A method and device for evaluating customer indicators

Abstract

The present application discloses a method and device for evaluating customer indicators, wherein the method comprises: obtaining designated information of target customers; performing feature processing on the designated information of target customers to obtain a feature vector corresponding to the target customers; inputting the feature vector corresponding to the target customers into the currently trained latest final evaluation model, processing the feature vector corresponding to the target customers by the currently latest final evaluation model to obtain a target indicator corresponding to the target customers; wherein the currently latest final evaluation model is obtained by weighting an initial evaluation model obtained by the current round of training and a final evaluation model obtained by the previous round of training using a weight adjustment coefficient corresponding to the current round of training in advance; the initial model obtained by the current round of training is obtained by training using the current training samples and the designated information of each current training sample and the corresponding weights; the weight corresponding to the current training sample is determined based on the misclassification loss corresponding to the current training sample and the product usage.

Description

Technical Field

[0001] The present application relates to the technical field of indicator evaluation, and in particular to a method and device for evaluating customer indicators. Background Art

[0002] In order to discover valuable customers or determine the risks of customers, it is often necessary to calculate corresponding indicators for customers, so as to reflect the value, risks, etc. of customers through indicators.

[0003] With the continuous development of neural network technology, corresponding models are now being used to evaluate customer indicators. However, as the business continues to develop, the accuracy of customer indicators analyzed by the model will become lower and lower. Therefore, when the accuracy of the existing model is too low, a new model will be rebuilt, and then the specified information of the current customer sample will be used to train the constructed model to obtain a new model.

[0004] However, since the cost of reconstructing the model is relatively high, reconstruction is generally performed only when the model accuracy is low. Therefore, the existing method usually makes the accuracy of the model in use relatively low. In addition, directly building a new model using the specified information of the current customer sample does not consider the historical information and the differences between different customer samples, so that the accuracy of the obtained model is not high enough. Summary of the invention

[0005] Based on the above-mentioned deficiencies of the prior art, the present application provides a method and device for evaluating customer indicators to solve the problem of relatively low accuracy of the existing evaluation methods.

[0006] In order to achieve the above objectives, this application provides the following technical solutions:

[0007] The first aspect of the present application provides a method for evaluating customer indicators, comprising:

[0008] Acquire designated information of target customers; wherein the designated information refers to information of target indicators used for evaluation;

[0009] Performing feature processing on the designated information of the target customer to obtain a feature vector corresponding to the target customer;

[0010] The feature vector corresponding to the target customer is input into the trained current latest final evaluation model, and the feature vector corresponding to the target customer is processed by the current latest final evaluation model to obtain the target indicator corresponding to the target customer; wherein the current latest final evaluation model is obtained by weighting the initial evaluation model obtained in this round of training and the final evaluation model obtained in the previous round of training in advance using the weight adjustment coefficient corresponding to this round of training; the initial model obtained in this round of training is trained using the current training samples and the specified information of each of the current training samples and the corresponding weights; the weight corresponding to the current training sample is determined based on the misclassification loss corresponding to the current training sample and the product usage.

[0011] Optionally, in the above-mentioned customer indicator evaluation method, the training method of the current latest final evaluation model includes:

[0012] Use the current stock of customers as current training samples;

[0013] Obtaining specified information of each of the current training samples;

[0014] If this round of training is the first round of training, the weight corresponding to each of the current training samples is determined by using the preset value of the misclassification loss and the product usage corresponding to each of the current training samples;

[0015] Using the specified information of the current training sample and the corresponding weights, the constructed evaluation model is trained to obtain a final evaluation model of the first round of training;

[0016] If the current round of training does not belong to the first round of training, the weights corresponding to the current training samples are determined based on the misclassification loss and product usage corresponding to the current training samples; wherein the misclassification loss corresponding to the training samples is the deviation between the predicted value of the current training samples obtained by the latest historical evaluation model and the actual value of the current training samples; the latest historical evaluation model is the final evaluation model obtained from the previous round of training; the greater the misclassification loss and product usage, the greater the weight corresponding to the current training samples;

[0017] Using the specified information of the current training sample and the corresponding weights, the latest historical evaluation model is trained to obtain an initial evaluation model for this round of training;

[0018] Predicting the error rate of the current training sample based on the initial evaluation model of the current round of training, and calculating the weight adjustment coefficient corresponding to the current round of training; wherein the smaller the error rate, the larger the weight adjustment coefficient;

[0019] The initial evaluation model of the current round of training and the latest historical evaluation model are weighted by using the weight adjustment coefficient corresponding to the current round of training to obtain the current latest final evaluation model.

[0020] Optionally, in the above-mentioned customer indicator evaluation method, determining the weight corresponding to each current training sample based on the misclassification loss and product usage corresponding to the current training sample includes:

[0021] For each of the current training samples, if the current training sample is an old sample, calculate the square of the difference between the predicted value of the current training sample by the latest historical evaluation model and the actual value of the current training sample to obtain the misclassification loss corresponding to the current training sample;

[0022] The weight corresponding to the current training sample is obtained by multiplying the amplification score corresponding to the numerical interval of the misclassification loss corresponding to the current training sample by the amplification score corresponding to the user category to which the product usage corresponding to the current training sample belongs;

[0023] If the current training sample is a new sample, a weighted mean is calculated based on the weights corresponding to the old samples;

[0024] The weight mean is determined as the weight corresponding to the current training sample.

