Model training method and device based on rejection inference, equipment and storage medium

Abstract

The embodiment of the invention provides a model training method and device based on rejection inference, equipment and a storage medium. The model training method based on rejection inference comprises the steps of training a plurality of different basic models based on the same training data set, predicting the same target data set based on the trained basic models to obtain a meta-feature matrix, training meta-models based on the meta-feature matrix, and training the meta-models based on the meta-feature matrix. And predicting a plurality of rejected samples based on each trained basic model and the trained meta-model to obtain negative samples in all rejected samples, adding negative sample data into the training data set, and returning to execute training of a plurality of different basic models based on the same training data set. According to the model training method based on rejection inference, the negative samples can be expanded, and the number of the negative samples in sample training is increased, so that the consistency and robustness of the model can be improved.

Description

Technical Field

[0001] Embodiments of the present disclosure relate to the technical field of model training, and in particular, to a model training method, apparatus, device, and storage medium based on rejection inference. Background Art

[0002] In the field of home improvement installment loans, user access prediction is an important link that financial institutions pay attention to in order to ensure loan security and compliance. Ensuring that the borrower's repayment ability and credit status meet the requirements is a key task in maintaining the asset quality of financial institutions.

[0003] Due to the particularity of home improvement installment scenarios, the samples available for training are relatively limited. This scarcity makes it challenging for traditional machine learning methods to achieve accurate user access predictions, especially when multi-dimensional financial information needs to be considered. There are relatively more customers with good credit and relatively fewer customers with bad credit, which leads to the model being directly trained on samples with fewer negative samples, which is more unstable and difficult to ensure the consistency and robustness of the model. Summary of the invention

[0004] The embodiments described in this article provide a model training method, apparatus, device and storage medium based on rejection inference, which can expand negative samples and increase the number of negative samples in sample training, thereby improving the consistency and robustness of the model.

[0005] In a first aspect, the present disclosure provides a model training method based on rejection inference, comprising:

[0006] Based on the same training data set, multiple different basic models are trained; based on each of the trained basic models, the same target data set is predicted to obtain a meta-feature matrix, wherein the target data set is a verification data set and / or a test data set, and a column in the meta-feature matrix is ​​a prediction result of a trained basic model on the target data set; based on the meta-feature matrix, a meta-model is trained; based on each of the trained basic models and the trained meta-model, multiple rejection samples are predicted to obtain negative samples among all the rejection samples; the negative sample data is added to the training data set, and the step of training multiple different basic models based on the same training data set is returned to execute.

[0007] In some embodiments of the present disclosure, before training a plurality of different basic models based on the same training data set, the method further includes:

[0008] The data information of all users who have applied for the service and the labels corresponding to each user are obtained, wherein the data information includes personal information, service information, credit information and repayment performance information; target features are extracted from the data information of each user to obtain the training data set.

[0009] In some embodiments of the present disclosure, obtaining data information of all users who have participated in applying for the service and tags corresponding to each user includes:

[0010] Collect the data information of all the users who have participated in the application business; and determine the label corresponding to each user based on the repayment performance information of each user.

[0011] In some embodiments of the present disclosure, extracting target features from the data information of each user to obtain the training data set includes:

[0012] Processing missing values, abnormal values ​​and duplicate values ​​in the data information of all the users; determining the target feature of each user based on the processed data information of each user; determining the training data set and the target data set based on the target features of all the users.

[0013] In some embodiments of the present disclosure, determining the target feature of each user based on the processed data information of each user includes:

[0014] Based on the processed data information of each of the users, feature derivation is performed to obtain initial features; the initial features of each of the users are encoded to obtain intermediate features that can be recognized by the model; and the target features of each of the users are screened out from the intermediate features of each of the users.

[0015] In some embodiments of the present disclosure, the multiple different base models are models of different algorithms, and / or models of different hyperparameter configurations of the same algorithm.

