Business prediction model training method and device and nonvolatile storage medium

Abstract

The invention discloses a business prediction model training method and device and a nonvolatile storage medium. The method comprises the following steps: acquiring historical business data of a business to be predicted, and determining a first class of training samples and a second class of training samples from the historical business data, the second class of training samples being more than the first class of training samples; randomly dividing the second type of training samples into a plurality of first sample sets, and fusing the plurality of first sample sets with the first type of training samples to obtain a plurality of second sample sets; and training a business prediction model according to the plurality of second sample sets, the business prediction model comprising a plurality of dichotomous base models, the number of the dichotomous base models being the same as the number of the second sample sets. According to the method and the device, the technical problem that the business result cannot be effectively predicted in a specific application scene due to low bipartite model training efficiency in the specific application scene in related technologies is solved.

Description

Technical Field

[0001] The present application relates to the field of electronic digital data processing, and more specifically, to a business prediction model training method, device and non-volatile storage medium. Background Art

[0002] When training a binary base model, it is usually required that the ratio between positive and negative samples is close to 1:1 to ensure training efficiency and training results. However, in actual application scenarios, there may be a serious imbalance between positive and negative samples. This results in the inability to guarantee the training efficiency of the binary base model in specific application scenarios in related technologies. This in turn results in the inability to effectively predict business results in specific application scenarios.

[0003] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention

[0004] The embodiments of the present application provide a business prediction model training method, device and non-volatile storage medium to at least solve the technical problem that business results cannot be effectively predicted in specific application scenarios due to the low efficiency of binary model training in specific application scenarios in the related technology.

[0005] According to one aspect of an embodiment of the present application, a business prediction model training method is provided, including: obtaining historical business data of a business to be predicted, and determining a first category of training samples and a second category of training samples from the historical business data, wherein the second category of training samples is more than the first category of training samples, and when the first category of training samples is a positive sample, the second category of training samples is a negative sample, and when the first category of training samples is a negative sample, the second category of training samples is a positive sample; randomly dividing the second category of training samples into multiple first sample sets, and fusing the multiple first sample sets with the first category of training samples, respectively, to obtain multiple second sample sets; training a business prediction model based on the multiple second sample sets, wherein the business prediction model includes multiple binary base models, the number of the binary base models is the same as the number of the second sample sets, and the business prediction model is used to process the samples to be predicted related to the business to be predicted, thereby obtaining a predicted business processing result of the business to be predicted.

[0006] Optionally, randomly dividing the second category training samples into multiple first sample sets includes: clustering the first category training samples to obtain clustering results, wherein the clustering results include cluster centers; determining the similarity between each second category training sample and the cluster center, and screening the second category training samples based on the similarity to obtain a first screening result, wherein the similarity between the second category training samples and the cluster center is the minimum value of the similarity between the second category training samples and each cluster; and randomly dividing the first screening result into multiple first sample sets.

[0007] Optionally, the method also includes: determining the similarity between each sample to be predicted and the cluster center, wherein the similarity between the sample to be predicted and the cluster center is the minimum value of the similarity between the sample to be predicted and each cluster center; screening the samples to be predicted based on the similarity corresponding to each sample to be predicted and a preset similarity threshold to obtain a second screening result, wherein the second screening result includes samples to be predicted whose corresponding similarity is greater than the preset similarity threshold; and using a business prediction model to process the second screening result.

[0008] Optionally, the method further includes: processing the samples to be predicted respectively through each binary base model to obtain the model output results of each binary base model; and determining the prediction business processing results according to the model output results of each binary base model.

[0009] Optionally, training the business prediction model based on multiple second sample sets includes: determining a binary base model corresponding to each second sample set; and training each binary base model using the second sample set corresponding to the binary base model.

[0010] Optionally, training with a second sample set corresponding to the binary base model includes: training the binary base model using a lightweight gradient boosting machine algorithm and the second sample set until the fluctuation value of an evaluation index of the binary base model is within a preset threshold range, wherein the fluctuation value of the evaluation index is the difference between the evaluation index of this round of training and the evaluation index of the previous round of training, and the evaluation index includes at least one of the following: precision, recall, and the harmonic mean of precision and recall.

