A method, device and storage medium for determining customer category

Abstract

The embodiments of this specification provide a method, device and storage medium for determining customer categories, which can be applied to the field of artificial intelligence technology. The method includes: obtaining a preset number of customer samples; the customer samples include positive samples and negative samples; the positive samples indicate that the customer category is a bad customer, and the negative samples indicate that the customer category is an ordinary customer; the negative samples are divided into multiple groups of negative samples, each group of negative samples is combined with the positive samples to form a sample subset, and multiple sample subsets are obtained; for different sample subsets, the feature vectors of the customer samples in each sample subset are extracted from different dimensions; based on the preset objective function, the feature vectors of the customer samples in each sample subset and the customer categories represented by the customer samples are fitted to obtain the classification conditions of the customer categories under each dimension; the category of the target customer is determined according to the classification conditions of the customer categories under each dimension, so as to improve the accuracy of the customer category determination.

Description

Technical Field

[0001] The embodiments of this specification relate to the field of artificial intelligence technology, and more particularly to a method, device and storage medium for determining a customer category. Background Art

[0002] With the development of the financial industry, the proportion of international balance of payments business in financial institutions is increasing. Due to the complexity of the international balance of payments business scenario, it is difficult to find bad customers in advance. If bad customers become more serious, it will have an adverse impact on financial institutions, resulting in a decline in the reputation of financial institutions and reduced profits. With the development of artificial intelligence technology, applying machine learning technology to classify and predict international balance of payments customers is a solution worth trying.

[0003] Machine learning technology can be used to model and learn from a large number of existing samples, and to predict unknown samples, thereby making predictions on the classification of international balance of payments customers.

[0004] However, the current mainstream machine learning methods still have shortcomings in the classification of international balance of payments customers, mainly in two aspects. First, the characteristics of samples in the international balance of payments scenario contain a lot of information, such as international trade information, enterprise information and business owner information. The existing mainstream methods directly splice these different categories of information into sample characteristics, which on the one hand leads to a particularly large feature dimension of the sample, which is easy to cause the "dimensionality disaster problem"; on the other hand, the feature scales and meanings of different categories of features are often quite different, and directly splicing them together is likely to make them lose their original physical meaning, resulting in the model failing to achieve the expected effect. Second, the existing mainstream methods regard the international balance of payments customer classification problem as a general category balance classification problem. In fact, in the international balance of payments customer classification scenario, the proportion of bad customers is often very low, and the number of positive and negative samples is extremely unbalanced. If it is treated as a category balance problem, the effect is not good. In this scenario, more attention should be paid to minority class samples, that is, the model should predict bad samples as accurately as possible while ensuring accuracy. Existing methods cannot meet this requirement. Summary of the invention

[0005] The purpose of the embodiments of this specification is to provide a method, device and storage medium for determining customer categories, so as to improve the accuracy of determining customer categories.

[0006] To solve the above problems, an embodiment of the present specification provides a method for determining customer categories, the method comprising: obtaining a preset number of customer samples; the customer samples include positive samples and negative samples; the positive samples indicate that the customer category is a bad customer, and the negative samples indicate that the customer category is an ordinary customer; the negative samples are divided into multiple groups of negative samples, each group of negative samples is combined with the positive samples to form a sample subset, and multiple sample subsets are obtained; for different sample subsets, feature vectors of customer samples in each sample subset are extracted from different dimensions; based on a preset objective function, the feature vectors of customer samples in each sample subset and the customer categories represented by the customer samples are fitted to obtain classification conditions for customer categories under each dimension; and the category of the target customer is determined according to the classification conditions for customer categories under each dimension.

[0007] To solve the above problems, an embodiment of the present specification also provides a customer category determination device, which includes: an acquisition module, used to acquire a preset number of customer samples; the customer samples include positive samples and negative samples; the positive samples indicate that the customer category is a bad customer, and the negative samples indicate that the customer category is an ordinary customer; a division module, used to divide the negative samples into multiple groups of negative samples, each group of negative samples is combined with the positive samples to form a sample subset, and multiple sample subsets are obtained; an extraction module, used to extract the feature vectors of the customer samples in each sample subset from different dimensions for different sample subsets; a fitting module, used to fit the feature vectors of the customer samples in each sample subset and the customer category represented by the customer samples based on a preset objective function, and obtain the classification conditions of the customer category under each dimension; a determination module, used to determine the category of the target customer according to the classification conditions of the customer category under each dimension.