[0016] In some embodiments of the present disclosure, the predicting of multiple rejection samples based on the trained basic models and the trained meta-model to obtain negative samples among all the rejection samples includes:

[0017] Input all rejected sample data into all the trained basic models to obtain the meta-feature matrix corresponding to all the rejected samples; input the meta-feature matrix corresponding to all the rejected samples into the trained meta-model to obtain the predicted probability corresponding to all the rejected samples, and the predicted probability is used to characterize the probability that the corresponding rejected sample is a positive category; determine the rejected sample whose predicted probability is less than the probability threshold as the negative sample

[0018] In a second aspect, the present disclosure provides a model training device based on rejection inference, comprising:

[0019] The training module is used to train multiple different base models based on the same training dataset.

[0020] A prediction module is used to predict the same target data set based on the trained basic models to obtain a meta-feature matrix, wherein the target data set is a verification data set and / or a test data set, and a column in the meta-feature matrix is ​​a prediction result of a trained basic model on the target data set.

[0021] The training module is also used to train the meta-model based on the meta-feature matrix.

[0022] The prediction module is further used to predict multiple rejection samples based on the trained basic models and the trained meta-model.

[0023] The determination module is used to obtain negative samples from all the rejected samples and add the negative sample data to the training data set.

[0024] In a third aspect, the present disclosure provides an electronic device, comprising a processor, wherein the processor is configured to execute a computer program stored in a memory, wherein the computer program, when executed by the processor, implements the steps of any one of the methods provided in the first aspect.

[0025] In a fourth aspect, the present disclosure provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of any method provided in the first aspect are implemented.

[0026] In the technical solution provided by the present invention, multiple different basic models are trained based on the same training data set, the same target data set is predicted based on the trained basic models to obtain a meta-feature matrix, a meta-model is trained based on the meta-feature matrix, multiple rejection samples are predicted based on the trained basic models and the trained meta-model to obtain negative samples among all the rejection samples, the negative sample data is added to the training data set, and the training of multiple different basic models based on the same training data set is returned to expand the negative samples and increase the number of negative samples in the sample training, thereby improving the consistency and robustness of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the drawings of the embodiments will be briefly described below. It should be noted that the drawings described below only relate to some embodiments of the present disclosure, but are not intended to limit the present disclosure, wherein:

[0028] Figure 1 A flowchart of a model training method based on rejection inference provided in an embodiment of the present disclosure;

[0029] Figure 2A flowchart of another model training method based on rejection inference provided in an embodiment of the present disclosure;

[0030] Figure 3 A flowchart of another model training method based on rejection inference provided in an embodiment of the present disclosure;

[0031] Figure 4 A flowchart of another model training method based on rejection inference provided in an embodiment of the present disclosure;

[0032] Figure 5 A schematic diagram of the structure of a model training device based on rejection inference provided in an embodiment of the present disclosure;

[0033] Figure 6 A schematic diagram of the structure of an electronic device provided in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0034] In order to make the purpose, technical solution and advantages of the embodiments of the present disclosure clearer, the technical solution of the embodiments of the present disclosure will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present disclosure, rather than all of the embodiments. Based on the described embodiments of the present disclosure, all other embodiments obtained by those skilled in the art without creative work also fall within the scope of protection of the present disclosure.

[0035] Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by those skilled in the art to which the subject matter of the present disclosure belongs. It will be further understood that terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with their meanings in the context of the specification and the relevant art, and will not be interpreted in an idealized or overly formal form unless otherwise explicitly defined herein. In addition, terms such as "first" and "second" are only used to distinguish one component (or a portion of a component) from another component (or another portion of a component).

[0036] The method provided in the present disclosure is applied to scenarios with a small number of negative samples, for example, in the scenario of home improvement installment loans, there are relatively more customers with good credit, i.e., positive samples, and relatively fewer customers with bad credit, i.e., negative samples. The application business in the present disclosure may be a home improvement installment loan application business. All customers who have participated in the application business in the present disclosure are customers whose applications have been rejected and customers whose applications have been approved. The rejected samples in the present disclosure are customers whose applications have been rejected.