[0011] Optionally, the business to be predicted includes a user broadband churn risk prediction business and a potential customer mining result prediction of a preset business, wherein, when the business to be predicted is a user broadband churn risk prediction business, the positive sample is the sample data of customers who no longer apply for broadband services within a preset time period, and the negative sample is the sample data of customers who continue to apply for broadband services within the preset time period; when the business to be predicted is a potential customer mining result prediction of a preset business, the positive sample is the sample data of customers who apply for the preset business, and the negative sample is the sample data of customers who do not apply for the preset business.

[0012] According to another aspect of an embodiment of the present application, a business prediction model training device is also provided, including: a first processing module, used to obtain historical business data of the business to be predicted, and determine a first type of training samples and a second type of training samples from the historical business data, wherein the second type of training samples are more than the first type of training samples, and when the first type of training samples are positive samples, the second type of training samples are negative samples, and when the first type of training samples are negative samples, the second type of training samples are positive samples; a second processing module, used to randomly divide the second type of training samples into multiple first sample sets, and fuse the multiple first sample sets with the first type of training samples respectively to obtain multiple second sample sets; a third processing module, used to train a business prediction model based on the multiple second sample sets, wherein the business prediction model includes multiple binary base models, the number of the binary base models is the same as the number of the second sample sets, and the business prediction model is used to process the samples to be predicted related to the business to be predicted, so as to obtain a predicted business processing result of the business to be predicted.

[0013] According to another aspect of an embodiment of the present application, a non-volatile storage medium is further provided, in which a program is stored, wherein when the program is running, a device where the non-volatile storage medium is located is controlled to execute a business prediction model training method.

[0014] According to another aspect of an embodiment of the present application, there is also provided an electronic device, including: a memory and a processor, the processor being used to run a program stored in the memory, wherein the business prediction model training method is executed when the program is running.

[0015] According to another aspect of an embodiment of the present application, a computer program product is also provided, including a computer program, which implements the steps of the business prediction model training method when executed by a processor.

[0016] In an embodiment of the present application, historical business data of the business to be predicted is obtained, and a first type of training samples and a second type of training samples are determined from the historical business data, wherein the second type of training samples are more than the first type of training samples, and when the first type of training samples are positive samples, the second type of training samples are negative samples, and when the first type of training samples are negative samples, the second type of training samples are positive samples; the second type of training samples are randomly divided into multiple first sample sets, and the multiple first sample sets are respectively fused with the first type of training samples to obtain multiple second sample sets; the business prediction model is trained according to the multiple second sample sets, wherein the business prediction model includes The invention comprises a plurality of binary base models, the number of the binary base models is the same as the number of the second sample set, the business prediction model is used to process the samples to be predicted related to the business to be predicted, so as to obtain the predicted business processing result of the business to be predicted, and the purpose of balancing the ratio of the first category of training samples and the second category of training samples in each sample set is achieved by dividing the second category of training samples and fusing them with the first category of training samples, thereby achieving the technical effect of improving the training efficiency of the binary model, and further solving the technical problem that the business results cannot be effectively predicted in a specific application scenario due to the low training efficiency of the binary model in the related technology in a specific application scenario. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0018] Figure 1 It is a structural schematic diagram of a computer terminal (mobile terminal) provided according to an embodiment of the present application;

[0019] Figure 2 It is a flowchart of a business prediction model training method provided according to an embodiment of the present application;

[0020] Figure 3 is a positive sample clustering elbow diagram provided according to an embodiment of the present application;

[0021] Figure 4 is a schematic diagram of data distribution provided according to an embodiment of the present application;

[0022] Figure 5 is a schematic diagram of the training results of each binary base model provided according to the embodiment of the present application;

[0023] Figure 6 It is a flowchart of a model training process provided according to an embodiment of the present application;

[0024] Figure 7is a schematic diagram of a modeling effect provided according to an embodiment of the present application;

[0025] Figure 8 It is a schematic diagram of the modeling effect of the prior art provided according to the embodiment of the present application;

[0026] Fig. 9 It is a flowchart of a model prediction process provided according to an embodiment of the present application;

[0027] Fig.10 It is a structural diagram of a business prediction model training device provided according to an embodiment of the present application. DETAILED DESCRIPTION

[0028] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0029] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0030] The binary base model based on the traditional machine learning algorithm generally requires a positive-negative sample ratio close to 1:1. However, in the construction of actual marketing application models, there is often a serious imbalance in the positive-negative sample ratio, such as: the broadband churn warning model that predicts the monthly user broadband churn risk, and the cold start scenarios for potential customer mining in some new business scenarios.

