Model training method and device for identifying customer qualification, equipment and medium
By preprocessing the income data of multiple sample customers and optimizing the multi-classification neural network model, the weight value of the revenue level interval is calculated, and the accuracy of customer income assessment in financial risk control is solved, reducing risk control costs and improving risk control The accuracy and interpretability of the assessment were conducted.
Patent Information
- Application Number
- CN202411145386.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-20
- Publication Date
- 2024-12-13
AI Technical Summary
The existing technology is difficult to accurately evaluate customer revenue levels in financial risk control, resulting in large errors and sample distribution does not conform to business awareness, which increases risk control costs.
By obtaining the income data of multiple sample customers for preprocessing, a multi-classification neural network model is constructed, and the weight value of each revenue level interval is calculated through parameter adjustment and optimization to evaluate customer qualifications, reduce errors and improve accuracy.
It realizes accurate identification of customer qualifications, reduces risk control costs, and improves the accuracy and interpretability of evaluation.
Smart Images
Figure CN119128716A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of bank data processing, and in particular to a model training method, device, equipment and medium for identifying customer qualifications. Background Art
[0002] In the field of financial risk control, it is crucial to review the qualifications of the customer, such as income level, work situation, age, marital status, etc. Only after obtaining this information and making a comprehensive assessment can the customer be approved and financial risks can be better controlled. However, in actual business, data that measures the income level of customers is relatively difficult to obtain. Even if there are a few third-party data vendors that can provide it, the query fee is extremely high. In the face of a large number of online application customers in the Internet financial lending business, if all third-party data is queried, it will undoubtedly greatly increase the risk control cost. Therefore, how to evaluate and predict the income level of customers under the premise of controllable costs, understand customers more comprehensively, make better judgments on customers, and control financial risks is a major difficulty in current Internet financial risk control.
[0003] In most existing solutions, the customer's qualifications are inferred through other customer data, such as using bank card loan and repayment information and provident fund information in the People's Bank of China's credit report to calculate by simple coefficient multiplication and addition, or to build a simple linear regression model to predict the customer's specific income. Although this method is simple to operate, it has a large error, and the sample distribution of its prediction results does not conform to business cognition and is difficult to explain. Therefore, in the final actual business application, it is of limited help.
[0004] Therefore, the present invention provides a new solution to solve this problem. Summary of the invention
[0005] In view of the shortcomings of the prior art, the purpose of the present invention is to provide a model training method, device, equipment and medium for identifying customer qualifications, which effectively solves the problems of large errors and defects in sample distribution in the existing technical solutions.
[0006] The technical solution is that the present application provides a model training method for identifying customer qualifications, the method comprising:
[0007] Obtaining income data of multiple sample customers, and preprocessing the income data of the multiple sample customers to obtain multiple model input features and classification ratios; the preprocessing includes filling, conversion and elimination; the classification ratio is the ratio of the number of sample customers corresponding to each income level interval after the income of the multiple sample customers is classified into levels;
[0008] The obtained multiple input features are sent to a pre-established customer qualification identification model; the customer qualification identification model is pre-established by using a multi-classification neural network and sample customer income data, and is optimized by a parameter adjustment network;
[0009] The customer qualification identification model obtains multiple prediction scores based on the received multiple input features to select the maximum prediction score; the prediction score corresponds to the weight value of each income level interval given by the customer qualification identification model based on the sample weight algorithm network;
[0010] The customer qualification identification model is evaluated based on the maximum predicted score of each sample customer and the graded proportion of the sample customers, so as to identify the customer qualifications based on the evaluated customer qualification identification model.
[0011] Furthermore, the weight values corresponding to the income level intervals are obtained based on the sample weight algorithm network in the following specific steps:
[0012] The customer qualification identification model assigns the same initial weight value to each income level interval;
[0013] The customer qualification identification model adjusts the initial weight value of each income level interval multiple times according to the pre-stored income data of sample customers, until the weight value corresponding to the income level interval corresponds to the graded proportion of the income data of the sample customers.