[0025] Optionally, in the above-mentioned method for evaluating customer indicators, the weighted average is calculated based on the weights corresponding to the old samples, including:

[0026] Calculate the sum of the average value of the weights corresponding to the current training samples corresponding to the customers in the old samples that have issued business opportunities and the average value of the weights corresponding to the current training samples corresponding to the customers in the old samples that have not issued business opportunities, to obtain a total average value of the weights;

[0027] The weight mean is obtained by multiplying the average value of the weights by the proportion of the customers who have not been issued business opportunities in the old samples.

[0028] Optionally, in the above-mentioned method for evaluating customer indicators, the initial evaluation model based on the current round of training predicts the error rate of the current training sample, and calculates the weight adjustment coefficient corresponding to the current round of training, including:

[0029] Sorting the prediction values ​​of the initial evaluation model of the current round of training for each of the current training samples from large to small to obtain a sorting result;

[0030] Calculating the error rate of the predicted values ​​in the top preset percentage in the sorting results;

[0031] Calculating the logarithm of the difference between the reciprocal of the error rate and 1 to obtain a logarithmic value;

[0032] Divide the logarithmic value by 2 to obtain the weight adjustment coefficient corresponding to the current round of training.

[0033] Optionally, in the above-mentioned customer indicator evaluation method, the weight adjustment coefficient corresponding to the current round of training is used to weight the initial evaluation model of the current round of training and the latest historical evaluation model to obtain the current latest final evaluation model, including:

[0034] The product of the initial evaluation model of this round of training and the first weight is added to the product of the latest historical evaluation model and the second weight to obtain the current latest final evaluation model; wherein the first weight is the ratio of the weight adjustment coefficient corresponding to the current round of training to the weight adjustment coefficient corresponding to the current round of training plus one; the second weight is the ratio of 1 to the weight adjustment coefficient corresponding to the current round of training plus one.

[0035] A second aspect of the present application provides a device for evaluating customer indicators, including:

[0036] A first acquisition unit is used to acquire designated information of a target customer; wherein the designated information refers to information of a target indicator used for evaluation;

[0037] A feature processing unit, used to perform feature processing on the designated information of the target customer to obtain a feature vector corresponding to the target customer;

[0038] An input unit is used to input the feature vector corresponding to the target customer into the trained current latest final evaluation model, and process the feature vector corresponding to the target customer through the current latest final evaluation model to obtain the target indicator corresponding to the target customer; wherein the current latest final evaluation model is obtained by weighting the initial evaluation model obtained in this round of training and the final evaluation model obtained in the previous round of training in advance using the weight adjustment coefficient corresponding to this round of training; the initial model obtained in this round of training is trained using the current training samples and the specified information of each of the current training samples and the corresponding weights; the weight corresponding to the current training sample is determined based on the misclassification loss corresponding to the current training sample and the product usage.

[0039] Optionally, in the above-mentioned client indicator evaluation device, it further includes:

[0040] A sample determination unit, used for taking the current stock of customers as current training samples;

[0041] A second acquisition unit, used to acquire specified information of each current training sample;

[0042] A first weight determination unit, configured to determine a weight corresponding to each of the current training samples by using a preset misclassification loss value and a product usage amount corresponding to each of the current training samples if the current round of training is the first round of training;

[0043] A first training unit is used to train the constructed evaluation model using the specified information of the current training sample and the corresponding weights to obtain a final evaluation model of the first round of training;

[0044] The second weight determination unit is used to determine the weights corresponding to each current training sample based on the misclassification loss and product usage corresponding to the current training sample if the current round of training does not belong to the first round of training; wherein the misclassification loss corresponding to the training sample is the deviation between the predicted value of the current training sample obtained by the latest historical evaluation model and the actual value of the current training sample; the latest historical evaluation model is the final evaluation model obtained by the previous round of training; the greater the misclassification loss and product usage, the greater the weight corresponding to the current training sample;

[0045] A second training unit is used to train the latest historical evaluation model using the specified information of the current training sample and the corresponding weights to obtain an initial evaluation model for this round of training;

[0046] A coefficient calculation unit, used to predict the error rate of the current training sample based on the initial evaluation model of the current round of training, and calculate the weight adjustment coefficient corresponding to the current round of training; wherein, the smaller the error rate, the larger the weight adjustment coefficient;

[0047] The weighting unit is used to weight the initial evaluation model of the current round of training and the latest historical evaluation model using the weight adjustment coefficient corresponding to the current round of training to obtain the current latest final evaluation model.

[0048] Optionally, in the above-mentioned client indicator evaluation device, the second weight determination unit includes:

[0049] A first calculation unit is used to calculate the square of the difference between the predicted value of the current training sample by the latest historical evaluation model and the actual value of the current training sample if the current training sample is an old sample, so as to obtain the misclassification loss corresponding to the current training sample;

[0050] A second calculation unit is used to multiply the amplification score corresponding to the numerical interval of the misclassification loss corresponding to the current training sample by the amplification score corresponding to the user category to which the product usage corresponding to the current training sample belongs, to obtain a weight corresponding to the current training sample;

[0051] A third calculation unit, configured to calculate a weighted mean based on the weights corresponding to the old samples if the current training sample is a new sample;

[0052] The second weight determination subunit is used to determine the weight mean as the weight corresponding to the current training sample.