[0008] To solve the above problems, an embodiment of the present specification also provides an electronic device, including: a memory for storing a computer program; a processor for executing the computer program to achieve: obtaining a preset number of customer samples; the customer samples include positive samples and negative samples; the positive samples indicate that the customer category is a bad customer, and the negative samples indicate that the customer category is an ordinary customer; the negative samples are divided into multiple groups of negative samples, each group of negative samples is combined with the positive samples to form a sample subset, and multiple sample subsets are obtained; for different sample subsets, the feature vectors of the customer samples in each sample subset are extracted from different dimensions; based on a preset objective function, the feature vectors of the customer samples in each sample subset and the customer category represented by the customer samples are fitted to obtain the classification conditions of the customer category under each dimension; the category of the target customer is determined according to the classification conditions of the customer category under each dimension.

[0009] To solve the above problems, the embodiments of the present specification also provide a computer-readable storage medium having computer instructions stored thereon, which when executed implement the following: obtaining a preset number of customer samples; the customer samples include positive samples and negative samples; the positive samples indicate that the customer category is a bad customer, and the negative samples indicate that the customer category is an ordinary customer; the negative samples are divided into multiple groups of negative samples, each group of negative samples is combined with the positive samples to form a sample subset, and multiple sample subsets are obtained; for different sample subsets, feature vectors of the customer samples in each sample subset are extracted from different dimensions; based on a preset objective function, the feature vectors of the customer samples in each sample subset and the customer category represented by the customer samples are fitted to obtain the classification conditions of the customer category under each dimension; and the category of the target customer is determined according to the classification conditions of the customer category under each dimension.

[0010] It can be seen from the technical solutions provided in the above embodiments of this specification that, in the embodiments of this specification, a preset number of customer samples can be obtained; the customer samples include positive samples and negative samples; the positive samples indicate that the customer category is a bad customer, and the negative samples indicate that the customer category is an ordinary customer; the negative samples are divided into multiple groups of negative samples, each group of negative samples is combined with the positive samples to form a sample subset, and multiple sample subsets are obtained; for different sample subsets, the feature vectors of the customer samples in each sample subset are extracted from different dimensions; based on the preset objective function, the feature vectors of the customer samples in each sample subset and the customer categories represented by the customer samples are fitted to obtain the classification conditions of the customer categories under each dimension; the category of the target customer is determined according to the classification conditions of the customer categories under each dimension. The method provided in the embodiments of this specification divides the negative samples and combines them with the positive samples into sample subsets, which is conducive to alleviating the imbalance of the classification problem, and divides the features of different categories into different dimensions, making full use of the complementary information of different dimensions, thereby improving the accuracy of customer category determination. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In order to more clearly illustrate the embodiments of this specification 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 some embodiments recorded in this specification. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0012] Figure 1 This is a schematic diagram of the training process of a classification model in a scenario example of this manual;

[0013] Figure 2 This is a schematic diagram of the testing process of a classification model in a scenario example of this specification;

[0014] Figure 3A flowchart of a method for determining a customer category according to an embodiment of the present specification;

[0015] Figure 4 This is a functional structure diagram of an electronic device according to an embodiment of this specification;

[0016] Figure 5 This is a functional structure diagram of a device for determining a customer category according to an embodiment of this specification. DETAILED DESCRIPTION

[0017] The following will be combined with the drawings in the embodiments of this specification to clearly and completely describe the technical solutions in the embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Based on the embodiments in this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this specification.

[0018] This specification provides a scenario example. In this scenario example, taking the customer as an international balance of payments customer as an example, the classification model can be trained and used to determine the category of the international balance of payments customer. Specifically, the customer can be determined to be a bad customer, that is, a customer with credit problems, or the customer can be determined to be an ordinary customer, that is, a customer with credit problems.