[0014] Furthermore, the customer qualification identification model adjusts the initial weight value of each income level interval multiple times according to the income data of the pre-stored sample customers, including:
[0015] After roughly adjusting and iterating the initial weight value of each income level interval with a preset step length, it is determined whether the weight value after the rough adjustment and iteration corresponds to the graded proportion of the sample customers;
[0016] If not, the preset step size is halved and the process is repeated until the weight value corresponding to the income level interval corresponds to the graded proportion of the sample customers.
[0017] Furthermore, the specific steps of the customer qualification identification model constructed by a multi-classification neural network and the income data of sample customers and optimized by a parameter adjustment network are:
[0018] The initial customer qualification recognition model is constructed using the sample customer's income data and a multi-classification neural network, and the sample customer's income data is used as the target label;
[0019] The initial customer qualification identification model is trained and optimized multiple times based on the parameter adjustment network and the target label to obtain the customer qualification identification model.
[0020] Furthermore, the customer qualification identification model is evaluated based on the maximum predicted score of each sample customer and the graded proportion of the sample customers, including:
[0021] Perform a two-dimensional cross-section of the maximum predicted score of each sample customer and the graded proportion of the sample customers to obtain a score table;
[0022] The accuracy of the customer qualification identification model is calculated based on the score table to evaluate the customer qualification identification model based on the accuracy.
[0023] Further, the calculating the accuracy of the customer qualification identification model based on the score table to evaluate the customer qualification identification model based on the accuracy includes:
[0024] Selecting first sample customers whose number of maximum predicted scores is the same as the graded proportion of sample customers from the diagonal line of the score table, and determining the number of first sample customers;
[0025] The accuracy of the customer qualification identification model is determined by the ratio of the number of first sample customers to the sample customers.
[0026] Further, the evaluating the customer qualification identification model based on accuracy includes:
[0027] Determining whether the accuracy of the customer qualification identification model reaches a preset accuracy of the customer qualification identification model;
[0028] If not, it is impossible to identify the customer qualifications based on the evaluated customer qualification identification model.
[0029] The present application also provides a model training device for identifying customer qualifications, the device comprising:
[0030] An acquisition module is used to acquire income data of multiple sample customers, and pre-process the income data of multiple sample customers to obtain multiple model input features and classification ratios; the pre-processing includes filling, conversion and elimination; the classification ratio is the ratio of the number of sample customers corresponding to each income level interval after the income of multiple sample customers is divided into levels;
[0031] A sending module is used to send the obtained multiple input features to a pre-established customer qualification identification model; the customer qualification identification model is pre-established by using a multi-classification neural network and income data of sample customers, and is optimized by a parameter adjustment network;
[0032] A receiving module, used for the customer qualification identification model to obtain multiple prediction scores based on the received multiple input features to select the maximum prediction score; the prediction score corresponds to the weight value of each income level interval given by the customer qualification identification model based on the sample weight algorithm network;
[0033] The identification module is used to evaluate the customer qualification identification model based on the maximum predicted score of each sample customer and the graded proportion of the sample customers, so as to identify the customer qualifications based on the evaluated customer qualification identification model.
[0034] The present application also provides an electronic device, comprising: a processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor and the memory communicate through the bus, and when the machine-readable instructions are executed by the processor, any step of a model training method for identifying customer qualifications is performed.
[0035] The present application also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the computer program executes any one of the steps of the model training method for identifying customer qualifications.
