample, and a preset number of samples ranked at the top are selected to obtain the evaluation index value through the evaluation function calculation. In an exemplary embodiment, training samples in the front m ranking intervals after the model output result is ranked from small to large can be selected for calculating the evaluation index value, and m is the upper limit of the estimated passing rate in the actual service.
Referring to fig. 2, fig. 2 is a schematic view of an electronic device 100 according to an embodiment of the present disclosure, where the electronic device 100 may be, but is not limited to, a server, a personal computer, and other devices with data processing capability. The electronic device 100 includes a reject-bailing model training apparatus 110, a machine-readable storage medium 120, and a processor 130.
The machine-readable storage medium 120 and the processor 130 are electrically connected to each other directly or indirectly to enable data transmission or interaction. For example, the components may be electrically connected to each other via one or more communication buses or signal lines. The reject-and-salvage model training device 110 includes at least one software function module that can be stored in the machine-readable storage medium 120 in the form of software or firmware (firmware) or solidified in an Operating System (OS) of the electronic device 100. The processor 130 is configured to execute executable modules stored in the machine-readable storage medium 120, such as software function modules and computer programs included in the reject-fishing model training apparatus 110.
The machine-readable storage medium 120 may be, but is not limited to, a Random Access Memory (RAM), a Read Only Memory (ROM), a Programmable Read-Only Memory (PROM), an Erasable Read-Only Memory (EPROM), an electrically Erasable Read-Only Memory (EEPROM), and the like. The machine-readable storage medium 120 is used for storing a program, and the processor 130 executes the program after receiving an execution instruction.
The processor 130 may be an integrated circuit chip having signal processing capabilities. The Processor may be a general-purpose Processor, and includes a Central Processing Unit (CPU), a Network Processor (NP), and the like; but may also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA) or other programmable logic device, discrete gate or transistor logic device, discrete hardware components. The various methods, steps, and logic blocks disclosed in the embodiments of the present application may be implemented or performed. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.
Referring to fig. 3, the embodiment further provides a customer refusing and bailing model training device, which includes at least one functional module that can be stored in a machine-readable storage medium in a software form. Functionally, the customer refusal and salvage model training device may include a data obtaining module 111, a feature processing module 112, a model processing module 113, and a parameter adjusting module 114.
The data obtaining module 111 is configured to obtain a plurality of user data training samples, where the user data training samples include a positive sample and a negative sample, a sample label of the positive sample represents a user who does not need to be rewound, and a sample label of the negative sample represents a user who needs to be rewound.
In this embodiment, the data obtaining module 111 may be configured to execute step S110 shown in fig. 1, and for a detailed description of the data obtaining module 111, reference may be made to the description of step S110.
The feature processing module 112 is configured to perform feature extraction and feature screening on the user data training sample, and determine a modeling variable.
In this embodiment, the feature processing module 112 may be configured to execute step S120 shown in fig. 1, and for a detailed description of the feature processing module 112, reference may be made to the description of step S120.
The model processing module 113 is configured to input the user data training sample and the model entering variable into a model to be trained for processing, so as to obtain an output result of the user data training sample.
In this embodiment, the model processing module 113 may be configured to execute step S130 shown in fig. 1, and reference may be made to the description of step S130 for a detailed description of the model processing module 113.
The parameter adjusting module 114 is configured to adjust the model parameters of the model to be trained according to the sample label of the user data training sample, the output result, and a preset objective function.
In this embodiment, the parameter adjustment module 114 may be configured to execute step S140 shown in fig. 1, and for a detailed description of the parameter adjustment module 114, reference may be made to the description of step S140.
Wherein the loss function of the model to be trained comprises a first function item representing the loss function values of all positive samples and a second function item representing the loss function values of all negative samples, the first function item comprises a first penalty coefficient representing when a positive sample is predicted as a negative sample, the second function item comprises a second penalty coefficient representing when a negative sample is predicted as a positive sample, and the first penalty coefficient is greater than the second penalty coefficient.
In some possible implementations, the formula of the objective function is as follows:
wherein, yiThe authentic label representing the user data training sample i,
representing the probability that the prediction label of the user data training sample i is a positive sample;
data item
Is said first function term, w0Representing the first penalty coefficient;
the modulation coefficients of the loss function values calculated by different positive samples are represented, r represents the weight adjustment proportion, and r is more than or equal to 0;
data item
Is said second function term, w1Representing the second penalty coefficient;
and the modulation coefficients representing the loss function values calculated by different negative samples, r representing the weight adjustment proportion, and r being more than or equal to 0.