[0031] The related art usually attempts to adjust the samples or class weights, or up / downsample the samples. However, this method causes changes in the data distribution structure, and in practical applications, the methods in the related art are prone to model overfitting, or insufficient model training due to negative sampling.

[0032] In order to solve the above problems, relevant solutions are provided in the embodiments of the present application, which are described in detail below.

[0033] According to an embodiment of the present application, a method embodiment of a business prediction model training method is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0034] The method embodiments provided in the embodiments of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Figure 1 The hardware structure block diagram of a computer terminal (or mobile device) for implementing a business prediction model training method is shown. Figure 1 As shown, the computer terminal 10 (or mobile device 10) may include one or more (102a, 102b, ..., 102n are used to illustrate) processors 102 (the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 104 for storing data, and a transmission module 106 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the BUS bus), a network interface, a power supply and / or a camera. It can be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations are shown.

[0035] It should be noted that the one or more processors 102 and / or other data processing circuits described above may generally be referred to herein as "data processing circuits". The data processing circuits may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. In addition, the data processing circuit may be a single independent processing module, or may be incorporated in whole or in part into any of the other components in the computer terminal 10 (or mobile device). As described in the embodiments of the present application, the data processing circuit acts as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0036] The memory 104 can be used to store software programs and modules of application software, such as program instructions / data storage devices corresponding to the business prediction model training method in the embodiment of the present application. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, that is, the above-mentioned business prediction model training method is realized. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories may be connected to the computer terminal 10 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0037] The transmission device 106 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the computer terminal 10. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0038] The display may be, for example, a touch screen liquid crystal display (LCD) that enables a user to interact with a user interface of the computer terminal 10 (or mobile device).

[0039] In the above operating environment, the present application embodiment provides a business prediction model training method, such as Figure 2 As shown, the method comprises the following steps:

[0040] Step S202, obtaining historical business data of the business to be predicted, and determining a first type of training samples and a second type of training samples from the historical business data, wherein the second type of training samples are more than the first type of training samples, and when the first type of training samples are positive samples, the second type of training samples are negative samples, and when the first type of training samples are negative samples, the second type of training samples are positive samples;

[0041] In the technical solution provided in step S202, the business to be predicted includes a user broadband churn risk prediction business and a potential customer mining result prediction of a preset business, wherein, in the case where the business to be predicted is a user broadband churn risk prediction business, the positive sample is the sample data of customers who no longer handle the broadband service within the preset time period, and the negative sample is the sample data of customers who continue to handle the broadband service within the preset time period; in the case where the business to be predicted is a potential customer mining result prediction of a preset business, the positive sample is the sample data of customers who handle the preset service, and the negative sample is the sample data of customers who do not handle the preset service.

[0042] As an optional implementation, the step of randomly dividing the second category training samples into multiple first sample sets includes: clustering the first category training samples to obtain clustering results, wherein the clustering results include cluster centers; determining the similarity between each second category training sample and the cluster center, and screening the second category training samples based on the similarity to obtain a first screening result, wherein the similarity between the second category training samples and the cluster center is the minimum value of the similarity between the second category training samples and each cluster; and randomly dividing the first screening result into multiple first sample sets.

[0043] Step S204, randomly dividing the second type of training samples into a plurality of first sample sets, and fusing the plurality of first sample sets with the first type of training samples respectively to obtain a plurality of second sample sets;

[0044] In the technical solution provided in step S204, assuming that the business prediction model includes 5 binary base models, the second-category training samples can be randomly divided into 5 first sample sets. By first deleting the second-category training samples that are too similar to the first-category training samples according to the similarity, and dividing the second-category training samples into multiple sample sets and then merging each sample set with the first-category training samples, the ratio between the first-category training samples and the second-category training samples in each of the second sample sets finally obtained will be closer to 1:1 than that in the original data set.