[0036] The present invention achieves the following beneficial effects:
[0037] The present invention proposes a model training method for identifying customer qualifications. The method first obtains income data of multiple sample customers, and pre-processes the income data of the multiple sample customers to obtain multiple model input features and graded proportions; the pre-processing includes filling, conversion and elimination; the graded proportion is the proportion of the number of sample customers corresponding to each income grade interval after the income of the multiple sample customers is graded; secondly, the obtained multiple model input features are sent to a pre-established customer qualification identification model; the customer qualification identification model is pre-constructed by a multi-classification neural network and the income data of the sample customers, and is optimized by a parameter adjustment network; then the customer qualification identification model obtains multiple prediction scores based on the received multiple model input features to select the maximum prediction score; the prediction score corresponds to the weight value of each income grade interval of the customer qualification identification model based on the sample weight algorithm network; finally, the customer qualification identification model is evaluated based on the maximum prediction score of each sample customer and the graded proportion of the sample customer, so as to identify the customer qualifications based on the evaluated customer qualification identification model, and the income data of the sample customers is used as a basis, and the customer qualification identification model is constructed in a way to reduce the existence of errors, ensure the accuracy of identifying customer qualifications, and avoid the problems of large errors and defects in sample distribution in the existing technical solutions. BRIEF DESCRIPTION OF THE DRAWINGS
[0038] Figure 1 It is a flow chart of a model training method for identifying customer qualifications provided by the present invention;
[0039] Figure 2It is a structural schematic diagram of a model training device for identifying customer qualifications of the present invention. DETAILED DESCRIPTION
[0040] For the above and other technical contents, features and effects of the present invention, please refer to the following attached Figure 1-2 The detailed description of the embodiments will clearly show that the structural contents mentioned in the following embodiments are all based on the drawings in the specification.
[0041] The existing method of inferring customer qualifications through other customer data, such as using bank card loan and repayment details and provident fund information in the People's Bank of China's credit report, to calculate by simple coefficient multiplication and addition, or to build a simple linear regression model to predict the customer's specific income. Although this method is simple to operate, it has a large error, and the sample distribution of its prediction results does not conform to business cognition and is difficult to explain. Therefore, in the final actual business application, it is of limited help.
[0042] Based on this, the present application provides a model training method, device, equipment and medium for identifying customer qualifications to solve this problem.
[0043] Exemplary embodiments of the present invention will be described below with reference to the accompanying drawings.
[0044] First, a model training method for identifying customer qualifications provided in this application is described.
[0045] The present application embodiment provides a model training method for identifying customer qualifications, such as Figure 1 As shown, the method includes:
[0046] S101, obtaining income data of multiple sample customers, and preprocessing the income data of the multiple sample customers to obtain multiple model input features and classification ratios; the preprocessing includes filling, conversion and elimination; the classification ratio is the ratio of the number of sample customers corresponding to each income level interval after the income of the multiple sample customers is classified into levels;
[0047] S102, the obtained multiple input features are sent to a pre-established customer qualification identification model; the customer qualification identification model is pre-established by using a multi-classification neural network and income data of sample customers, and is optimized by a parameter adjustment network;
[0048] S103, the customer qualification identification model obtains multiple prediction scores based on the received multiple input features to select the maximum prediction score; the prediction score corresponds to the weight value of each income level interval given by the customer qualification identification model based on the sample weight algorithm network;
[0049] S104. Evaluate the customer qualification identification model based on the maximum predicted score of each sample customer and the graded proportion of the sample customers, so as to identify the customer qualifications based on the evaluated customer qualification identification model.
[0050] In step S101, the income data of the sample customers include the credit data of the People's Bank of China, consumption data, provident fund data, and text data. The above data are all provided by the sample customers, and can also be other income data that can be easily obtained. The above income data is preprocessed, and the preprocessing includes filling, conversion and elimination. The filling is to fill the missing value when the difference between the two values before and after the income data is large, and the judgment basis for the large difference between the two values before and after can be that the difference reaches a preset threshold, or the difference does not conform to the law between the income data; the conversion is to convert the text data into numerical data, and the elimination is to remove the missing value existing in the income data. The outliers including abnormally large values and abnormally small values are eliminated to ensure the accuracy of the income data, and then the income data of each sample customer is processed into multiple model input features of X1~Xn according to its characteristics, and the income level of the sample customers is divided into intervals to obtain multiple income level intervals. The divided intervals are set according to the actual situation. Table 1 is a schematic income level interval. The number of sample customers falling into the income level interval is counted, and the graded proportion of each income level interval is obtained according to the number of sample customers falling into each income level interval. The graded proportion is expressed in terms of the number of sample customers.