[0037] Figure 1 A flowchart of a model training method based on rejection inference provided in an embodiment of the present disclosure is shown in FIG. Figure 1As shown, the specific steps of the model training method based on rejection inference include:

[0038] S101, training multiple different basic models based on the same training data set.

[0039] Exemplarily, the multiple different base models are models of different algorithms, and / or models of different hyperparameter configurations of the same algorithm, to ensure diversity in learning tasks. The multiple different base models can be models of different algorithms. For example, the multiple different base models include a first base model, a second base model, and a third base model, wherein the first base model is a random forest model, the second base model is a support vector machine model, and the third base model is an extreme gradient boosting tree model.

[0040] In some other embodiments, the multiple different base models may be models of different hyperparameter configurations of the same algorithm. For example, the multiple different base models include a first base model, a fourth base model, and a fifth base model, wherein the first base model is a random forest model of a first hyperparameter configuration, the fourth base model is a random forest model of a second hyperparameter configuration, and the fifth base model is a random forest model of a third hyperparameter configuration.

[0041] In some other implementations, some of the multiple different basic models are models of different algorithms, and some are models of different hyperparameter configurations of the same algorithm. For example, the multiple different basic models include the first basic model, the second basic model, the third basic model, the fourth basic model, and the fifth basic model in the above embodiment.

[0042] It should be noted that this embodiment only uses the first basic model, the second basic model, the third basic model, the fourth basic model and the fifth basic model to exemplify multiple different basic models, and does not limit the number of basic models. The random forest model, the support vector machine model and the extreme gradient boosting tree model are all exemplary descriptions of the basic models, and the specific types of basic models in practical applications are not limited thereto.

[0043] The training data set includes multiple sets of training data, each set of training data includes multiple different target features. Among them, a set of training data is the relevant data of a user who has participated in the application service. Each time the basic model is trained, a set of training data and its corresponding labels are input into the basic model. The basic model predicts the set of training data input at the same time, and obtains the loss value based on the prediction result and label of the set of training data. The parameters of the basic model are adjusted in the direction of reducing the loss value, thereby achieving the purpose of training the basic model.

[0044] Based on the above training process, a set of training data can be input into multiple different basic models respectively, and multiple different basic models can be trained to obtain multiple trained basic models.

[0045] S102, predicting the same target data set based on the trained basic models to obtain a meta-feature matrix.

[0046] The target data set is a validation data set and / or a test data set, and a column in the meta-feature matrix is ​​a prediction result of a trained basic model on the target data set.

[0047] Exemplarily, the target data set includes multiple groups of target data, each group of target data includes multiple different target features, and the types and quantities of target features in a group of target data and a group of training data are consistent. The target data set can be a verification data set, or a test data set, or both a verification data set and a test data set.

[0048] If the target dataset is a validation dataset, then the target data is validation data. If the target dataset is a test dataset, then the target data is test data. If the target dataset is both a validation dataset and a test dataset, then the target data is either test data or validation data.

[0049] A set of target data is input into each trained basic model, and each trained basic model makes a prediction for the set of target data input at the same time to obtain a prediction result. In this way, all target data in the target data set are input into each trained basic model, and each trained basic model makes a prediction for all target data in the target data set to obtain multiple prediction results. The multiple prediction results of each trained basic model for the target data set are used as a column matrix, and the column matrices corresponding to all trained basic models can be constructed into a meta-feature matrix.

[0050] For example, the target data set includes first target data, second target data and third target data, and the multiple different basic models include the first basic model, the second basic model and the third basic model. All target data in the target data set are input into the trained first basic model, the trained first basic model predicts the first target data to obtain a first prediction result, the trained first basic model predicts the second target data to obtain a second prediction result, and the trained first basic model predicts the third target data to obtain a third prediction result.