[0045] For example, in the broadband off-grid warning scenario, the process of using this method to process sample data to balance positive and negative samples is as follows:

[0046] Step 1: Screen positive samples, the original positive and negative sample ratio of the model: positive samples (offline in the next month): 22020, negative samples (not offline in the next month): 2975608, initial positive and negative sample ratio: 1:135;

[0047] Step 2: Clustering of positive samples of training data set. According to the elbow method, Figure 3 As shown, the number of clusters of training positive samples is set. The number of clusters of training positive samples can be determined by the user, for example, it can be set to 4.

[0048] Specifically, the Elbow Method, also known as the Elbow Rule, is a simple method for determining the number of clusters, which is widely used in cluster analysis, especially in the K-means algorithm. This method can provide an optional appropriate number of clusters by evaluating the changes in cluster quality.

[0049] Step 3: Calculate the similarity between the cluster centers of the negative samples in the training set and each positive sample, and determine the minimum similarity between the cluster centers of the negative samples in the training set and each positive sample as the final similarity;

[0050] Optionally, when calculating the similarity, the similarity between the negative sample of the training set and each positive sample cluster center can be reflected by the similarity distance. And the maximum value of the similarity distance between the negative sample of the training set and each positive sample cluster center is determined to be the similarity distance of the negative sample of the training set, wherein the similarity distance can be used to determine the similarity, and the similarity distance is inversely proportional to the similarity, that is, the larger the similarity distance, the smaller the similarity.

[0051] Step 4: Filter negative samples according to the preset threshold;

[0052] As an optional implementation, when filtering negative samples, a similarity distance threshold, such as 0.9, can be set. Negative samples with a similarity distance less than the threshold are discarded. After screening, the number of negative samples in this example is reduced to 1967438, a decrease of about 33.88%. The data distribution is as follows Figure 4 As shown in the figure. After filtering the negative samples, the positive samples (those who left the network in the next month) are 22020, the negative samples (those who did not leave the network in the next month) are 1967438, and the positive-negative sample ratio is 1:89, which is an increase of 51.69%. It can be seen that this method can significantly improve the problem of imbalance between positive and negative samples.

[0053] Step 5: Divide the filtered negative samples into multiple sets and fuse them with the positive sample training set respectively, so that the positive-negative sample ratio can be increased to 1:18.

[0054] Step S206, training the business prediction model based on multiple second sample sets, wherein the business prediction model includes multiple binary base models, the number of binary base models is the same as the number of second sample sets, and the business prediction model is used to process samples to be predicted related to the business to be predicted, thereby obtaining a predicted business processing result of the business to be predicted.

[0055] In the technical solution provided in step S206, the step of training the business prediction model based on multiple second sample sets includes: determining a binary base model corresponding to each second sample set; and training each binary base model using the second sample set corresponding to the binary base model.

[0056] As an optional implementation, the step of training with a second sample set corresponding to the binary base model includes: training the binary base model with a lightweight gradient boosting machine algorithm and the second sample set until the fluctuation value of the evaluation index of the binary base model is within a preset threshold range, wherein the fluctuation value of the evaluation index is the difference between the evaluation index of this round of training and the evaluation index of the previous round of training, and the evaluation index includes at least one of the following: precision, recall, and the harmonic mean of precision and recall.

[0057] Optional, such as Figure 5 As shown, the above evaluation indicators include evaluation indicators of each model on the training set and evaluation indicators on the test set. Figure 5 These are various evaluation indicators after each binary base model is stabilized in an actual training scenario.

[0058] It should be noted that the Light Gradient Boosting Machine (LightGBM) algorithm is a method for optimizing machine learning model parameters based on gradient boosting trees, which has the characteristics of high performance and low memory usage. It can help find the best parameter combination to improve model performance.

[0059] As an optional implementation, after the model training is completed using the Light GBM algorithm, the grid search cross-validation method (Grid Cross-Validation, GridCV) can also be used to optimize the model parameters, including the following steps:

[0060] The first step is to define the parameter space: First, we need to determine the range of parameters to be optimized. These parameters can be learning rate, number of trees, maximum depth, minimum number of leaf samples, etc.

[0061] Step 2: Grid search: Grid search is an exhaustive search method that tries all possible parameter combinations in a given parameter space. In this way, the best parameter combination can be found.