In some possible implementations, the apparatus further includes a coefficient adjustment module.
The coefficient adjusting module is used for setting the first penalty coefficient, the second penalty coefficient and the weight adjusting proportion according to the proportion of positive and negative samples in the user data training samples; or performing a grid search using the user data training samples to determine the first penalty coefficient, the second penalty coefficient, and the weight adjustment scale.
In summary, according to the training method, the training device and the training electronic device for the customer refusing and bailing model provided by the application, the loss function used in the model is set to include the first function item representing the loss function values of all positive samples and the second function item representing the loss function values of all negative samples, the first function item includes the first penalty coefficient representing that the positive samples are predicted to be the negative samples, the second function item includes the second penalty coefficient representing that the negative samples are predicted to be the positive samples, and the first penalty coefficient is greater than the second penalty coefficient. The accuracy of the model for predicting the positive sample is improved, the using effect of the model is improved, and the risk of the model for fishing back the client in the actual use is ensured to be within a controllable range. So, can be so that to "predict the negative sample with the positive sample" the attention degree be higher than "predict the negative sample as the positive sample" to improve the accuracy that the model was just sample prediction, promote the model result of use, ensure that the model returns the risk of dragging for the customer in the in-service use and in controllable range.
In the embodiments provided in the present application, it should be understood that the disclosed apparatus and method may be implemented in other ways. The apparatus embodiments described above are merely illustrative, and for example, the flowchart and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods and computer program products according to various embodiments of the present application. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.
In addition, functional modules in the embodiments of the present application may be integrated together to form an independent part, or each module may exist separately, or two or more modules may be integrated to form an independent part.
The functions, if implemented in the form of software functional modules and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present application or portions thereof that substantially contribute to the prior art may be embodied in the form of a software product stored in a storage medium and including instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present application. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and other various media capable of storing program codes.
It is noted that, herein, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, 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. The term "comprising", without further limitation, means that the element so defined is not excluded from the group consisting of additional identical elements in the process, method, article, or apparatus that comprises the element.
The above description is only for various embodiments of the present application, but the scope of the present application is not limited thereto, and any person skilled in the art can easily conceive of changes or substitutions within the technical scope of the present application, and all such changes or substitutions are included in the scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.
Claims (10)
1. A customer refusal and salvage model training method is characterized by comprising the following steps:
the method comprises the steps of obtaining a plurality of user data training samples, wherein the user data training samples comprise positive samples and negative samples, sample labels of the positive samples represent users who do not need to be salvaged, and sample labels of the negative samples represent users who need to be salvaged;
carrying out feature extraction and feature screening on the user data training sample, and determining a mode entering variable;
inputting the user data training sample and the model entering variable into a model to be trained for processing to obtain an output result of the user data training sample;
adjusting the model parameters of the model to be trained according to the sample label of the user data training sample, the output result and a preset target function;
wherein the loss function of the model to be trained comprises a first function item representing the loss function values of all positive samples and a second function item representing the loss function values of all negative samples, the first function item comprises a first penalty coefficient representing when a positive sample is predicted as a negative sample, the second function item comprises a second penalty coefficient representing when a negative sample is predicted as a positive sample, and the first penalty coefficient is greater than the second penalty coefficient.
2. The method of claim 1, wherein the objective function is formulated as follows:
wherein, yiThe authentic label representing the user data training sample i,
representing the probability that the prediction label of the user data training sample i is a positive sample;
data item
Is said first function term, w0Representing the first penalty coefficient;
the modulation coefficients of the loss function values calculated by different positive samples are represented, r represents the weight adjustment proportion, and r is more than or equal to 0;
data item
Is said second function term, w1Representing the second penalty coefficient;
and the modulation coefficients representing the loss function values calculated by different negative samples, r representing the weight adjustment proportion, and r being more than or equal to 0.
3. The method of claim 2, further comprising:
setting the first penalty coefficient, the second penalty coefficient and the weight adjustment proportion according to the proportion of positive and negative samples in the user data training samples; or
Performing a grid search using the user data training samples to determine the first penalty coefficient, second penalty coefficient, and weight adjustment scale.