[0051] All target data in the target data set are input into the trained second basic model, the trained second basic model predicts the first target data to obtain the fourth prediction result, the trained second basic model predicts the second target data to obtain the fifth prediction result, and the trained second basic model predicts the third target data to obtain the sixth prediction result. All target data in the target data set are input into the trained third basic model, the trained third basic model predicts the first target data to obtain the seventh prediction result, the trained third basic model predicts the second target data to obtain the eighth prediction result, and the trained third basic model predicts the third target data to obtain the ninth prediction result.

[0052] Thus, the first basic model corresponds to the column matrix The second basic model corresponds to the column matrix The third basic model corresponds to the column matrix Then the element feature matrix is

[0053] S103, training a meta-model based on the meta-feature matrix.

[0054] Exemplarily, the meta-model can be a simple linear model or a more complex model, such as a decision tree model or a random forest model. A row of prediction results in the meta-feature matrix corresponds to a label. Each time the meta-model is trained, a row of prediction results and their corresponding labels are input into the meta-model. The meta-model predicts the row of prediction results, and a loss value is obtained based on the prediction results and labels of the row of prediction results. The parameters of the meta-model are adjusted in the direction of reducing the loss value to achieve the purpose of training the meta-model, thereby obtaining a trained meta-model.

[0055] S104, predicting multiple rejection samples based on the trained basic models and the trained meta-model to obtain negative samples among all the rejection samples.

[0056] Exemplarily, users whose service applications are rejected are screened out from all users, and the users whose service applications are rejected are rejected samples, and the multiple target features corresponding to the rejected samples are rejected sample data.

[0057] As a specific description of a possible implementation method when executing S104, Figure 2 As shown, Figure 2 A flowchart of another model training method based on rejection inference provided in an embodiment of the present disclosure.

[0058] S201, input all rejected sample data into all trained basic models to obtain the meta-feature matrix corresponding to all rejected samples.

[0059] Exemplarily, each rejected sample data is input into all the trained basic models, and each trained basic model predicts each rejected sample. The prediction results of each trained basic model for all rejected samples are used as a column of the meta-feature matrix, thereby obtaining the meta-feature matrix corresponding to all rejected samples.

[0060] For example, all rejected samples include the first rejected sample and the second rejected sample, and correspondingly, all rejected sample data include the first rejected sample data and the second rejected sample data. All basic models include the first basic model, the second basic model and the third basic model, all rejected samples are input into the trained first basic model, the trained first basic model predicts the first rejected sample to obtain the tenth prediction result, and the trained first basic model predicts the second rejected sample to obtain the eleventh prediction result.

[0061] All rejected samples are input into the trained second basic model, and the trained second basic model predicts the twelfth prediction result for the first rejected sample, and the trained second basic model predicts the thirteenth prediction result for the first rejected sample. All rejected samples are input into the trained third basic model, and the trained third basic model predicts the fourteenth prediction result for the first rejected sample, and the trained third basic model predicts the fifteenth prediction result for the second rejected sample.

[0062] Thus, based on the above embodiment, the meta-feature matrix corresponding to all rejected samples is obtained as follows:

[0063] S202, inputting the meta-feature matrices corresponding to all rejected samples into the trained meta-model to obtain the prediction probabilities corresponding to all rejected samples.

[0064] Among them, the predicted probability is used to characterize the probability that the corresponding rejected sample is a positive class.

[0065] Exemplarily, the meta-feature matrix corresponding to all rejected samples is input into the trained meta-model, and the trained meta-model predicts each row of preset results in the meta-feature matrix to obtain the predicted probability of each rejected sample. The predicted probability is the probability that the corresponding rejected sample is good, that is, the greater the predicted probability, the greater the probability that the corresponding rejected sample is a positive sample.

[0066] For example, By inputting into the trained meta-model, the first predicted probability P1 of the first rejected sample and the second predicted probability P2 of the second rejected sample can be obtained.

[0067] S203, determining the rejected samples whose predicted probabilities are less than the probability threshold as negative samples.