[0019] In this scenario example, the training process of the classification model is as follows Figure 1 As shown, the following steps may be specifically included: first, feature information related to international balance of payments customers is obtained from the data warehouse, and the features are divided into three dimensions according to categories, namely, international trade dimension, enterprise information dimension and enterprise owner information dimension, wherein the international trade dimension includes information such as total transaction volume, foreign-related income, overseas remittance amount and its increment, overseas remittance number and its increment, etc., the enterprise information dimension includes basic enterprise information, daily average assets and its year-on-year, enterprise account inflow and outflow amount, number and its year-on-year, etc., and the enterprise owner information dimension includes age, education level, whether it is an important customer, enterprise owner account inflow and outflow amount, number and its year-on-year, etc.; after data preprocessing and feature engineering processing to obtain training samples, the negative samples are divided into 3 points, and are respectively combined with positive samples into sample subsets; wherein the positive sample indicates that the customer category is a bad customer, and the negative sample indicates that the customer category is an ordinary customer. The three sample subsets respectively take feature information of the international trade dimension, enterprise information dimension and enterprise owner information dimension. During the training process, different feature information of the three dimensions of the training samples is used. The positive samples in the three dimensions are mutually optimized using the dimensional equivalence constraints designed in this scenario example. By minimizing the empirical loss of the model and the dimensional equivalence constraints, three sub-classifiers are obtained, and the classifier weights are calculated based on their accuracy for the positive samples in the training set. Finally, the three sub-classifiers are voted according to the weights to obtain the final classification model.

[0020] In this scenario example, the model training process can be divided into three parts: data preprocessing, feature engineering, and model building and training. Each part is described below.

[0021] In this scenario example, the data preprocessing part may include a data selection step and a data preprocessing step.

[0022] The data selection step may include: classifying the relevant characteristics related to the classification of international balance of payments customers into three categories: international trade information, enterprise information and enterprise owner information. The data scope can be determined by category, thereby determining the data table involved.

[0023] The data preprocessing step may include: observing the data columns related to international trade information, enterprise information and business owner information in the data table. Connecting the relevant data columns related to the classification of international balance of payments customers in different tables according to the customer ID to form the original features. For columns with incorrect data types, first convert them into the correct data type. For columns with missing values, fill them in a certain way, such as filling the missing values ​​of numerical features with the column mean, and filling the missing values ​​of non-numerical features with "unknown".

[0024] In this scenario example, the feature engineering part may include constructing an international trade dimension, constructing an enterprise information dimension, constructing an enterprise owner information dimension, and constructing training samples.

[0025] The step of constructing the international trade dimension includes: the basic features of the international trade dimension may include information such as the total transaction volume, foreign-related income, the amount of overseas remittances and its increment, the number of overseas remittances and its increment, etc. Derivative features are constructed based on this information, including time series aggregation features and time series historical features. The time series aggregation features represent the transaction data of the customer in the historical time period, and the time series historical features represent the transaction data of the customer at the historical time point.

[0026] Among them, the time series aggregation feature F agg The construction method is as follows:

[0027] F agg =[f(feature) time ,time=1,2,3,4,5,6,1-2,1-3,1-4,1-5,1-6]

[0028] Feature data f(feature) timeThe Mean() average value, Max() maximum value, Min() minimum value, and Std() standard deviation of the transaction data are taken respectively, and the time periods are the previous month, previous two months, previous three months, previous four months, previous five months, previous six months, previous second month, previous third month, previous fourth month, previous fifth month, and previous sixth month respectively.

[0029] Time series historical features F his The construction method is as follows,

[0030] F his =[feature time ,time=1,2,3,4,5,6]

[0031] The time periods are the first month before, the second month before, the third month before, the fourth month before, the fifth month before, and the sixth month before.

[0032] The construction of the enterprise information dimension includes: the enterprise information dimension includes basic information of the enterprise, daily average assets and its year-on-year growth, the amount of inflow and outflow of the enterprise account, the number of transactions and its year-on-year growth, etc. Based on this information, derived features are constructed, including time series aggregation features and time series historical features. Specifically, the construction method of time series aggregation features and time series historical features is similar to the construction method of time series aggregation features and time series historical features in the step of constructing the international trade dimension, and will not be repeated here.

[0033] The construction of the business owner information dimension includes: the business owner information dimension includes age, education level, whether it is an important customer, the amount of inflow and outflow of the business owner account, the number of transactions and its year-on-year growth, etc. Based on this information, derived features are constructed, including time series aggregation features and time series historical features. Specifically, the construction method of time series aggregation features and time series historical features is similar to the construction method of time series aggregation features and time series historical features in the step of constructing the international trade dimension, and will not be repeated here.

[0034] The constructed training samples include: the labels of the samples are 1 (positive sample) and 0 (negative sample), representing bad customers and ordinary customers respectively, and the characteristics include three dimensions, namely, international trade dimension, enterprise information dimension and business owner information dimension. The positive sample labeling rules are business quick loan overdue, personal credit default, personal loan credit card default, and corporate loan default, and only one of them is satisfied. Negative samples are customers without the above conditions.