[0053] Optionally, in the above-mentioned client indicator evaluation device, the third calculation unit includes:

[0054] A fourth calculation unit is used to calculate the sum of the average value of the weights corresponding to the current training samples corresponding to the customers who have issued business opportunities in the old samples and the average value of the weights corresponding to the current training samples corresponding to the customers who have not issued business opportunities in the old samples, to obtain a total average value of the weights;

[0055] The fifth calculation unit is used to multiply the average value of the weights by the proportion of the number of customers who have not been issued business opportunities in the old samples to obtain the mean value of the weights.

[0056] Optionally, in the above-mentioned client indicator evaluation device, the coefficient unit includes:

[0057] A sorting unit, used to sort the prediction values ​​of the initial evaluation model of the current round of training for each of the current training samples from large to small to obtain a sorting result;

[0058] An error rate calculation unit, used to calculate the error rate of the predicted values ​​in the top preset percentage in the sorting results;

[0059] a sixth calculation unit, configured to calculate the logarithm of the difference between the reciprocal of the error rate and 1 to obtain a logarithmic value;

[0060] The seventh calculation unit is used to divide the logarithmic value by 2 to obtain the weight adjustment coefficient corresponding to the current round of training.

[0061] Optionally, in the above-mentioned client indicator evaluation device, the weighting unit includes:

[0062] A weighting subunit is used to add the product of the initial evaluation model of the current round of training and the first weight, and the product of the latest historical evaluation model and the second weight, to obtain the current latest final evaluation model; wherein the first weight is the ratio of the weight adjustment coefficient corresponding to the current round of training to the weight adjustment coefficient corresponding to the current round of training plus one; the second weight is the ratio of 1 to the weight adjustment coefficient corresponding to the current round of training plus one.

[0063] The present application provides an evaluation of customer indicators, obtains the designated information of the target indicator for evaluation of the target customer, then performs feature processing on the designated information of the target customer, obtains the feature vector corresponding to the target customer, and finally inputs the feature vector corresponding to the target customer into the trained current latest final evaluation model, processes the feature vector corresponding to the target customer through the current latest final evaluation model, and obtains the target indicator corresponding to the target customer. Since the current latest final evaluation model pre-uses the weight adjustment coefficient corresponding to the current round of training, the initial evaluation model obtained by the current round of training and the final evaluation model obtained by the previous round of training are weighted, and are no longer directly reconstructed, so the previous historical information is effectively taken into account, which can effectively improve the accuracy of the model. In addition, the misclassification loss and product usage are also introduced to calculate the weight corresponding to the current training sample, and then the current training sample and the designated information of each current training sample and the corresponding weight training are used to obtain the initial model obtained by the current round of training, and the difference of the samples is considered, so that the initial model obtained by the current round of training can learn more information from the error instance, and thus the accuracy is higher, and more attention can be paid to valuable customers. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0065] Figure 1 A flowchart of a currently latest final evaluation model training method provided in an embodiment of the present application;

[0066] Figure 2 A flowchart of a method for determining a weight corresponding to a current training sample provided in an embodiment of the present application;

[0067] Figure 3 A flowchart of a method for calculating a weighted mean provided in an embodiment of the present application;

[0068] Figure 4 A flowchart of a method for calculating a weight adjustment coefficient corresponding to the current round of training provided in an embodiment of the present application;

[0069] Figure 5 A flowchart of a method for evaluating customer indicators provided in an embodiment of the present application;

[0070] Figure 6 A schematic diagram of the structure of a device for evaluating customer indicators provided in an embodiment of the present application. DETAILED DESCRIPTION

[0071] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions 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 are within the scope of protection of this application.

[0072] In this application, relational terms such as first and second, etc. are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the statement "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or device including the element.

[0073] The present application provides a method for evaluating customer indicators, which solves the problem of relatively low accuracy of existing evaluation methods.

[0074] First of all, it should be explained that the customer indicator evaluation method provided in this application obtains the current latest final evaluation model through continuous iteration, and then evaluates the customer indicators using the current final evaluation model.

[0075] Optionally, in order to implement the customer indicator evaluation method provided in the present application, the present application embodiment provides a current latest final evaluation model training method, such as Figure 1 As shown, the following steps are included:

[0076] S101. Use the current stock of customers as current training samples.

[0077] It should be noted that the customers of an enterprise will change over time, so the training samples of each iteration training round are the latest, so that the trained model is more suitable for the current customers, that is, the model has the best effect. Among them, each customer is regarded as a current training sample, that is, one customer corresponds to one training sample.

[0078] S102: Obtain designated information of each current training sample.

[0079] The designated information refers to the information of the target indicators designated for evaluation, such as the basic information of the customer, transaction information, loan information, etc., which is selected from the factors affecting the target designation according to the different target indicators. The target indicators are the indicator values ​​designated for evaluating the customer value or risk, such as risk level, contract signing probability, value level, etc.