[0068] For example, since the larger the predicted probability, the larger the probability that the corresponding rejected sample is a positive sample, a probability threshold can be set to determine the samples corresponding to the predicted probability greater than or equal to the probability threshold as positive samples, and the samples corresponding to the predicted probability less than the probability threshold as negative samples. In this way, negative samples can be screened out from rejected samples.

[0069] For example, based on the above embodiment, if the first predicted probability P1 is less than the probability threshold, the first rejected sample is a negative sample, and if the first predicted probability P1 is greater than or equal to the probability threshold, the first rejected sample is a positive sample. If the second predicted probability P2 is less than the probability threshold, the second rejected sample is a negative sample. If the second predicted probability P2 is greater than or equal to the probability threshold, the second rejected sample is a positive sample.

[0070] S105, adding negative sample data to the training data set.

[0071] Return to execute S101.

[0072] For example, negative sample data is added to the training data set to expand the negative samples in the training samples. The basic model and meta-model are further trained based on the newly added negative sample data in the training data set, which can increase the number of negative samples in the sample training, thereby improving the consistency and robustness of the model.

[0073] In the disclosed embodiment, multiple different basic models are trained based on the same training data set, the same target data set is predicted based on the trained basic models to obtain a meta-feature matrix, a meta-model is trained based on the meta-feature matrix, multiple rejection samples are predicted based on the trained basic models and the trained meta-model to obtain negative samples among all the rejection samples, the negative sample data is added to the training data set, and the training of multiple different basic models based on the same training data set is returned to execute, so that the negative samples can be expanded and the number of negative samples in the sample training can be increased, thereby improving the consistency and robustness of the model.

[0074] In some embodiments, Figure 3 A flowchart of another model training method based on rejection inference provided in an embodiment of the present disclosure is provided. Figure 3 for Figure 2 Based on the illustrated embodiment, before executing S101, the specific steps of the model training method based on rejection inference also include:

[0075] S301, obtaining data information of all users who have participated in applying for the service and tags corresponding to each user.

[0076] Among them, data information includes personal information, business information, credit information and repayment performance information.

[0077] Exemplarily, personal information includes information such as age, income and family status, business information includes information such as cooperative institutions and loan information, credit information includes information such as credit history, debt status and repayment record, and repayment performance information is historical data information, including information such as on-time repayment and overdue repayment. In other embodiments, data information may also include other information, such as third-party data.

[0078] Each user corresponds to a user identification code, and a user identification code corresponds to a set of data information. The data information of all users who have applied for the service is collected, and the label corresponding to each user is determined based on the repayment performance information in the data information corresponding to each user. The label is good or bad. For example, users with serious overdue payments are defined as bad, and users with normal repayments or fewer overdue days are defined as good.

[0079] S302, extracting target features from the data information of each user to obtain a training data set.

[0080] For example, as a specific description of a possible implementation method when executing S302, Figure 4 As shown, Figure 4 A flowchart of another model training method based on rejection inference is provided in an embodiment of the present disclosure.

[0081] S401, processing missing values, abnormal values ​​and duplicate values ​​in the data information of all users.

[0082] For example, the missing values ​​in the data information of all users are analyzed to understand the distribution of missing values ​​in each feature and determine whether the missing values ​​are caused by data collection problems or other problems. Depending on the reasons, a single value, mean, or maximum value can be selected for filling. For example, when the missing values ​​are greater than 50% due to data collection problems, a single value can be used for filling.

[0083] Detect outliers in all users' data information based on statistics and visualization tools to ensure data accuracy and consistency. For detected outliers, you can choose to delete the outliers and replace them with missing values, or you can delete the outliers and use other methods such as interpolation to fill them.

[0084] Check all users' data information for duplicate records, for example, check whether user identification codes are duplicated. For duplicate values, you can choose to keep the first occurrence, delete all duplicate records, keep the last occurrence, or keep all records, depending on your business needs.

[0085] S402: Determine target features of each user based on the processed data information of each user.