[0035] In this scenario example, the model building and training part may include the following steps.

[0036] Step 1: Divide the negative samples into 3 parts, and combine each part with the positive samples to obtain 3 sample subsets, each sample subset has features of different dimensions. For example, the first sample subset takes the features of the international trade dimension and discards the features of other dimensions; the second sample subset takes the features of the enterprise information dimension and discards the features of other dimensions; the third sample subset takes the features of the business owner information dimension and discards the features of other dimensions. Preferably, the number of positive samples is 3 times the number of negative samples. The negative samples can be divided into 3 parts, and the negative samples of each part are combined with all the positive samples to obtain 3 sample subsets, so that the ratio of negative samples to positive samples in each sample subset is 1:1. The subset obtained in this way is just balanced between positive and negative classes, which reduces the difficulty of model learning and is conducive to improving the accuracy and recall rate of the model for positive samples.

[0037] Step 2: Design of dimensional equivalence constraints.

[0038]

[0039] Where R mv represents the dimension equivalence constraint; X + represents a positive sample, represents the i-th dimension of the positive sample, f i Denotes the sub-classifier under the i-th dimension. The dimensional equivalence constraint can be used to make the classification results of positive samples by sub-classifiers under different dimensions as consistent as possible, so that the classifier can make full use of the complementary information between different dimensions.

[0040] Step 3: Objective function design.

[0041] L=R emp +αR mv +λR reg

[0042] Where L represents the objective function; R emp Indicates experience loss; R mv Represents the dimension equivalence constraint; R reg represents L2 regularization loss; α and λ represent hyperparameters, which are used to adjust the weights of each item. Expand each item in the formula as follows:

[0043]

[0044] Among them, X i represents the i-th dimension of the training sample, y represents the label of the training sample, and W i Represents the classifier f i The corresponding coefficient.

[0045] Step 4: Model optimization. Specifically, the gradient descent method can be used to solve the optimization problem of the objective function and obtain sub-classifiers in different dimensions.

[0046] Step 5: Calculate the weight of the subclassifier. Specifically, the weight of the subclassifier can be calculated according to the degree of fit of the subclassifier to the positive samples in the training set. Suppose the number of positive samples in the training set is N. + , subclassifier f i The number of correct predictions is N i , subclassifier f i The weight calculation method of is as follows:

[0047]

[0048] Among them, w i Represents the subclassifier f i The weight of .

[0049] The final classification model F(X) is as follows:

[0050]

[0051] Step 6: Model testing. For the test sample x, divide the test sample x into three dimensions, input the discriminant function, and obtain the output result of the model. A result greater than or equal to 0.5 indicates that the customer is a bad customer, and a result less than 0.5 indicates that the customer is an ordinary customer.

[0052]

[0053] Specifically, the model testing process is as follows: Figure 2 As shown in the figure. First, obtain the feature information related to international balance of payments customers from the data warehouse. The features are divided into three dimensions according to categories, namely international trade dimension, enterprise information dimension and enterprise owner information dimension. Then perform data preprocessing and feature engineering on the test samples. Use the features of the three dimensions of the customer to be predicted to construct the test sample. Input the test sample into the classification model to obtain the prediction result.

[0054] It can be seen from the technical solution provided by the scenario example in the above specification that in this scenario example, the model divides the features into international trade dimension, enterprise information dimension and enterprise owner information dimension according to the category of the features, and divides the features of different categories into different dimensions. During the training process, the positive samples in the three dimensions are mutually optimized using the dimensional equivalence constraint. By minimizing the empirical loss of the model and the dimensional equivalence constraint, three sub-classifiers are obtained, and the classifier weights are calculated according to their accuracy for the positive samples in the training set. Finally, the three sub-classifiers are voted according to the weights to obtain the final classification model. Due to the extreme imbalance in the number of positive and negative samples in the training samples, the negative samples are divided into 3 points, and they are combined with the positive samples into sample subsets, which is conducive to alleviating the imbalance of the classification problem. During the training process of the model, the model's learning of positive samples can be mutually optimized, so that the model pays more attention to positive samples during the learning process, and improves the recall rate of the model's prediction of positive samples. The final model integrates the results of the three sub-models to improve the prediction accuracy of the model on the test samples.