[0080] S103: Determine whether this round of training is the first round of training.

[0081] Since in this application, misclassification loss and product usage are introduced to determine the weight corresponding to the current training sample, the model focuses on the samples corresponding to customers whose classification errors are greatly different from the actual ones and whose value is higher, that is, customers with larger product usage, during the learning process. However, the misclassification loss refers to the misclassification loss of the model, and during the first round of training, the constructed evaluation model has not been trained and used, so there is no misclassification loss. Therefore, in an embodiment of the present application, a preset value of the misclassification loss will be set for the first round of training, that is, a preset value will be used to replace the misclassification loss. Therefore, it is necessary to first determine whether this round of training is the first round of training. If it is the first round of training, execute step S104. If this round of training is not the first round of training, execute step S106.

[0082] S104: Determine the weight corresponding to each current training sample by using the preset value of the misclassification loss and the product usage corresponding to each current training sample.

[0083] Optionally, the preset value of the misclassification loss can be set to 1. Optionally, during training, the misclassification loss is usually used to determine the amplification score of the misclassification loss, and then the weight corresponding to the sample is determined, so the preset value of the misclassification loss can be considered as the initial value of the method score of the misclassification loss.

[0084] Optionally, the product of the preset misclassification loss value and the product usage amount corresponding to each current training sample may be determined as the weight corresponding to each current training sample.

[0085] S105: Use the specified information of the current training sample and the corresponding weights to train the constructed evaluation model to obtain the final evaluation model of the first round of training.

[0086] Since this is the first round of training, the previously untrained model is superimposed on the currently trained model, so the trained model is directly used as the final evaluation model.

[0087] S106: Determine the weights corresponding to each current training sample based on the misclassification loss corresponding to the current training sample and the product usage.

[0088] Among them, the misclassification loss corresponding to the training sample is the deviation between the predicted value of the current training sample obtained by the latest historical evaluation model and the actual value of the current training sample.

[0089] The latest historical evaluation model obtains the predicted value of the current training sample, that is, the predicted value obtained when the latest historical evaluation model makes a prediction for the customer corresponding to the current training sample.

[0090] It should be noted that the latest historical evaluation model is the final evaluation model obtained from the previous round of training. In addition, the greater the misclassification loss, the greater the weight corresponding to the current training sample, so that the training process focuses on training samples with large misclassification losses, so that more information can be learned from the wrong examples, thereby more effectively improving the accuracy of the evaluation model.

[0091] Similarly, the greater the usage of a product, the greater the weight corresponding to the current training sample, so that the training process focuses on training samples with large product usage. Since product usage reflects the value of the customer, the model can more accurately mine high-value customers.

[0092] Optionally, the product of the misclassification loss corresponding to the current training sample and the corresponding product usage amount may be used as the weight corresponding to the current training sample.

[0093] Optionally, another embodiment of the present application provides a specific implementation of step S106, such as Figure 2 As shown, including:

[0094] S201: For each current training sample, determine whether the current training sample is an old sample.

[0095] It should be noted that old samples refer to samples that have been used for training in the previous round of training, that is, the customers corresponding to the samples in the previous round of training are already customers in the target customer set. New samples are samples that are newly added for training in this round of training, that is, the users corresponding to the samples are included in the target customer set in this round of training.

[0096] Since the corresponding misclassification loss and product usage cannot be obtained for the new sample, it is necessary to rely on the corresponding misclassification loss and product usage of the old sample for determination. Therefore, when the current training sample is determined to be a new sample, step S204 is executed. If the current training sample is determined to be an old sample, step S202 is executed.

[0097] S202: Calculate the square of the difference between the predicted value of the current training sample by the latest historical evaluation model and the actual value of the current training sample to obtain the misclassification loss corresponding to the current training sample.

[0098] S203: multiply the amplification score corresponding to the numerical interval of the misclassification loss corresponding to the current training sample by the amplification score corresponding to the user category to which the product usage corresponding to the current training sample belongs, to obtain a weight corresponding to the current training sample.

[0099] Optionally, in an embodiment of the present application, in order to facilitate the processing of misclassification losses and product usage, in an embodiment of the present application, a binning method is used to perform normalization processing, and then the weights corresponding to the training samples are determined based on the binning results.

[0100] Specifically, the misclassification loss is divided into multiple intervals according to equal intervals, and each interval corresponds to an amplification score. For example, as shown in Table 1, the misclassification loss is divided into four intervals: [0, 0.25], [0.25, 0.5], [0.5, 0.75], [0.75, 1], each interval corresponds to β, 2β, 3β: 4β, which can be determined by the accuracy of the latest historical evaluation model.

[0101] Table 1

[0102]

[0103] Similarly, customers are divided into multiple categories based on their product usage, and each category corresponds to an amplification score. Figure 1 As shown, customers are divided into active customers, customers who use products, customers who do not use products, and customers who have not signed contracts according to the amount of product usage. These four types correspond to: β, 2β, 3β: 4β. The weight corresponding to the training sample is the product of the two amplification scores, so the weight corresponding to the training sample finally obtained can be shown in Table 1. Of course, other methods can also be used to determine the weight corresponding to the training sample based on the misclassification loss and the amount of product usage, which should also fall within the scope of protection of this application.