[0086] Exemplarily, feature derivation is performed based on the processed data information of each user to obtain initial features of each user. For example, feature derivation is performed based on the credit information of each user to obtain initial features of multiple accounts, credit cards and loans.

[0087] Encode the initial features of each user to obtain intermediate features that the model can recognize. For example, one-hot encode the initial features to convert them into binary representations so that the model can interpret them correctly.

[0088] Filter out the target features of each user from the intermediate features of each user. For example, through correlation analysis, IV value or model, combined with expert experience, filter out the target features with discrimination and business explanation from the intermediate features of each user. Usually, the number of intermediate features is large, and the number of target features obtained through feature screening is relatively small, which can reasonably reduce the amount of training data.

[0089] S403: Determine a training data set and a target data set based on target features of all users.

[0090] Exemplarily, the target features of all users are divided into two data sets in proportion, for example, the target features of all users can be divided according to 7:3, and the target features of 70% of the users constitute the training data set, and the target features of 30% of the users constitute the target data set.

[0091] The present disclosure also provides a model training device based on rejection inference, Figure 5 is a schematic diagram of a model training device based on rejection inference provided by an embodiment of the present disclosure, such as Figure 5 As shown, the model training device based on rejection inference includes:

[0092] The training module 110 is used to train multiple different basic models based on the same training data set.

[0093] The prediction module 120 is used to predict the same target data set based on the trained basic models to obtain a meta-feature matrix, wherein the target data set is a validation data set and / or a test data set, and a column in the meta-feature matrix is ​​a prediction result of a trained basic model on the target data set.

[0094] The training module 110 is further used to train the meta-model based on the meta-feature matrix.

[0095] The prediction module 120 is further used to predict a plurality of rejection samples based on the trained basic models and the trained meta-model.

[0096] The determination module 130 is used to obtain negative samples from all the rejected samples and add the negative sample data to the training data set.

[0097] In some embodiments, the rejection inference-based model training apparatus further comprises:

[0098] The acquisition module is used to acquire the data information of all users who have applied for the service and the labels corresponding to each user. The data information includes personal information, service information, credit information and repayment performance information.

[0099] The determination module 130 is further configured to extract target features from the data information of each user to obtain the training data set.

[0100] In some embodiments, the acquisition module is further used to collect the data information of all the users who have participated in the application business, and determine the label corresponding to each of the users based on the repayment performance information of each of the users.

[0101] In some embodiments, the determination module 130 is further used to process missing values, abnormal values ​​and duplicate values ​​in the data information of all the users, determine the target features of each user based on the processed data information of each user, and determine the training data set and the target data set based on the target features of all the users.

[0102] In some embodiments, the determination module 130 is further used to derive features based on the processed data information of each of the users to obtain initial features, encode the initial features of each of the users to obtain intermediate features recognizable by the model, and filter out the target features of each of the users from the intermediate features of each of the users.

[0103] In some embodiments, the multiple different base models are models of different algorithms, and / or models of the same algorithm with different hyperparameter configurations.

[0104] In some embodiments, the determination module 130 is further used to input all the rejected sample data into all the trained basic models to obtain the meta-feature matrix corresponding to all the rejected samples; input the meta-feature matrix corresponding to all the rejected samples into the trained meta-model to obtain the predicted probability corresponding to all the rejected samples, and the predicted probability is used to characterize the probability that the corresponding rejected sample is a positive category; and determine the rejected sample whose predicted probability is less than the probability threshold as the negative sample.

[0105] The device provided in the embodiment of the present disclosure is used to execute the steps provided in any of the above method embodiments, has functional modules corresponding to the method embodiments, and has the beneficial effects of the method embodiments, which will not be repeated here.

[0106] The present disclosure further provides an electronic device, comprising: a processor, wherein the processor is used to execute a computer program stored in a memory, and the computer program implements the steps of the method embodiment of the present disclosure when executed by the processor.