[0055] See also Figure 3 . This specification embodiment provides a method for determining a customer category. In the embodiment of this specification, the subject that executes the customer category determination method may be an electronic device with a logical operation function, and the electronic device may be a server. The server may be an electronic device with a certain computing and processing capability. It may have a network communication unit, a processor, a memory, etc. Of course, the server is not limited to the above-mentioned electronic device with a certain entity, and it may also be software running in the above-mentioned electronic device. The server may also be a distributed server, which may be a system with multiple processors, memories, network communication modules, etc. operating in collaboration. Alternatively, the server may also be a server cluster formed by several servers. The method may include the following steps.

[0056] S310: Obtain a preset number of customer samples; the customer samples include positive samples and negative samples; the positive samples indicate that the customer category is a bad customer, and the negative samples indicate that the customer category is an ordinary customer.

[0057] In some embodiments, the customer may be an enterprise involved in international balance of payments business, for example, an international balance of payments customer.

[0058] In some embodiments, the server may obtain characteristic data of international balance of payments customers from a preset data warehouse. The characteristic data may be transaction data of customers, and labels may be established for customers based on the transaction data to obtain customer samples. Among them, bad customers may be used as positive samples, and ordinary customers may be used as negative samples; the labels of customer samples may be, for example, 1 (positive sample) and 0 (negative sample) to represent bad customers and ordinary customers, respectively.

[0059] In some embodiments, the bad customer meets at least one of the following conditions: business quick loan overdue, personal credit default, personal loan credit card default, corporate loan default; the ordinary customer is a customer without the above conditions. By distinguishing bad customers from ordinary customers and establishing positive and negative samples, the prerequisite for determining the customer category can be established, so that bad customers and ordinary customers can be accurately identified.

[0060] S320: Divide the negative samples into multiple groups of negative samples, and combine each group of negative samples with the positive samples to form a sample subset, thereby obtaining multiple sample subsets.

[0061] In some embodiments, negative samples can be divided into multiple groups, and each group of negative samples is combined with the positive samples to form a sample subset, thereby obtaining multiple sample subsets. For example, the customer sample includes negative samples {x1, x2, x3, x4, x5, x6} and positive samples {y1, y2, y3, y4}. The negative samples can be divided into three groups to obtain {x1, x2}, {x3, x4} and {x5, x6}, and then {x1, x2}, {x3, x4} and {x5, x6} are respectively combined with positive samples to form sample subsets {x1, x2, y1, y2, y3, y4}, {x3, x4, y1, y2, y3, y4} and {x5, x6, y1, y2, y3, y4}.

[0062] In some embodiments, the ratio of the number of positive samples to the number of negative samples is 3:1; accordingly, the negative samples are equally divided into multiple groups, each group of negative samples is combined with the positive samples to form a sample subset, and multiple sample subsets are obtained; the ratio of negative samples to positive samples in each sample subset is 1:1. Due to the particularity of the scene, there are often fewer positive samples (bad customers) and more negative samples (ordinary customers). The obtained data set is often unbalanced, so it is preferred that the negative samples are 3 times the positive samples, so that the ratio of positive and negative samples in the sample subset constructed in the next step is exactly 1:1. Divide the negative samples into 3 parts and combine them with the positive samples to obtain 3 sample subsets respectively. The subsets obtained in this way are exactly positive and negative balanced, which can reduce the learning difficulty of the subsequent fitting process and help improve the fitting accuracy.

[0063] S330: For different sample subsets, extract feature vectors of customer samples in each sample subset from different dimensions.

[0064] In some embodiments, the number of sample subsets is three; correspondingly, for different sample subsets, the feature vectors of customer samples in each sample subset are extracted from the international trade dimension, the enterprise information dimension, and the business owner information dimension. Among them, the international trade dimension includes information such as the total transaction volume, foreign-related income, the amount of overseas remittances and its increment, the number of overseas remittances and its increment, etc., the enterprise information dimension includes basic enterprise information, average daily assets and its year-on-year, the amount of inflow and outflow of enterprise accounts, the number of transactions and its year-on-year, etc., and the business owner information dimension includes age, education level, whether it is an important customer, the amount of inflow and outflow of the business owner account, the number of transactions and its year-on-year, etc.