[0104] S204: Calculate the weight mean based on the weights corresponding to the current training samples corresponding to the old samples.

[0105] It should be noted that the calculated weighted mean may be an arithmetic mean or a weighted mean.

[0106] Optionally, another embodiment of the present application provides a specific implementation of step S204, such as Figure 3 As shown, the following steps are included:

[0107] S301, calculating the sum of the average value of the weights corresponding to the current training samples corresponding to the customers who have issued business opportunities in the old samples and the average value of the weights corresponding to the current training samples corresponding to the customers who have not issued business opportunities in the old samples, to obtain the total average value of the weights.

[0108] Among them, sending business opportunity customers can be understood as executing actions to promote the customer to become a conversion customer, such as pushing advertisements.

[0109] S302: Multiply the average value of the weights by the proportion of customers to whom business opportunities have not been issued in the old samples to obtain the mean value of the weights.

[0110] Since the total average of the weights of old customers can reflect the weights of new samples, and new customers are customers who have not been issued business opportunities, the average of the weights is multiplied by the proportion of customers who have not been issued business opportunities in the old samples as the weight mean. Then, step S205 is executed to determine the weight mean as the weight corresponding to the current training sample.

[0111] S205: Determine the weight mean as the weight corresponding to the current training sample.

[0112] S107: Use the specified information of the current training sample and the corresponding weights to train the latest historical evaluation model to obtain an initial evaluation model for this round of training.

[0113] S108. Predict the error rate of the current training sample based on the initial evaluation model of this round of training, and calculate the weight adjustment coefficient corresponding to this round of training.

[0114] Among them, the smaller the error rate, the larger the weight adjustment coefficient, so that the final evaluation model obtained after the final weighted summation is more inclined to the model with a low error rate in the initial evaluation model of this round of training and the latest historical evaluation model, thereby ensuring the effect of the final evaluation model.

[0115] Optionally, another embodiment of the present application provides a specific implementation of step S108, such as Figure 4 As shown, including:

[0116] S401, sorting the prediction values ​​of the initial evaluation model of this round of training for each current training sample from large to small to obtain a sorting result.

[0117] S402: Calculate the error rate of the predicted values ​​that are in the first preset percentage in the sorting results.

[0118] It should be noted that since the results obtained from the evaluation model often focus on customers with relatively high predicted values, in this embodiment of the application, the error rate is only calculated for part of the results, that is, the error rate is only calculated for the predicted values ​​that are ranked in the front preset percentage, for example, the error rate is only calculated for the top 5% of the predicted values.

[0119] S403 . Calculate the logarithm of the difference between the reciprocal of the error rate and 1 to obtain a logarithmic value.

[0120] S404: Divide the logarithmic value by 2 to obtain the weight adjustment coefficient corresponding to this round of training.

[0121] S109: Using the weight adjustment coefficient corresponding to this round of training, the initial evaluation model of this round of training and the latest historical evaluation model are weighted to obtain the current latest final evaluation model.

[0122] It should be noted that since the initial evaluation model of this round of training is the latest model, when performing weighted summation, the larger the weight adjustment coefficient, the greater the proportion of the initial evaluation model of this round of training. The current latest final evaluation model is the final evaluation model obtained in this round of training.

[0123] Optionally, a specific implementation of step S109 is:

[0124] The product of the initial evaluation model of this round of training and the first weight is added to the product of the latest historical evaluation model and the second weight to obtain the current final evaluation model.

[0125] The first weight is the ratio of the weight adjustment coefficient corresponding to the current round of training to the weight adjustment coefficient corresponding to the current round of training plus 1. The second weight is the ratio of 1 to the weight adjustment coefficient corresponding to the current round of training plus 1.

[0126] Based on the latest final evaluation model obtained through the above training, the present application embodiment provides a method for evaluating customer indicators, such as Figure 5 As shown, including:

[0127] S501. Obtain designated information of target customers.

[0128] The designated information refers to the information of the target indicator used for evaluation, such as the basic information of the customer, transaction information, loan information, etc., which is selected from the factors affecting the target designation according to the different target indicators. The target indicator is the indicator value designated for evaluating the customer value or risk, such as risk level, contract signing probability, value level, etc.

[0129] S502: Perform feature processing on the designated information of the target customer to obtain a feature vector corresponding to the target customer.

[0130] S503: Input the feature vector corresponding to the target customer into the trained latest final evaluation model, process the feature vector corresponding to the target customer through the latest final evaluation model, and obtain the target indicator corresponding to the target customer.

[0131] Among them, the current latest final evaluation model refers to the final evaluation model obtained by the latest round of training.

[0132] Among them, the current latest final evaluation model is obtained by weighting the initial evaluation model obtained in this round of training and the final evaluation model obtained in the previous round of training using the weight adjustment coefficient corresponding to this round of training in advance. The initial model obtained in this round of training is trained using the current training samples and the specified information of each of the current training samples and the corresponding weights. The weight corresponding to the current training sample is determined based on the misclassification loss corresponding to the current training sample and the product usage. It should be noted that the more specific training method of the current latest final evaluation model can be referred to accordingly. Figure 1 The illustrated embodiment.