4. The method of claim 1, further comprising:
training the model to be trained for multiple times by using the user data training sample and the model entering variable to obtain different trained models; wherein, different coefficients used in each training are subjected to grid search;
calculating the evaluation index value of each trained model through a preset evaluation function;
and taking the trained model with the highest evaluation index value as a customer refusing and fishing model after training.
5. The method according to claim 4, wherein the evaluation index of the merit function includes an area under the curve AUC index or a Kolmogorov-Smirnov index.
6. The method of claim 4, wherein the step of calculating an evaluation index value for each of the trained models by a predetermined evaluation function comprises:
and aiming at each trained model, sequencing the user data training samples according to the sequence of the output result of each user data training sample from small to large, and selecting a preset number of samples ranked at the top to obtain an evaluation index value through calculation of the evaluation function.
7. A guest refusing and salvage model training device is characterized by comprising:
the data acquisition module is used for acquiring a plurality of user data training samples, wherein the user data training samples comprise positive samples and negative samples, the sample labels of the positive samples represent users who do not need to be salvaged, and the sample labels of the negative samples represent users who need to be salvaged;
the characteristic processing module is used for carrying out characteristic extraction and characteristic screening on the user data training sample and determining a mode entering variable;
the model processing module is used for inputting the user data training sample and the model entering variable into a model to be trained for processing to obtain an output result of the user data training sample;
the parameter adjusting module is used for adjusting the model parameters of the model to be trained according to the sample labels of the user data training samples, the output result and a preset target function;
wherein the loss function of the model to be trained comprises a first function item representing the loss function values of all positive samples and a second function item representing the loss function values of all negative samples, the first function item comprises a first penalty coefficient representing when a positive sample is predicted as a negative sample, the second function item comprises a second penalty coefficient representing when a negative sample is predicted as a positive sample, and the first penalty coefficient is greater than the second penalty coefficient.
8. The apparatus of claim 7, wherein the objective function is formulated as follows:
wherein, yiThe authentic label representing the user data training sample i,
representing the probability that the prediction label of the user data training sample i is a positive sample;
data item
Is said first function term, w0Representing the first penalty coefficient;
the modulation coefficients of the loss function values calculated by different positive samples are represented, r represents the weight adjustment proportion, and r is more than or equal to 0;
data item
Is said second function term, w1Representing the second penalty coefficient;
and the modulation coefficients representing the loss function values calculated by different negative samples, r representing the weight adjustment proportion, and r being more than or equal to 0.
9. The apparatus of claim 8, further comprising:
the coefficient adjusting module is used for setting the first penalty coefficient, the second penalty coefficient and the weight adjusting proportion according to the proportion of positive and negative samples in the user data training samples; or performing a grid search using the user data training samples to determine the first penalty coefficient, the second penalty coefficient, and the weight adjustment scale.
10. An electronic device comprising a processor and a machine-readable storage medium having stored thereon machine-executable instructions that, when executed by the processor, implement the method of any of claims 1-6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110713287.2A CN113313582A (en) | 2021-06-25 | 2021-06-25 | Guest refusing and reflashing model training method and device and electronic equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110713287.2A CN113313582A (en) | 2021-06-25 | 2021-06-25 | Guest refusing and reflashing model training method and device and electronic equipment |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113313582A true CN113313582A (en) | 2021-08-27 |
Family
ID=77380447
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110713287.2A Pending CN113313582A (en) | 2021-06-25 | 2021-06-25 | Guest refusing and reflashing model training method and device and electronic equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113313582A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114331696A (en) * | 2021-12-31 | 2022-04-12 | 北京瑞莱智慧科技有限公司 | Risk assessment method, device and storage medium |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108830645A (en) * | 2018-05-31 | 2018-11-16 | 厦门快商通信息技术有限公司 | A kind of visitor's attrition prediction method and system |