[0107] Figure 6 A schematic diagram of the structure of an electronic device provided by the present disclosure is shown in FIG. Figure 6 A block diagram of an exemplary electronic device suitable for implementing the embodiments of the present disclosure is shown. Figure 6 The electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0108] like Figure 6 As shown, the electronic device 12 is in the form of a general purpose computing device. The components of the electronic device 12 may include, but are not limited to: one or more processors 16, a system memory 28, and a bus 18 connecting various system components (including the system memory 28 and the processor 16).

[0109] Bus 18 represents one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, a processor or a local bus using any of a variety of bus architectures. By way of example, these architectures include, but are not limited to, an Industry Standard Architecture (ISA) bus, a Micro Channel Architecture (MAC) bus, an Enhanced ISA bus, a Video Electronics Standards Association (VESA) local bus, and a Peripheral Component Interconnect (PCI) bus.

[0110] The electronic device 12 typically includes a variety of computer system readable media. These media can be any media that can be accessed by the electronic device 12, including volatile and non-volatile media, removable and non-removable media.

[0111] The system memory 28 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 30 and / or cache memory 32. The electronic device 12 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, the storage system 34 may be used to read and write non-removable, non-volatile magnetic media (commonly referred to as "hard drives"). A disk drive for reading and writing removable non-volatile disks (such as "floppy disks"), and an optical disk drive for reading and writing removable non-volatile optical disks (such as CD-ROMs, DVD-ROMs, or other optical media) may be provided. In these cases, each drive may be connected to the bus 18 via one or more data media interfaces. The system memory 28 may include at least one program product having a set (e.g., at least one) of program modules that are configured to perform the functions of the various embodiments of the present disclosure.

[0112] A program / utility 40 having a set (at least one) of program modules 42 may be stored, for example, in system memory 28, such program modules 42 including, but not limited to, an operating system, one or more application programs, other program modules, and program data, each of which or some combination thereof may include an implementation of a network environment. Program modules 42 generally perform the functions and / or methods of the embodiments described in the embodiments of the present disclosure.

[0113] The processor 16 executes various functional applications and data processing by running at least one program among a plurality of programs stored in the system memory 28 , such as implementing the method embodiment provided in the embodiment of the present disclosure.

[0114] The present disclosure also provides a computer-readable storage medium having a computer program stored thereon, and the computer program implements the steps of the above method embodiment when executed by a processor.

[0115] Any combination of one or more computer-readable media can be used. Computer-readable media can be computer-readable signal media or computer-readable storage media. Computer-readable storage media can be, for example, but not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or devices, or any combination of the above. More specific examples (non-exhaustive list) of computer-readable storage media include: electrical connections with one or more wires, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the above. In this document, computer-readable storage media can be any tangible medium containing or storing a program, which can be used by an instruction execution system, device or device or used in combination with it.

[0116] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, which carry computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0117] The program code embodied on the computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0118] Computer program code for performing the operations of the present disclosure may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages, such as Java, Smalltalk, C++, and conventional procedural programming languages, such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN) domain, or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0119] The present disclosure also provides a computer program product. When the computer program product is run on a computer, the computer is enabled to execute the steps of implementing the above method embodiments.

[0120] The flow chart and block diagram in the accompanying drawings show the possible architecture, function and operation of the device and method according to multiple embodiments of the present disclosure. In this regard, each square box in the flow chart or block diagram can represent a part of a module, a program segment or an instruction, and a part of the module, a program segment or an instruction includes one or more executable instructions for realizing the specified logical function. In some alternative implementations, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two continuous square boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of special hardware and computer instructions.

[0121] Unless the context clearly indicates otherwise, the singular form of the words used herein and in the appended claims includes the plural and vice versa. Thus, when referring to the singular, the plural form of the corresponding term is generally included. Similarly, the words "comprise" and "include" are to be interpreted as inclusive rather than exclusive. Likewise, the terms "include" and "or" should be interpreted as inclusive unless such interpretation is expressly prohibited herein. Where the term "example" is used herein, particularly when it is located after a group of terms, the "example" is merely exemplary and illustrative and should not be considered exclusive or comprehensive.