[0062] The third step is cross-validation: In order to evaluate the performance of different parameter combinations, a cross-validation method can be used. Cross-validation usually divides the data set into several subsets, and then uses one of the subsets as the test set and the rest as the training set in turn. In this way, each subset will be used as a test set once, thereby obtaining a more reliable performance estimate.

[0063] Step 4: Performance evaluation: During each cross-validation process, the performance of the model needs to be evaluated. Usually, indicators such as accuracy, recall, and F1 score are used to measure the performance of the model.

[0064] Step 5, Result Analysis: After grid search and cross validation are completed, the best parameter combination can be found. Based on these parameters, a model with better performance can be trained.

[0065] Step 6: Model training: Finally, retrain the model using the best parameter combination found until the iteration stopping condition is met.

[0066] As an optional implementation, the complete process of training the model in this application is as follows: Figure 6 As shown, the following steps are included:

[0067] The first step is to screen the positive training samples;

[0068] The second step is to train the clustering of the positive sample set and obtain multiple cluster centers C1, C2...Cn;

[0069] The third step is to calculate the similarity distance L1, L2, ..., Ln between the training negative samples and each cluster center;

[0070] The fourth step is to determine the maximum similarity distance between each cluster center as the similarity distance corresponding to the training negative sample;

[0071] The fifth step is to set the similarity distance threshold and screen each training negative sample according to the threshold;

[0072] Step 6: randomly divide the negative samples into multiple sample sets and fuse them with the positive samples respectively;

[0073] Step 7: Use the LightGBM method and the fused sample set to train each base model;

[0074] Step 8: Use the GridCV method to adjust the parameters of each base model;

[0075] In the ninth step, the Voting ensemble selection voting method is used to integrate the prediction results of each base model and output the final prediction results.

[0076] In some embodiments of the present application, the precision, recall and F1-Score of the model trained by the method provided by the present application on the training set and the test set are as follows: Figure 7 The precision, recall and F1-Score of the related technology on the same training set and test set are shown in Figure 8 It can be seen that the F1-Score of the method provided in the embodiment of the present application is improved by 5.23pp compared with the method in the related art, and the difference between the indicators of the training and test sets is small, and the model is more robust. In other words, the training effect of the model provided by the method of the present application is significantly better than the training effect in the related art.

[0077] In some embodiments of the present application, after the business prediction model training is completed, the business prediction model can be used to process the samples to be predicted in the following manner: determine the similarity between each sample to be predicted and the cluster center, wherein the similarity between the sample to be predicted and the cluster center is the minimum value of the similarity between the sample to be predicted and each cluster center; screen the samples to be predicted based on the similarity corresponding to each sample to be predicted and a preset similarity threshold to obtain a second screening result, wherein the second screening result includes samples to be predicted whose corresponding similarity is greater than the preset similarity threshold; and use the business prediction model to process the second screening result.

[0078] In some embodiments of the present application, after the business prediction model training is completed, the business prediction model can also be used to process the samples to be predicted in the following manner: process the samples to be predicted by each binary base model respectively to obtain the model output results of each binary base model; determine the predicted business processing results based on the model output results of each binary base model.

[0079] As an optional implementation method, the output results of each binary base model can be integrated by using the majority voting method to obtain the final judgment result and verify the final indicator effect of the model.

[0080] Specifically, when the prediction results of multiple models are integrated to obtain the final result by majority voting, the prediction results of multiple models can be combined to improve the overall prediction accuracy and robustness. Specifically, there are two main voting strategies: hard voting and soft voting.

[0081] Hard Voting: In hard voting, each model independently predicts the input data and outputs a class label. Then, the predictions of all models are collected and the final prediction is determined by majority voting, that is, the class with the most votes is selected as the final prediction. Hard voting is simple and easy to implement, but it ignores the probability information that the model may give.

[0082] Soft Voting: Unlike hard voting, soft voting takes into account the probability information predicted by each model. Each model outputs a probability distribution, which represents the probability that the input data belongs to each category. These probabilities are then averaged or weighted, and the final prediction is the category with the highest average probability. Soft voting can consider the confidence of each model in more detail and can usually provide more accurate predictions than hard voting.

[0083] As an optional implementation, when implementing voting ensemble, different types of models can be selected to increase diversity, which can capture different aspects of the data and thus improve the overall performance.