[0065] Existing methods directly splice these different categories of information together into sample features, which, on the one hand, makes the sample feature dimension particularly large, easily causing the "dimensionality disaster problem". On the other hand, the feature scales and meanings of different categories of features often differ greatly. Directly splicing them together can easily cause them to lose their original physical meanings, resulting in the determination of customer categories failing to achieve the expected results. For different sample subsets, the feature vectors of customer samples in each sample subset are extracted from the international trade dimension, enterprise information dimension, and business owner information dimension, respectively, so that the features of different categories are divided into different dimensions, making full use of the complementary information of different dimensions, which can improve the accuracy of customer category determination.

[0066] In some embodiments, the feature vector includes time series aggregate features and time series historical features; the time series aggregate features represent the transaction data of the customer within the historical time period, and the time series historical features represent the transaction data of the customer at the historical time point. Specifically, because some features are related to time (such as 'month-end balance'), for a customer, there will be an 'month-end balance' every month. The purpose of constructing time series aggregate features is to obtain the state of the feature over a period of time in the past. The purpose of constructing time series historical features is to obtain the state of the feature at a certain point in time in the past. For example: for the feature 'month-end balance', construct the average of the feature in the past 1-3 months, and construct the value of the feature in the past 3 months. By constructing time series aggregate features and time series historical features, it is possible to analyze the relationship between customer categories as bad customers and ordinary customers and these features, thereby improving the accuracy of the subsequent fitting process.

[0067] In some embodiments, the time series aggregation feature extraction method includes:

[0068] F agg =[f(feature) time ,time=1,2,3,4,5,6,1-2,1-3,1-4,1-5,1-6]

[0069] Among them, F aggRepresents the time series aggregation feature, feature data f(feature) time Take the average, maximum, minimum and standard deviation of the transaction data respectively, and the time periods are the previous month, previous two months, previous three months, previous four months, previous five months, previous six months, previous second month, previous third month, previous fourth month, previous fifth month and previous sixth month respectively;

[0070] The time series historical feature extraction method includes:

[0071] F his =[feature time ,time=1,2,3,4,5,6]

[0072] Among them, F his Represents the historical characteristics of the time series, feature data f(feature) time The average value, maximum value, minimum value and standard deviation of the transaction data are taken respectively, and the time periods are the first month before, the second month before, the third month before, the fourth month before, the fifth month before, and the sixth month before.

[0073] Through the above-mentioned time series aggregation feature and time series historical feature extraction method, it is possible to extract features from the original data to the maximum extent for use in the fitting process. By taking the average, maximum, minimum, and standard deviation of the transaction data, feature extraction from multiple different dimensions is achieved, further improving the subsequent fitting accuracy.

[0074] S340: Based on a preset objective function, the feature vectors of the customer samples in each sample subset and the customer categories represented by the customer samples are fitted to obtain classification conditions of the customer categories under each dimension.

[0075] In some embodiments, the objective function includes at least a dimension equivalence constraint term:

[0076]

[0077] Where R mv (X + ) represents the dimension equivalence constraint; X + represents a positive sample, represents the i-th dimension of the positive sample, Represents the classification condition under the i-th dimension. Through the dimensional equivalence constraint, the classification results of the positive samples under the classification conditions under different dimensions can be kept as consistent as possible, so that the classification conditions can make full use of the complementary information between different dimensions, thereby improving the accuracy of classification. Among them, the classification condition corresponds to the sub-classifier in the above scenario example.

[0078] In some embodiments, the objective function includes:

[0079] L=R emp +αR mv +λR reg

[0080] Among them, L represents the objective function, R emp represents the experience loss, R mv represents the dimension equivalence constraint, R reg represents the L2 regularization loss, α and λ represent hyperparameters used to adjust the weights of each item. Expand the formula as follows:

[0081]

[0082] Among them, X i represents the i-th dimension of the training sample, y represents the label of the training sample, and W i Indicates the classification condition f i Corresponding coefficients. Through L2 regularization loss, empirical loss and dimensional equivalence constraints, the complementary information between different dimensions is fully explored, and the model pays more attention to positive samples, making the classification conditions more accurate for customers.

[0083] In some embodiments, the gradient descent method can be used to solve the optimization problem of the objective function to obtain classification conditions under different dimensions.

[0084] S350: Determine the category of the target customer according to the classification conditions of the customer category under each dimension.

[0085] In some embodiments, in order to further improve the accuracy of customer category determination, the weights of the classification conditions under each dimension may be calculated to finally obtain a comprehensive classification condition, and the comprehensive classification condition is used to determine the category of the target customer. Specifically, determining the category of the target customer according to the classification conditions of the customer category under each dimension may include the following steps.