[0051] Table 1. Correspondence between income level range and income
[0052]
[0053] In step S102, multiple input features obtained by feature processing based on the income data of the sample customers are sent to a pre-established customer qualification identification model; the customer qualification identification model is pre-constructed by a multi-classification neural network and the income data of the sample customers, the income data of the sample customers is also pre-acquired, and the income data is pre-processed, such as row cleaning, conversion and normalization, to eliminate noise, missing values and outliers, and ensure the comparability between different features. Before building the customer qualification identification model, it is necessary to select the features that have the greatest impact on the customer's qualification level from the income data, and further process and convert the selected features to extract more useful information or improve the performance of the model. For example, new features such as annual income growth rate and income stability index can be created, and the multi-classification neural network is used as the architecture of the customer qualification identification model. The multi-classification neural network can be a lightgbm algorithm. In order to ensure the accuracy of the customer qualification identification model, the obtained customer qualification identification model is optimized using a parameter adjustment network and sample data.
[0054] In the specific implementation process of step S102, there is an embodiment in which the specific steps of the customer qualification identification model constructed by the multi-classification neural network and the income data of the sample customers and optimized by the parameter adjustment network are:
[0055] S1021. Construct an initial customer qualification recognition model using the sample customer's income data and a multi-classification neural network, and use the sample customer's income data as the target label;
[0056] S1022: training and optimizing the initial customer qualification identification model multiple times based on the parameter adjustment network and the target label to obtain the customer qualification identification model.
[0057] In steps S1021-S1022, an initial customer qualification identification model is constructed using the income data of sample customers and a multi-classification neural network. In order to improve the accuracy of the initial customer qualification identification model, the income data of sample customers is used as a target label. The target label is used to train and learn the initial customer qualification identification model multiple times, and the parameters of the initial customer qualification identification model are adjusted using a parameter adjustment network so that the recognition result output by the adjusted customer qualification identification model is closer to the true value, thereby improving the recognition result of the customer qualification identification model.
[0058] In step S103, the customer qualification identification model receives a plurality of input features, and performs prediction based on the received plurality of input features to obtain a plurality of prediction scores, wherein the plurality of prediction scores correspond to a plurality of income level intervals, that is, one prediction score corresponds to one income level interval, and one income level interval has a prediction score output by the customer qualification identification model, and the plurality of prediction scores output by the customer qualification identification model are sorted, either in ascending order or in descending order, and the maximum value of the prediction score is selected therefrom, and the corresponding income level interval into which the sample customer falls is determined based on the maximum prediction score, so as to obtain the customer's qualification according to the income level interval into which the sample customer falls, and the prediction score corresponds to the weight value of each income level interval provided by the customer qualification identification model based on the sample weight algorithm network, that is, according to the correspondence between the prediction score and the weight value of each income level interval, the income level interval into which the sample customer falls can be obtained according to the prediction score.
[0059] In the specific implementation process of step S103, there is an embodiment in which the weight value corresponding to the income level interval is obtained based on the sample weight algorithm network. The specific steps are:
[0060] S1031, the customer qualification identification model assigns the same initial weight value to each income level interval;
[0061] S1032. The customer qualification identification model adjusts the initial weight value of each income level interval multiple times according to the pre-stored income data of the sample customers, until the weight value corresponding to the income level interval corresponds to the graded proportion of the income data of the sample customers.