[0084] In some embodiments of the present application, the complete process of using the above-trained business prediction model to process the sample to be tested is as follows: Fig. 9 As shown, the following steps are included:

[0085] The first step is to calculate the similarity distance L1, L2, ... Ln between the sample to be tested and each cluster center;

[0086] The second step is to determine the maximum similarity distance between each cluster center as the similarity distance S corresponding to the sample to be tested;

[0087] The third step is to set the similarity distance threshold S0, and screen each sample to be tested according to the threshold, and only retain the samples to be tested with S≥S0;

[0088] The fourth step is to input the samples to be tested into each binary base model;

[0089] The fifth step is to use the Voting ensemble selection voting method to integrate the prediction results of each base model and output the final prediction results.

[0090] The method adopts a method of obtaining historical business data of the business to be predicted, and determining a first type of training samples and a second type of training samples from the historical business data, wherein the second type of training samples are more than the first type of training samples, and when the first type of training samples are positive samples, the second type of training samples are negative samples, and when the first type of training samples are negative samples, the second type of training samples are positive samples; randomly dividing the second type of training samples into a plurality of first sample sets, and respectively fusing the plurality of first sample sets with the first type of training samples to obtain a plurality of second sample sets; training a business prediction model according to the plurality of second sample sets, wherein the business prediction model includes a plurality of binary base models, the number of the binary base models is the same as the number of the second sample sets, and the business prediction model is used to process the samples to be predicted related to the business to be predicted, so as to obtain a predicted business processing result of the business to be predicted, and by dividing the second type of training samples and fusing them with the first type of training samples, the purpose of balancing the proportion of the first type of training samples and the second type of training samples in each sample set is achieved, thereby achieving the technical effect of improving the training efficiency of the binary model, and further solving the technical problem that the business results cannot be effectively predicted in a specific application scenario due to the low training efficiency of the binary model in the related technology in a specific application scenario.

[0091] Furthermore, through the method provided in the embodiment of the present application, the samples can be preliminarily screened to narrow the range of modeling samples and balance the ratio of positive and negative samples, thereby improving the index effect of the final output of the model.

[0092] The present application embodiment provides a business prediction model training device, Fig.10 is a schematic diagram of the structure of the device. Fig.10 It can be seen that the device includes: a first processing module 100, which is used to obtain historical business data of the business to be predicted, and determine the first category of training samples and the second category of training samples from the historical business data, wherein the second category of training samples is more than the first category of training samples, and when the first category of training samples is a positive sample, the second category of training samples is a negative sample, and when the first category of training samples is a negative sample, the second category of training samples is a positive sample; a second processing module 102, which is used to randomly divide the second category of training samples into multiple first sample sets, and merge the multiple first sample sets with the first category of training samples respectively to obtain multiple second sample sets; a third processing module 104, which is used to train a business prediction model based on the multiple second sample sets, wherein the business prediction model includes multiple binary base models, the number of the binary base models is the same as the number of the second sample sets, and the business prediction model is used to process the samples to be predicted related to the business to be predicted, so as to obtain the predicted business processing result of the business to be predicted.

[0093] In some embodiments of the present application, the step of the first processing module 100 randomly dividing the second category training samples into multiple first sample sets includes: clustering the first category training samples to obtain clustering results, wherein the clustering results include cluster centers; determining the similarity between each second category training sample and the cluster center, and screening the second category training samples based on the similarity to obtain a first screening result, wherein the similarity between the second category training samples and the cluster center is the minimum value of the similarity between the second category training samples and each cluster; and randomly dividing the first screening result into multiple first sample sets.

[0094] In some embodiments of the present application, the step of the third processing module 104 training the business prediction model based on multiple second sample sets includes: determining a binary base model corresponding to each second sample set; and training each binary base model using the second sample set corresponding to the binary base model.

[0095] In some embodiments of the present application, the step of training the second sample set corresponding to the binary base model adopted by the third processing module 104 includes: training the binary base model using a lightweight gradient boosting machine algorithm and the second sample set until the fluctuation value of the evaluation index of the binary base model is within a preset threshold range, wherein the fluctuation value of the evaluation index is the difference between the evaluation index of the current round of training and the evaluation index of the previous round of training, and the evaluation index includes at least one of the following: precision, recall, and the harmonic mean of precision and recall.