[0133] Optionally, a new round of model training may be started at a preset time interval, or when the accuracy of the latest final evaluation model is greater than a threshold, a new round of model training may be started. Of course, other strategies may also be used to start new training.

[0134] The embodiment of the present application provides an evaluation of a customer index, obtains the designated information of the target index for evaluation of the target customer, then performs feature processing on the designated information of the target customer, obtains the feature vector corresponding to the target customer, and finally inputs the feature vector corresponding to the target customer into the trained current latest final evaluation model, processes the feature vector corresponding to the target customer by the current latest final evaluation model, and obtains the target index corresponding to the target customer. Since the current latest final evaluation model pre-uses the weight adjustment coefficient corresponding to the current round of training, the initial evaluation model obtained by the current round of training and the final evaluation model obtained by the previous round of training are weighted, and are no longer directly reconstructed, so the previous historical information is effectively taken into account, which can effectively improve the accuracy of the model. In addition, the misclassification loss and the product usage are also introduced to calculate the weight corresponding to the current training sample, and then the current training sample and the designated information of each current training sample and the corresponding weight training are used to obtain the initial model obtained by the current round of training, so that the initial model obtained by the current round of training can learn more information from the error instance, and thus the accuracy is higher, and more attention can be paid to valuable customers.

[0135] Another embodiment of the present application provides a device for evaluating customer indicators, such as Figure 6 As shown, including:

[0136] The first acquisition unit 601 is used to acquire designated information of target customers.

[0137] The designated information refers to the information of the target indicator used for evaluation.

[0138] The feature processing unit 602 is used to perform feature processing on the designated information of the target customer to obtain a feature vector corresponding to the target customer.

[0139] The input unit 603 is used to input the feature vector corresponding to the target customer into the trained and latest final evaluation model, and process the feature vector corresponding to the target customer through the latest final evaluation model to obtain the target indicator corresponding to the target customer.

[0140] Among them, the latest final evaluation model is obtained by weighting the initial evaluation model obtained in this round of training and the final evaluation model obtained in the previous round of training using the weight adjustment coefficient corresponding to this round of training in advance. The initial model obtained in this round of training is trained using the current training samples and the specified information of each current training sample and the corresponding weights. The weight corresponding to the current training sample is determined based on the misclassification loss corresponding to the current training sample and the product usage.

[0141] Optionally, in the client indicator evaluation device provided in another embodiment of the present application, the device further includes:

[0142] The sample determination unit is used to use the current stock of customers as current training samples.

[0143] The second acquisition unit is used to acquire the specified information of each current training sample.

[0144] The first weight determination unit is used to determine the weight corresponding to each current training sample by using the preset value of the misclassification loss and the product usage corresponding to each current training sample if the current round of training is the first round of training.

[0145] The first training unit is used to train the constructed evaluation model using the specified information of the current training sample and the corresponding weights to obtain the final evaluation model of the first round of training.

[0146] The second weight determination unit is used to determine the weights corresponding to each current training sample based on the misclassification loss and product usage corresponding to the current training sample if the current round of training does not belong to the first round of training. The misclassification loss corresponding to the training sample is the deviation between the predicted value of the current training sample obtained by the latest historical evaluation model and the actual value of the current training sample. The latest historical evaluation model is the final evaluation model obtained in the previous round of training. The greater the misclassification loss and product usage, the greater the weight corresponding to the current training sample.

[0147] The second training unit is used to train the latest historical evaluation model using the specified information of the current training sample and the corresponding weights to obtain an initial evaluation model for this round of training.

[0148] The coefficient calculation unit is used to predict the error rate of the current training sample based on the initial evaluation model of this round of training, and calculate the weight adjustment coefficient corresponding to this round of training. The smaller the error rate, the larger the weight adjustment coefficient.

[0149] The weighting unit is used to weight the initial evaluation model of this round of training and the latest historical evaluation model using the weight adjustment coefficient corresponding to this round of training to obtain the current latest final evaluation model.

[0150] Optionally, in the client indicator evaluation device provided in another embodiment of the present application, the second weight determination unit includes:

[0151] The first calculation unit is used to calculate the square of the difference between the predicted value of the current training sample by the latest historical evaluation model and the actual value of the current training sample if the current training sample is an old sample, so as to obtain the misclassification loss corresponding to the current training sample.

[0152] The second calculation unit is used to multiply the amplification score corresponding to the numerical interval of the misclassification loss corresponding to the current training sample by the amplification score corresponding to the user category to which the product usage corresponding to the current training sample belongs, to obtain the weight corresponding to the current training sample.

[0153] The third calculation unit is used to calculate the weight mean based on the weights corresponding to each old sample if the current training sample is a new sample.

[0154] The second weight determination subunit is used to determine the weight mean as the weight corresponding to the current training sample.

[0155] Optionally, in the client indicator evaluation device provided in another embodiment of the present application, the third calculation unit includes:

[0156] The fourth calculation unit is used to calculate the sum of the average weights of the current training samples corresponding to the customers with business opportunities in the old samples and the average weights of the current training samples corresponding to the customers with no business opportunities in the old samples, to obtain the total average weights.

[0157] The fifth calculation unit is used to multiply the average value of the weights by the proportion of customers to whom business opportunities have not been issued in the old samples to obtain the mean value of the weights.