| CN109919202A (en) * | 2019-02-18 | 2019-06-21 | 新华三技术有限公司合肥分公司 | Disaggregated model training method and device |
| CN110321422A (en) * | 2018-03-28 | 2019-10-11 | 腾讯科技（深圳）有限公司 | Method, method for pushing, device and the equipment of on-line training model |
| CN110728323A (en) * | 2019-10-12 | 2020-01-24 | 中诚信征信有限公司 | Target type user identification method and device, electronic equipment and storage medium |
| CN111310814A (en) * | 2020-02-07 | 2020-06-19 | 支付宝(杭州)信息技术有限公司 | Method and device for training business prediction model by utilizing unbalanced positive and negative samples |
| CN111966831A (en) * | 2020-08-18 | 2020-11-20 | 创新奇智(上海)科技有限公司 | Model training method, text classification device and network model |
| CN112686218A (en) * | 2021-03-09 | 2021-04-20 | 北京世纪好未来教育科技有限公司 | Training method and device of text detection model, readable storage medium and equipment |
| CN112819024A (en) * | 2020-07-10 | 2021-05-18 | 腾讯科技（深圳）有限公司 | Model processing method, user data processing method and device and computer equipment |
-
2021
- 2021-06-25 CN CN202110713287.2A patent/CN113313582A/en active Pending
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110321422A (en) * | 2018-03-28 | 2019-10-11 | 腾讯科技（深圳）有限公司 | Method, method for pushing, device and the equipment of on-line training model |
| CN108830645A (en) * | 2018-05-31 | 2018-11-16 | 厦门快商通信息技术有限公司 | A kind of visitor's attrition prediction method and system |
| CN109919202A (en) * | 2019-02-18 | 2019-06-21 | 新华三技术有限公司合肥分公司 | Disaggregated model training method and device |
| CN110728323A (en) * | 2019-10-12 | 2020-01-24 | 中诚信征信有限公司 | Target type user identification method and device, electronic equipment and storage medium |
| CN111310814A (en) * | 2020-02-07 | 2020-06-19 | 支付宝(杭州)信息技术有限公司 | Method and device for training business prediction model by utilizing unbalanced positive and negative samples |
| CN112819024A (en) * | 2020-07-10 | 2021-05-18 | 腾讯科技（深圳）有限公司 | Model processing method, user data processing method and device and computer equipment |
| CN111966831A (en) * | 2020-08-18 | 2020-11-20 | 创新奇智(上海)科技有限公司 | Model training method, text classification device and network model |
| CN112686218A (en) * | 2021-03-09 | 2021-04-20 | 北京世纪好未来教育科技有限公司 | Training method and device of text detection model, readable storage medium and equipment |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114331696A (en) * | 2021-12-31 | 2022-04-12 | 北京瑞莱智慧科技有限公司 | Risk assessment method, device and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107103171B (en) | Modeling method and device of machine learning model | |
| CN111507831A (en) | Credit risk automatic assessment method and device | |
| CN112102073A (en) | Credit risk control method and system, electronic device and readable storage medium | |
| CN110930218B (en) | Method and device for identifying fraudulent clients and electronic equipment | |
| CN110866832A (en) | Risk control method, system, storage medium and computing device | |
| CN112488716A (en) | Abnormal event detection system | |
| CN108197795B (en) | Malicious group account identification method, device, terminal and storage medium | |
| CN111275338A (en) | Method, device, equipment and storage medium for judging enterprise fraud behaviors | |
| CN111435357A (en) | Gas station customer churn prediction method, gas station customer churn prediction device, electronic equipment and storage medium | |
| CN113313582A (en) | Guest refusing and reflashing model training method and device and electronic equipment | |
| CN111126264A (en) | Image processing method, device, equipment and storage medium | |
| CN114004691A (en) | Line scoring method, device, equipment and storage medium based on fusion algorithm | |
| CN112348685A (en) | Credit scoring method, device, equipment and storage medium | |
| CN113112347A (en) | Determination method of hasty collection decision, related device and computer storage medium | |
| CN116976995A (en) | Multi-target recommendation processing method and device | |
| CN110852322A (en) | Method and device for determining region of interest | |
| CN113065892B (en) | Information pushing method, device, equipment and storage medium | |
| CN111899092B (en) | Business data screening method and device based on two-channel model | |
| CN113554099A (en) | Method and device for identifying abnormal commercial tenant | |
| CN110766544A (en) | Credit risk detection method and device, storage medium and electronic device | |
| CN113537666B (en) | Evaluation model training method, evaluation and business auditing method, device and equipment | |
| CN112258093B (en) | Data processing method and device for risk level, storage medium and electronic equipment | |
| CN114140140B (en) | Scene screening method, device and equipment | |
| CN111598256B (en) | Processing method and device for default purchase behavior of target client | |
| CN109947927B (en) | Information processing method and device |
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 |