[0096] In some embodiments of the present application, the service to be predicted includes a user broadband churn risk prediction service and a potential customer mining result prediction of a preset service, wherein, in the case where the service to be predicted is a user broadband churn risk prediction service, the positive sample is the sample data of customers who no longer apply for broadband services within a preset time period, and the negative sample is the sample data of customers who continue to apply for broadband services within a preset time period;

[0097] In the case where the business to be predicted is the potential customer mining result prediction of the preset business, the positive sample is the sample data of customers who handle the preset business, and the negative sample is the sample data of customers who do not handle the preset business.

[0098] In some embodiments of the present application, the business prediction model training device is also used to: determine the similarity between each sample to be predicted and the cluster center, wherein the similarity between the sample to be predicted and the cluster center is the minimum value of the similarity between the sample to be predicted and each cluster center; screen the samples to be predicted based on the similarity corresponding to each sample to be predicted and a preset similarity threshold to obtain a second screening result, wherein the second screening result includes samples to be predicted whose corresponding similarity is greater than the preset similarity threshold; and use the business prediction model to process the second screening result.

[0099] In some embodiments of the present application, the business prediction model training device is also used to: process the samples to be predicted respectively through each binary base model to obtain the model output results of each binary base model; and determine the predicted business processing results based on the model output results of each binary base model.

[0100] It should be noted that the various modules in the above-mentioned business prediction model training device can be program modules (for example, a set of program instructions that implement a certain specific function) or hardware modules. For the latter, it can be expressed in the following forms, but is not limited to this: the expression form of each of the above-mentioned modules is a processor, or the functions of each of the above-mentioned modules are implemented by a processor.

[0101] According to an embodiment of the present application, a non-volatile storage medium is also provided, in which a program is stored, wherein when the program is running, the device where the non-volatile storage medium is located is controlled to execute the following business prediction model training method: historical business data of the business to be predicted is obtained, and a first type of training samples and a second type of training samples are determined from the historical business data, wherein the second type of training samples are more than the first type of training samples, and when the first type of training samples are positive samples, the second type of training samples are negative samples, and when the first type of training samples are negative samples, the second type of training samples are positive samples; the second type of training samples are randomly divided into multiple first sample sets, and the multiple first sample sets are respectively fused with the first type of training samples to obtain multiple second sample sets; the business prediction model is trained based on the multiple second sample sets, wherein the business prediction model includes multiple binary base models, and the number of the binary base models is the same as the number of the second sample sets, and the business prediction model is used to process the samples to be predicted related to the business to be predicted, thereby obtaining a predicted business processing result of the business to be predicted.

[0102] According to an embodiment of the present application, an electronic device is also provided, comprising: a memory and a processor, the processor being used to run a program stored in the memory, wherein the following steps of a business prediction model training method are executed when the program is running: obtaining historical business data of the business to be predicted, and determining a first category of training samples and a second category of training samples from the historical business data, wherein the second category of training samples is more than the first category of training samples, and when the first category of training samples is a positive sample, the second category of training samples is a negative sample, and when the first category of training samples is a negative sample, the second category of training samples is a positive sample; randomly dividing the second category of training samples into multiple first sample sets, and merging the multiple first sample sets with the first category of training samples respectively to obtain multiple second sample sets; training a business prediction model based on the multiple second sample sets, wherein the business prediction model includes multiple binary base models, and the number of the binary base models is the same as the number of the second sample sets, and the business prediction model is used to process the samples to be predicted related to the business to be predicted, thereby obtaining a predicted business processing result of the business to be predicted.

[0103] In the above embodiments of the present application, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0104] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0105] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0106] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0107] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the relevant technology or all or part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, RandomAccess Memory), mobile hard disk, disk or optical disk, etc. Various media that can store program codes.

[0108] The above is only a preferred implementation of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A business prediction model training method, characterized in that: include: Acquire historical business data of the business to be predicted, and determine a first type of training samples and a second type of training samples from the historical business data, wherein the second type of training samples are more than the first type of training samples, and when the first type of training samples are positive samples, the second type of training samples are negative samples, and when the first type of training samples are negative samples, the second type of training samples are positive samples; Randomly dividing the second type of training samples into a plurality of first sample sets, and fusing the plurality of first sample sets with the first type of training samples respectively to obtain a plurality of second sample sets; The business prediction model is trained based on multiple second sample sets, wherein the business prediction model includes multiple binary base models, the number of the binary base models is the same as the number of the second sample sets, and the business prediction model is used to process samples to be predicted related to the business to be predicted, so as to obtain a predicted business processing result of the business to be predicted.