[0158] Optionally, in the client indicator evaluation device provided in another embodiment of the present application, the coefficient unit includes:

[0159] The sorting unit is used to sort the prediction values ​​of the initial evaluation model of this round of training for each current training sample from large to small to obtain a sorting result.

[0160] The error rate calculation unit is used to calculate the error rate of the prediction value that is in the first preset percentage in the sorting result.

[0161] The sixth calculation unit is used to calculate the logarithm of the difference between the reciprocal of the error rate and 1 to obtain a logarithmic value.

[0162] The seventh calculation unit is used to divide the logarithmic value by 2 to obtain the weight adjustment coefficient corresponding to the current round of training.

[0163] Optionally, in the client indicator evaluation device provided in another embodiment of the present application, the weighting unit includes:

[0164] The weighting subunit is used to add the product of the initial evaluation model of this round of training and the first weight, and the product of the latest historical evaluation model and the second weight, to obtain the latest final evaluation model. Among them, the first weight is the ratio of the weight adjustment coefficient corresponding to this round of training to the weight adjustment coefficient corresponding to this round of training plus one. The second weight is the ratio of 1 to the weight adjustment coefficient corresponding to this round of training plus one.

[0165] It should be noted that the specific working process of each unit provided in the above embodiments of the present application can refer to the corresponding steps in the above method embodiments, which will not be repeated here.

[0166] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0167] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for evaluating customer indicators, characterized in that: include: Acquire designated information of target customers; wherein the designated information refers to information of target indicators used for evaluation; Performing feature processing on the designated information of the target customer to obtain a feature vector corresponding to the target customer; The feature vector corresponding to the target customer is input into the trained current latest final evaluation model, and the feature vector corresponding to the target customer is processed by the current latest final evaluation model to obtain the target indicator corresponding to the target customer; wherein, when the current round of training is the first round of training, the current latest final evaluation model is obtained by training the constructed evaluation model using the specified information of the current training sample and the corresponding weight; the weight corresponding to the current training sample is determined by the preset value of the misclassification loss and the product usage corresponding to the current training sample; when the current round of training is not the first round of training, the current latest final evaluation model is obtained by weighting the initial evaluation model obtained in the current round of training and the final evaluation model obtained in the previous round of training using the weight adjustment coefficient corresponding to the current round of training; the initial model obtained in the current round of training is obtained by training using the current training sample and the specified information and corresponding weights of each of the current training samples; the weight corresponding to the current training sample is determined based on the misclassification loss and product usage corresponding to the current training sample.

2. The method according to claim 1, characterized in that The current latest final evaluation model training method includes: Use the current stock of customers as current training samples; Obtaining specified information of each of the current training samples; If this round of training is the first round of training, the weight corresponding to each of the current training samples is determined by using the preset value of the misclassification loss and the product usage corresponding to each of the current training samples; Using the specified information of the current training sample and the corresponding weights, the constructed evaluation model is trained to obtain a final evaluation model of the first round of training; If the current round of training does not belong to the first round of training, the weights corresponding to the current training samples are determined based on the misclassification loss and product usage corresponding to the current training samples; wherein the misclassification loss corresponding to the training samples is the deviation between the predicted value of the current training samples obtained by the latest historical evaluation model and the actual value of the current training samples; the latest historical evaluation model is the final evaluation model obtained from the previous round of training; the greater the misclassification loss and product usage, the greater the weight corresponding to the current training samples; Using the specified information of the current training sample and the corresponding weights, the latest historical evaluation model is trained to obtain an initial evaluation model for this round of training; Predicting the error rate of the current training sample based on the initial evaluation model of the current round of training, and calculating the weight adjustment coefficient corresponding to the current round of training; wherein the smaller the error rate, the larger the weight adjustment coefficient; The initial evaluation model of the current round of training and the latest historical evaluation model are weighted by using the weight adjustment coefficient corresponding to the current round of training to obtain the current latest final evaluation model.

3. The method according to claim 2, characterized in that The determining, based on the misclassification loss and product usage corresponding to the current training samples, the weights corresponding to the current training samples, includes: For each of the current training samples, if the current training sample is an old sample, calculate the square of the difference between the predicted value of the current training sample by the latest historical evaluation model and the actual value of the current training sample to obtain the misclassification loss corresponding to the current training sample; The weight corresponding to the current training sample is obtained by multiplying the amplification score corresponding to the numerical interval of the misclassification loss corresponding to the current training sample by the amplification score corresponding to the user category to which the product usage corresponding to the current training sample belongs; If the current training sample is a new sample, a weighted mean is calculated based on the weights corresponding to the old samples; The weight mean is determined as the weight corresponding to the current training sample.

4. The method according to claim 3, characterized in that The calculating the weighted mean based on the weights corresponding to the old samples includes: Calculate the sum of the average value of the weights corresponding to the current training samples corresponding to the customers in the old samples that have issued business opportunities and the average value of the weights corresponding to the current training samples corresponding to the customers in the old samples that have not issued business opportunities, to obtain a total average value of the weights; The weight mean is obtained by multiplying the average value of the weights by the proportion of the customers who have not been issued business opportunities in the old samples.