[0086] S351: Calculate the weight of the classification condition of the customer category under each dimension according to the fitting degree of the positive samples under each dimension.

[0087] Specifically, let the number of positive samples in the sample subset be N + , classification condition f i The number of correct predictions is N i , the classification condition f under the i-th dimension i The weight calculation method of is as follows:

[0088]

[0089] Among them, w i Represents the classification condition f under the i-th dimension i The weight of .

[0090] S352: Generate comprehensive classification conditions for customer categories according to the weights of the classification conditions for customer categories in each dimension.

[0091] Specifically, the comprehensive classification condition F(X) is as follows:

[0092]

[0093] S353: Extract feature vectors of target customers in various dimensions.

[0094] Specifically, the feature vectors of the target customers can be extracted from the international trade dimension, the enterprise information dimension, and the enterprise owner information dimension, respectively. The specific extraction method can refer to S330 and will not be described in detail here.

[0095] S354: Substitute the feature vectors of each dimension of the target customer into the comprehensive classification condition to obtain the category of the target customer.

[0096] Specifically, the category of the target customer can be obtained according to the following method:

[0097]

[0098] Among them, x represents the feature vector of the target customer, x i Represents the feature vector under the i-th dimension. A result greater than or equal to 0.5 represents that the customer is a bad customer, and a result less than 0.5 represents that the customer is an ordinary customer. Of course, the correspondence between the output result value and the customer category can also be set according to the actual situation, and the embodiments of this specification do not limit this. For example, a result greater than or equal to 0.3 represents that the customer is a bad customer, and a result less than 0.3 represents that the customer is an ordinary customer; for example, a result greater than or equal to 0.6 represents that the customer is a bad customer, and a result less than 0.6 represents that the customer is an ordinary customer.

[0099] By calculating the weight of the classification conditions of customer categories under each dimension according to the fitting degree of the positive samples under each dimension, the influence of each dimension on the determination of customer categories can be accurately determined, thereby achieving the complementarity of classification conditions under different dimensions and further improving the accuracy of customer category determination.

[0100] It can be seen from the technical solutions provided in the above embodiments of this specification that, in the embodiments of this specification, a preset number of customer samples can be obtained; the customer samples include positive samples and negative samples; the positive samples indicate that the customer category is a bad customer, and the negative samples indicate that the customer category is an ordinary customer; the negative samples are divided into multiple groups of negative samples, each group of negative samples is combined with the positive samples to form a sample subset, and multiple sample subsets are obtained; for different sample subsets, the feature vectors of the customer samples in each sample subset are extracted from different dimensions; based on the preset objective function, the feature vectors of the customer samples in each sample subset and the customer categories represented by the customer samples are fitted to obtain the classification conditions of the customer categories under each dimension; the category of the target customer is determined according to the classification conditions of the customer categories under each dimension. The method provided in the embodiments of this specification divides the negative samples and combines them with the positive samples into sample subsets, which is conducive to alleviating the imbalance of the classification problem, and divides the features of different categories into different dimensions, making full use of the complementary information of different dimensions, thereby improving the accuracy of customer category determination.

[0101] Figure 4 This is a functional structure diagram of an electronic device according to an embodiment of the present specification. The electronic device may include a memory and a processor.

[0102] In some embodiments, the memory can be used to store the computer program and / or module, and the processor implements various functions of the customer category determination method by running or executing the computer program and / or module stored in the memory, and calling the data stored in the memory. The memory may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system and at least one application required for a function; the data storage area may store data created according to the use of the user terminal. In addition, the memory may include a high-speed random access memory, and may also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0103] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc. The processor can execute the computer instructions to implement the following steps: obtain a preset number of customer samples; the customer samples include positive samples and negative samples; the positive samples indicate that the customer category is a bad customer, and the negative samples indicate that the customer category is an ordinary customer; the negative samples are divided into multiple groups of negative samples, each group of negative samples is combined with the positive samples to form a sample subset, and multiple sample subsets are obtained; for different sample subsets, the feature vectors of the customer samples in each sample subset are extracted from different dimensions; based on a preset objective function, the feature vectors of the customer samples in each sample subset and the customer categories represented by the customer samples are fitted to obtain the classification conditions of the customer categories under each dimension; the category of the target customer is determined according to the classification conditions of the customer categories under each dimension.

[0104] In the embodiments of this specification, the functions and effects specifically implemented by the electronic device can be explained in comparison with other embodiments and will not be repeated here.