[0062] In steps S1031-S1032, the customer qualification identification model assigns the same initial weight value to each income level interval. For the convenience of description, the customer qualification identification model divides the income levels into ten levels from 0 to 9, so the sum of multiple initial weight values is set to 1 as the initial weight value of each income level interval. 1 is evenly divided, that is, the original proportion of each income level interval is 10%. The customer qualification identification model is trained. It can be seen from Table 2 that in the first prediction distribution model_1 of the customer qualification identification model, the weight proportions of the three income levels 0, 1, and 2 are too high, and the weight proportions of the three income levels 7, 8, and 9 are too low. Therefore, it is necessary to adjust the weights of the income level intervals of the income data of the sample customers, and set the weights of the income levels 0, 1, and 2 in the original sample customers to be lowered, and the weights of the income levels 7, 8, and 9 to be higher. The customer qualification identification model is trained again to obtain the distribution of the second prediction distribution model_2 of the customer qualification identification model. After one adjustment, The weights of the three income levels 0, 1, and 2 are still too high, and the weights of the three income levels 7, 8, and 9 are still too low. Therefore, the sample weights are adjusted again, and the weight settings of the three income levels 0, 1, and 2 continue to be lowered, and the weight settings of the three income levels 7, 8, and 9 continue to be increased. In the third predicted distribution model_3 of the customer qualification identification model, the distribution of the third predicted distribution Model_3 is opposite to the previous one. The weights of the three income levels 0, 1, and 2 are too low, and the weights of the three income levels 7, 8, and 9 are too high. Therefore, the sample weights are continued to be adjusted, and the weights of the three income levels 0, 1, and 2 are increased, and the weights of the three income levels 7, 8, and 9 are decreased. And so on, by comparing the difference between the predicted distribution and the graded proportion of the income data of the original sample customers, the sample weights are continuously adjusted, and finally the error between the predicted distribution of model_N and the graded proportion of the income data of the original sample customers is lower than the preset error threshold, wherein the error threshold can be specifically set according to the actual situation.
[0063] Table 2 Weight adjustment process of customer qualification identification model
[0064]
[0065] In the specific implementation process of step S1032, there is an embodiment in which the customer qualification identification model adjusts the initial weight value of each income level interval multiple times according to the pre-stored income data of sample customers, including:
[0066] S10321. After roughly adjusting and iterating the initial weight value of each income level interval with a preset step length, determine whether the weight value after the rough adjustment and iteration corresponds to the graded proportion of the sample customers;
[0067] S10322. If not, the preset step size is halved and the process is repeated until the weight value corresponding to the income level interval corresponds to the graded proportion of the sample customers.
[0068] In steps S10321-S10322, 0.1, which is 10% of each income level interval, is used as the preset step size for rough adjustment. That is, when the distribution is too large, the weight of the corresponding income level interval is reduced by 0.1, and vice versa, the weight of the income level interval with a small distribution is increased by 0.1, and the iteration is continued to observe the results. When the classification with a high proportion of the original sample is adjusted to a low level, the step size needs to be changed to adjust the weight of the income level interval, and the step size becomes 0.05, that is, the weight of the low income level interval is increased by 0.05, and vice versa. Under the step size adjustment of 0.05, if the original high classification is now low, then the step size is adjusted again, and the step size is changed to 0.025 for adjustment. Continuously adjust the sample weights, and continuously change the change step size of the income level interval weights, so that the distribution of the final model prediction results is roughly equal to the weight of the income level interval, and then the end can be achieved. The customer qualification identification model of the Nth model_N obtained in the end is the final customer qualification identification model.
[0069] In step S104, based on the income grade interval into which the maximum predicted score of each sample customer falls, the number of sample customers in each income grade interval is obtained, and the number ratio of sample customers in each income grade interval is compared with the graded proportion of sample customers to evaluate the customer qualification identification model. The evaluation results include pass and fail. The sample customers are identified one by one based on the customer qualification identification model with a pass evaluation result after evaluation, thereby identifying the customer qualifications of the sample customers and ensuring the accuracy of the prediction.
[0070] In the specific implementation process of step S104, there is an embodiment in which the customer qualification identification model is evaluated based on the maximum predicted score of each sample customer and the graded proportion of the sample customers, including:
[0071] S1041, performing a two-dimensional cross-section of the maximum predicted score of each sample customer and the graded proportion of the sample customers to obtain a score table;
[0072] S1042: Calculate the accuracy of the customer qualification identification model based on the score table to evaluate the customer qualification identification model based on the accuracy.