5. The method according to claim 2, characterized in that: The initial evaluation model based on the current round of training predicts the error rate of the current training sample, and calculates the weight adjustment coefficient corresponding to the current round of training, including: Sorting the prediction values ​​of the initial evaluation model of the current round of training for each of the current training samples from large to small to obtain a sorting result; Calculating the error rate of the predicted values ​​in the top preset percentage in the sorting results; Calculating the logarithm of the difference between the reciprocal of the error rate and 1 to obtain a logarithmic value; Divide the logarithmic value by 2 to obtain the weight adjustment coefficient corresponding to the current round of training.

6. The method according to claim 2, characterized in that The weight adjustment coefficient corresponding to the current round of training is used to weight the initial evaluation model of the current round of training and the latest historical evaluation model to obtain the current latest final evaluation model, including: The product of the initial evaluation model of this round of training and the first weight is added to the product of the latest historical evaluation model and the second weight to obtain the current latest final evaluation model; wherein the first weight is the ratio of the weight adjustment coefficient corresponding to the current round of training to the weight adjustment coefficient corresponding to the current round of training plus one; the second weight is the ratio of 1 to the weight adjustment coefficient corresponding to the current round of training plus one.

7. A device for evaluating customer indicators, characterized in that: include: A first acquisition unit is used to acquire designated information of a target customer; wherein the designated information refers to information of a target indicator used for evaluation; A feature processing unit, used to perform feature processing on the designated information of the target customer to obtain a feature vector corresponding to the target customer; An input unit is used to input the feature vector corresponding to the target customer into the trained current latest final evaluation model, and process the feature vector corresponding to the target customer through the current latest final evaluation model to obtain the target indicator corresponding to the target customer; wherein, when the current round of training is the first round of training, the current latest final evaluation model is obtained by training the constructed evaluation model using the specified information of the current training sample and the corresponding weight; the weight corresponding to the current training sample is determined by using the preset value of the misclassification loss and the product usage corresponding to the current training sample; when the current round of training does not belong to the first round of training, the current latest final evaluation model is obtained by weighting the initial evaluation model obtained in the current round of training and the final evaluation model obtained in the previous round of training using the weight adjustment coefficient corresponding to the current round of training; the initial model obtained in the current round of training is obtained by training using the current training sample and the specified information and corresponding weights of each of the current training samples; the weight corresponding to the current training sample is determined based on the misclassification loss and product usage corresponding to the current training sample.

8. The device according to claim 7, characterized in that Also includes: A sample determination unit, used for taking the current stock of customers as current training samples; A second acquisition unit, used to acquire specified information of each current training sample; A first weight determination unit, configured to determine a weight corresponding to each of the current training samples by using a preset misclassification loss value and a product usage amount corresponding to each of the current training samples if the current round of training is the first round of training; A first training unit is used to train the constructed evaluation model using the specified information of the current training sample and the corresponding weights to obtain a final evaluation model of the first round of training; The second weight determination unit is used to determine the weights corresponding to each current training sample based on the misclassification loss and product usage corresponding to the current training sample if the current round of training does not belong to the first round of training; wherein the misclassification loss corresponding to the training sample is the deviation between the predicted value of the current training sample obtained by the latest historical evaluation model and the actual value of the current training sample; the latest historical evaluation model is the final evaluation model obtained by the previous round of training; the greater the misclassification loss and product usage, the greater the weight corresponding to the current training sample; A second training unit is used to train the latest historical evaluation model using the specified information of the current training sample and the corresponding weights to obtain an initial evaluation model for this round of training; A coefficient calculation unit, used to predict the error rate of the current training sample based on the initial evaluation model of the current round of training, and calculate the weight adjustment coefficient corresponding to the current round of training; wherein, the smaller the error rate, the larger the weight adjustment coefficient; The weighting unit is used to weight the initial evaluation model of the current round of training and the latest historical evaluation model using the weight adjustment coefficient corresponding to the current round of training to obtain the current latest final evaluation model.

9. The device according to claim 8, characterized in that The second weight determination unit includes: A first calculation unit is used to calculate the square of the difference between the predicted value of the current training sample by the latest historical evaluation model and the actual value of the current training sample if the current training sample is an old sample, so as to obtain the misclassification loss corresponding to the current training sample; A second calculation unit is used to multiply the amplification score corresponding to the numerical interval of the misclassification loss corresponding to the current training sample by the amplification score corresponding to the user category to which the product usage corresponding to the current training sample belongs, to obtain a weight corresponding to the current training sample; A third calculation unit, configured to calculate a weighted mean based on the weights corresponding to the old samples if the current training sample is a new sample; The second weight determination subunit is used to determine the weight mean as the weight corresponding to the current training sample.

10. The device according to claim 9, characterized in that The third computing unit comprises: A fourth calculation unit is used to calculate the sum of the average value of the weights corresponding to the current training samples corresponding to the customers who have issued business opportunities in the old samples and the average value of the weights corresponding to the current training samples corresponding to the customers who have not issued business opportunities in the old samples, to obtain a total average value of the weights; The fifth calculation unit is used to multiply the average value of the weights by the proportion of the number of customers who have not been issued business opportunities in the old samples to obtain the mean value of the weights.

Images