2. The business prediction model training method according to claim 1, characterized in that: Randomly dividing the second type of training samples into a plurality of first sample sets comprises: Clustering the first type of training samples to obtain a clustering result, wherein the clustering result includes a cluster center; Determine the similarity between each of the second-category training samples and the cluster center, and screen the second-category training samples according to the similarity to obtain a first screening result, wherein the similarity between the second-category training samples and the cluster center is the minimum value of the similarity between the second-category training samples and each of the clusters; The first screening results are randomly divided into a plurality of first sample sets.

3. The business prediction model training method according to claim 2, characterized in that: The business prediction model training method also includes: Determining the similarity between each of the samples to be predicted and the cluster center, wherein the similarity between the sample to be predicted and the cluster center is the minimum value of the similarity between the sample to be predicted and each of the cluster centers; The samples to be predicted are screened based on the similarities corresponding to the samples to be predicted and a preset similarity threshold, to obtain a second screening result, wherein the second screening result includes the samples to be predicted whose corresponding similarities are greater than the preset similarity threshold; The second screening result is processed using the business prediction model.

4. The business prediction model training method according to claim 1, characterized in that: The method further comprises: Processing the samples to be predicted respectively through each of the dichotomy base models to obtain model output results of each of the dichotomy base models; The predicted business processing result is determined according to the model output results of each of the dichotomy base models.

5. The business prediction model training method according to claim 1, characterized in that: Training the business prediction model according to the plurality of second sample sets includes: Determine the bipartite base model corresponding to each of the second sample sets; For each of the dichotomy base models, the second sample set corresponding to the dichotomy base model is used for training.

6. The business prediction model training method according to claim 5, characterized in that: The step of using the second sample set corresponding to the dichotomy base model for training comprises: The binary base model is trained using a lightweight gradient boosting machine algorithm and the second sample set until a fluctuation value of an evaluation index of the binary base model is within a preset threshold range, wherein the fluctuation value of the evaluation index is a difference between an evaluation index of a current round of training and an evaluation index of a previous round of training, and the evaluation index includes at least one of the following: precision, recall, and a harmonic mean of precision and recall.

7. The business prediction model training method according to claim 1, characterized in that: The services to be predicted include user broadband loss risk prediction services and potential customer mining result prediction of preset services, wherein: In the case where the service to be predicted is the user broadband loss risk prediction service, the positive sample is sample data of customers who no longer apply for broadband services within a preset time period, and the negative sample is sample data of customers who continue to apply for broadband services within the preset time period; In the case where the business to be predicted is a potential customer mining result prediction of a preset business, the positive sample is sample data of customers who handle the preset business, and the negative sample is sample data of customers who do not handle the preset business.

8. A business prediction device, characterized in that: include: A first processing module is used to obtain historical business data of the business to be predicted, and determine a first type of training samples and a second type of training samples from the historical business data, wherein the second type of training samples are more than the first type of training samples, and when the first type of training samples are positive samples, the second type of training samples are negative samples, and when the first type of training samples are negative samples, the second type of training samples are positive samples; A second processing module, configured to randomly divide the second type of training samples into a plurality of first sample sets, and fuse the plurality of first sample sets with the first type of training samples respectively to obtain a plurality of second sample sets; The third processing module is used to train the business prediction model based on multiple second sample sets, wherein the business prediction model includes multiple binary base models, and the number of the binary base models is the same as the number of the second sample sets. The business prediction model is used to process the samples to be predicted related to the business to be predicted, so as to obtain the predicted business processing results of the business to be predicted.

9. A non-volatile storage medium, characterized in that: The non-volatile storage medium stores a program, wherein when the program is running, the device where the non-volatile storage medium is located is controlled to execute the business prediction model training method described in any one of claims 1 to 7.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the business prediction model training method described in any one of claims 1 to 7 are implemented.

Images