[0073] In steps S1041-S1042, the number of maximum predicted scores of each sample customer falling into each income grade interval divided by the customer qualification identification model is counted, that is, one maximum predicted score represents one sample customer. At the same time, the graded proportion of sample customers, that is, the actual number of sample customers in each income grade interval and the number of maximum predicted scores of each sample customer falling into each income grade interval divided by the customer qualification identification model are two-dimensionally crossed to obtain a score table as shown in Table 3. In Table 3, the horizontal axis is the graded proportion of sample customers, and the vertical axis is the number of maximum predicted scores of each sample customer falling into each income grade interval divided by the customer qualification identification model. The accuracy of the customer qualification identification model is calculated based on the horizontal and vertical axes in the score table to evaluate the customer qualification identification model based on the accuracy, thereby improving the accuracy of the evaluated customer qualification identification model.
[0074] Table 3 Score table
[0075]
[0076] In the specific implementation process of step S1042, there is an embodiment in which the accuracy of the customer qualification identification model is calculated based on the score table to evaluate the customer qualification identification model based on the accuracy, including:
[0077] S104211. Select first sample customers whose number of maximum predicted scores is the same as the graded proportion of sample customers from the diagonal of the score table, and determine the number of first sample customers;
[0078] S104212. Determine the accuracy of the customer qualification identification model based on the proportion of the number of first sample customers to the total number of sample customers.
[0079] In steps S104211-S104212, the first sample customers whose number of maximum predicted scores is the same as the graded proportion of sample customers are selected from the diagonal of the score table. The completely overlapping values in the diagonal of the score table are the number of real incomes of sample customers accurately predicted by the customer qualification identification model. The sample customers accurately predicted in this part are the first sample customers, i.e., the numbers indicated by shaded areas in Table 3. The number of first sample customers is counted, and the proportion of first sample customers in sample customers is calculated. The accuracy of the customer qualification identification model in predicting the real income of sample customers is determined by the proportion.
[0080] In the specific implementation process of step S1042, there is an embodiment in which: the customer qualification identification model is evaluated based on accuracy, including:
[0081] S104221. Determine whether the accuracy of the customer qualification identification model reaches the preset accuracy of the customer qualification identification model;
[0082] S104222. If not, it is impossible to identify the customer qualifications based on the evaluated customer qualification identification model.
[0083] In steps S104221-S104222, a preset accuracy rate is set for the customer qualification identification model, and the preset accuracy rate is set specifically according to actual conditions. The higher the preset accuracy rate, the higher the accuracy rate of the customer qualification identification model. When the accuracy rate of the customer qualification identification model reaches the preset accuracy rate set for the customer qualification identification model, the evaluation result of the customer qualification identification model is passed, and the customer qualification identification model can be used to identify the customer's qualifications, wherein the customer's qualifications are determined by the customer's actual income. If not, the customer's qualifications cannot be identified based on the evaluated customer qualification identification model, and the customer qualification identification model is trained again.
[0084] Example 2
[0085] This application also provides a model training device for identifying customer qualifications, such as Figure 2 As shown, the device comprises:
[0086] An acquisition module is used to acquire income data of multiple sample customers, and pre-process the income data of multiple sample customers to obtain multiple model input features and classification ratios; the pre-processing includes filling, conversion and elimination; the classification ratio is the ratio of the number of sample customers corresponding to each income level interval after the income of multiple sample customers is divided into levels;
[0087] A sending module is used to send the obtained multiple input features to a pre-established customer qualification identification model; the customer qualification identification model is pre-established by using a multi-classification neural network and income data of sample customers, and is optimized by a parameter adjustment network;
[0088] A receiving module, used for the customer qualification identification model to obtain multiple prediction scores based on the received multiple input features to select the maximum prediction score; the prediction score corresponds to the weight value of each income level interval given by the customer qualification identification model based on the sample weight algorithm network;
[0089] The identification module is used to evaluate the customer qualification identification model based on the maximum predicted score of each sample customer and the graded proportion of the sample customers, so as to identify the customer qualifications based on the evaluated customer qualification identification model.
[0090] The model training device for identifying customer qualifications provided in this application can achieve the same technical effect as the above-mentioned model training method for identifying customer qualifications, and will not be repeated here.