[0122] Further aspects and scopes of adaptability become apparent from the description provided herein. It should be understood that various aspects of the present disclosure can be implemented individually or in combination with one or more other aspects. It should also be understood that the description and specific embodiments herein are intended for purposes of illustration only and are not intended to limit the scope of the present disclosure.

[0123] Several embodiments of the present disclosure are described in detail above, but it is obvious that those skilled in the art can make various modifications and variations to the embodiments of the present disclosure without departing from the spirit and scope of the present disclosure. The protection scope of the present disclosure is defined by the attached claims.

Claims

1. A model training method based on rejection inference, characterized in that: include: Train multiple different base models based on the same training dataset; Predicting the same target data set based on each of the trained basic models to obtain a meta-feature matrix, wherein the target data set is a verification data set and / or a test data set, and a column in the meta-feature matrix is ​​a prediction result of a trained basic model on the target data set; Training a meta-model based on the meta-feature matrix; Predicting a plurality of rejected samples based on the trained basic models and the trained meta-model to obtain negative samples among all the rejected samples; The negative sample data is added to the training data set, and the training of multiple different basic models based on the same training data set is returned to be executed.

2. The method according to claim 1, characterized in that Before training a plurality of different basic models based on the same training data set, the method further includes: Obtaining data information of all users who have applied for the service and labels corresponding to each user, including personal information, service information, credit information and repayment performance information; Target features are extracted from the data information of each of the users to obtain the training data set.

3. The method according to claim 2, characterized in that The acquisition of data information of all users who have applied for the service and the labels corresponding to each user includes: Collecting the data information of all the users who have participated in the application service; Based on the repayment performance information of each of the users, a label corresponding to each of the users is determined.

4. The method according to claim 2, characterized in that: Extracting target features from the data information of each user to obtain the training data set includes: Processing missing values, abnormal values ​​and duplicate values ​​in the data information of all the users; Determining the target feature of each of the users based on the processed data information of each of the users; The training data set and the target data set are determined based on the target features of all the users.

5. The method according to claim 4, characterized in that Determining the target feature of each user based on the processed data information of each user includes: Deriving features based on the processed data information of each of the users to obtain initial features; Encoding the initial features of each of the users to obtain intermediate features that can be recognized by the model; The target feature of each user is selected from the intermediate features of each user.

6. The method according to any one of claims 1 to 5, characterized in that: The multiple different basic models are models of different algorithms, and / or models of the same algorithm with different hyperparameter configurations.

7. The method according to any one of claims 1 to 5, characterized in that: The prediction of multiple rejection samples based on the trained basic models and the trained meta-model to obtain negative samples among all the rejection samples includes: Inputting all rejected sample data into all the trained basic models to obtain the meta-feature matrix corresponding to all the rejected samples; Inputting the meta-feature matrix corresponding to all the rejected samples into the trained meta-model to obtain the predicted probabilities corresponding to all the rejected samples, wherein the predicted probabilities are used to characterize the probabilities that the corresponding rejected samples are positive categories; The rejected samples whose predicted probabilities are less than a probability threshold are determined as negative samples.

8. A model training device based on rejection inference, characterized in that: include: The training module is used to train multiple different basic models based on the same training data set; A prediction module, used to predict the same target data set based on each of the trained basic models to obtain a meta-feature matrix; wherein the target data set is a verification data set and / or a test data set, and a column in the meta-feature matrix is ​​a prediction result of a trained basic model on the target data set; The training module is further used to train the meta-model based on the meta-feature matrix; The prediction module is further used to predict multiple rejection samples based on the trained basic models and the trained meta-model; The determination module is used to obtain negative samples from all the rejected samples and add the negative sample data to the training data set.

9. An electronic device, characterized in that: include: A processor, wherein the processor is used to execute a computer program stored in a memory, wherein the computer program, when executed by the processor, implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images