[0091] The receiving module comprises:
[0092] An assigning module, used for the customer qualification identification model to assign the same initial weight value to each income level interval;
[0093] The storage module is used for the customer qualification identification model to adjust the initial weight value of each income level interval multiple times according to the pre-stored income data of sample customers, until the weight value corresponding to the income level interval corresponds to the distribution ratio of the income data of the sample customers.
[0094] The receiving module further includes:
[0095] A rough adjustment module is used to perform rough adjustment and iteration for the initial weight value of each income level interval with a preset step size, and then determine whether the weight value after rough adjustment and iteration corresponds to the graded proportion of sample customers;
[0096] The halving module is used to cyclically halve the preset step size and re-roughly adjust and iterate if no, until the weight value corresponding to the income level interval corresponds to the graded proportion of the sample customers.
[0097] The sending module comprises:
[0098] A construction module is used to construct an initial customer qualification recognition model using the income data of sample customers and a multi-classification neural network, and the income data of sample customers is used as a target label;
[0099] The training module is used to train and optimize the initial customer qualification identification model multiple times based on the parameter adjustment network and the target label to obtain the customer qualification identification model.
[0100] The identification module comprises:
[0101] A crossover module is used to perform a two-dimensional crossover of the maximum predicted score of each sample customer and the graded proportion of the sample customers to obtain a score table;
[0102] The calculation module is used to calculate the accuracy of the customer qualification identification model based on the score table, so as to evaluate the customer qualification identification model based on the accuracy.
[0103] The identification module further comprises:
[0104] A determination module, used for selecting first sample customers whose number of maximum predicted scores is the same as the graded proportion of sample customers from the diagonal of the score table, and determining the number of first sample customers;
[0105] The occupation module is used to determine the accuracy of the customer qualification identification model based on the proportion of the number of first sample customers to the sample customers.
[0106] The identification module further includes:
[0107] A judgment module, used to judge whether the accuracy of the customer qualification identification model reaches the preset accuracy of the customer qualification identification model;
[0108] An evaluation module is used for, if not, it is impossible to identify the customer qualifications based on the evaluated customer qualification identification model.
[0109] Example 3
[0110] The present application also provides an electronic device, comprising: a processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor and the memory communicate through the bus, and when the machine-readable instructions are executed by the processor, the steps of a model training method for identifying customer qualifications as described in any one of the above are performed.
[0111] Example 4
[0112] The present application also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the computer program executes any one of the steps of the model training method for identifying customer qualifications.
[0113] Those skilled in the art can clearly understand that, for the convenience and simplicity of description, the specific working process of the system and device described above can refer to the corresponding process in the method embodiment, and will not be repeated in this application. In the several embodiments provided in this application, it should be understood that the disclosed system, device and method can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the modules is only a logical function division. There may be other division methods in actual implementation. For example, multiple modules 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 communication interfaces, indirect coupling or communication connection of devices or modules, which can be electrical, mechanical or other forms.
[0114] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
[0115] 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.
[0116] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a non-volatile computer-readable storage medium that is executable by a processor. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art or the 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 platform 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: various media that can store program codes, such as USB flash drives, mobile hard drives, ROM, RAM, magnetic disks, or optical disks.
[0117] The above are only specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.
Claims
1. A model training method for identifying customer qualifications, characterized in that: The method comprises: Obtaining income data of multiple sample customers, and preprocessing the income data of the multiple sample customers to obtain multiple model input features and classification ratios; the preprocessing includes filling, conversion and elimination; the classification ratio is the ratio of the number of sample customers corresponding to each income level interval after the income of the multiple sample customers is classified into levels; The obtained multiple input features are sent to a pre-established customer qualification identification model; the customer qualification identification model is pre-established by using a multi-classification neural network and sample customer income data, and is optimized by a parameter adjustment network; The customer qualification identification model obtains multiple prediction scores based on the received multiple input features to select the maximum prediction score; the prediction score corresponds to the weight value assigned by the customer qualification identification model to each income level interval based on the sample weight algorithm network; The customer qualification identification model is evaluated based on the maximum predicted score of each sample customer and the graded proportion of the sample customers, so as to identify the customer qualifications based on the evaluated customer qualification identification model.
2. The model training method for identifying customer qualifications according to claim 1, characterized in that: The weight values corresponding to the income level intervals are obtained based on the sample weight algorithm network in the following specific steps: The customer qualification identification model assigns the same initial weight value to each income level interval; The customer qualification identification model adjusts the initial weight value of each income level interval multiple times according to the pre-stored income data of sample customers, until the weight value corresponding to the income level interval corresponds to the distribution ratio of the income data of the sample customers.
3. The model training method for identifying customer qualifications as claimed in claim 2, characterized in that: The customer qualification identification model adjusts the initial weight value of each income level interval multiple times according to the income data of the pre-stored sample customers, including: After roughly adjusting and iterating the initial weight value of each income level interval with a preset step length, it is determined whether the weight value after the rough adjustment and iteration corresponds to the graded proportion of the sample customers; If not, the preset step size is halved and the process is repeated until the weight value corresponding to the income level interval corresponds to the graded proportion of the sample customers.
4. The model training method for identifying customer qualifications according to claim 1, characterized in that: The specific steps of the customer qualification identification model constructed by a multi-classification neural network and sample customer income data and optimized by a parameter adjustment network are as follows: The initial customer qualification recognition model is constructed using the sample customer's income data and a multi-classification neural network, and the sample customer's income data is used as the target label; The initial customer qualification identification model is trained and optimized multiple times based on the parameter adjustment network and the target label to obtain the customer qualification identification model.
5. The model training method for identifying customer qualifications according to claim 1, characterized in that: The customer qualification identification model is evaluated based on the maximum predicted score of each sample customer and the graded proportion of the sample customers, including: Perform a two-dimensional cross-section of the maximum predicted score of each sample customer and the graded proportion of the sample customers to obtain a score table; The accuracy of the customer qualification identification model is calculated based on the score table to evaluate the customer qualification identification model based on the accuracy.
6. The model training method for identifying customer qualifications as claimed in claim 5, characterized in that: The calculating the accuracy of the customer qualification identification model based on the score table to evaluate the customer qualification identification model based on the accuracy includes: Selecting first sample customers whose number of maximum predicted scores is the same as the graded proportion of sample customers from the diagonal line of the score table, and determining the number of first sample customers; The accuracy of the customer qualification identification model is determined by the ratio of the number of first sample customers to the sample customers.
7. The model training method for identifying customer qualifications as claimed in claim 5, characterized in that: The step of evaluating the customer qualification identification model based on accuracy includes: Determining whether the accuracy of the customer qualification identification model reaches a preset accuracy of the customer qualification identification model; If not, it is impossible to identify the customer qualifications based on the evaluated customer qualification identification model.
8. A model training device for identifying customer qualifications, characterized in that: The device comprises: An acquisition module is used to acquire income data of multiple sample customers, and pre-process the income data of multiple sample customers to obtain multiple model input features and classification ratios; the pre-processing includes filling, conversion and elimination; the classification ratio is the ratio of the number of sample customers corresponding to each income level interval after the income of multiple sample customers is divided into levels; A sending module is used to send the obtained multiple input features to a pre-established customer qualification identification model; the customer qualification identification model is pre-established by using a multi-classification neural network and income data of sample customers, and is optimized by a parameter adjustment network; A receiving module, used for the customer qualification identification model to obtain multiple prediction scores based on the received multiple input features to select the maximum prediction score; the prediction score corresponds to the weight value of each income level interval given by the customer qualification identification model based on the sample weight algorithm network; The identification module is used to evaluate the customer qualification identification model based on the maximum predicted score of each sample customer and the graded proportion of the sample customers, so as to identify the customer qualifications based on the evaluated customer qualification identification model.
9. An electronic device, characterized in that: include: A processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor and the memory communicate via the bus, and when the machine-readable instructions are executed by the processor, the steps of a model training method for identifying customer qualifications as described in any one of claims 1 to 7 are performed.
10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed by a processor, executes the steps of a model training method for identifying customer qualifications as described in any one of claims 1 to 7.