[0105] Figure 5 This is a functional structural diagram of a device for determining a customer category according to an embodiment of the present specification. The device may specifically include the following structural modules.

[0106] The acquisition module 510 is used to acquire a preset number of customer samples; the customer samples include positive samples and negative samples; the positive samples indicate that the customer category is a bad customer, and the negative samples indicate that the customer category is a normal customer;

[0107] A division module 520 is used to divide the negative samples into multiple groups of negative samples, each group of negative samples is combined with the positive samples to form a sample subset, and obtain multiple sample subsets;

[0108] An extraction module 530 is used to extract feature vectors of customer samples in each sample subset from different dimensions for different sample subsets;

[0109] A fitting module 540 is used to fit the feature vectors of the customer samples in each sample subset and the customer categories represented by the customer samples based on a preset objective function to obtain classification conditions of the customer categories under each dimension;

[0110] The determination module 550 is used to determine the category of the target customer according to the classification conditions of the customer category in each dimension.

[0111] The embodiments of the present specification also provide a computer-readable storage medium for a method for determining a customer category, wherein the computer-readable storage medium stores computer program instructions, which, when executed, implement the following: obtaining a preset number of customer samples; the customer samples include positive samples and negative samples; the positive samples indicate that the customer category is a bad customer, and the negative samples indicate that the customer category is an ordinary customer; dividing the negative samples into multiple groups of negative samples, each group of negative samples is combined with the positive samples to form a sample subset, and obtaining multiple sample subsets; for different sample subsets, extracting feature vectors of the customer samples in each sample subset from different dimensions; based on a preset objective function, fitting the feature vectors of the customer samples in each sample subset and the customer category represented by the customer samples to obtain classification conditions for the customer category under each dimension; and determining the category of the target customer according to the classification conditions for the customer category under each dimension.

[0112] In the embodiments of this specification, the above-mentioned storage medium includes but is not limited to random access memory (Random Access Memory, RAM), read-only memory (Read-Only Memory, ROM), cache (Cache), hard disk (Hard Disk Drive, HDD) or memory card (Memory Card). The memory can be used to store the computer program and / or module, and the memory can mainly include a storage program area and a storage data area, wherein the storage program area can store an operating system, at least one application required for a function, etc.; the storage data area can store data created according to the use of the user terminal, etc. In addition, the memory may include a high-speed random access memory and may also include a non-volatile memory. In the embodiments of this specification, the functions and effects specifically implemented by the program instructions stored in the computer-readable storage medium can be explained in comparison with other implementations, and will not be repeated here.

[0113] It should be noted that the customer category determination method, device and storage medium provided in the embodiments of this specification can be applied to the field of artificial intelligence technology. Of course, it can also be applied to the financial field, or any field other than the financial field. The embodiments of this specification do not limit the application field of the customer category determination method, device and storage medium.

[0114] It should be noted that the various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device embodiment and the equipment embodiment, since they are basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0115] After reading this specification, those skilled in the art can conceive of any combination of some or all of the embodiments listed in this specification without creative work, and these combinations are also within the scope of disclosure and protection of this specification.

[0116] In the 1990s, improvements to a technology could be clearly distinguished as hardware improvements (for example, improvements to the circuit structure of diodes, transistors, switches, etc.) or software improvements (improvements to the method flow). However, with the development of technology, many improvements to the method flow today can be regarded as direct improvements to the hardware circuit structure. Designers almost always obtain the corresponding hardware circuit structure by programming the improved method flow into the hardware circuit. Therefore, it cannot be said that an improvement in a method flow cannot be implemented using a hardware entity module. For example, a programmable logic device (PLD) (such as a field programmable gate array (FPGA)) is such an integrated circuit whose logical function is determined by the user's programming of the device. Designers can "integrate" a digital system on a PLD by programming it themselves, without having to ask a chip manufacturer to design and produce a dedicated integrated circuit chip. Moreover, nowadays, instead of manually making integrated circuit chips, this kind of programming is mostly implemented by "logic compiler" software, which is similar to the software compiler used when developing and writing programs, and the original code before compilation must also be written in a specific programming language, which is called hardware description language (HDL). There is not only one HDL, but many kinds, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc. The most commonly used ones are VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog2. Those skilled in the art should also know that it is only necessary to program the method flow slightly in the above-mentioned hardware description languages ​​and program it into the integrated circuit, and then it is easy to obtain the hardware circuit that implements the logic method flow.

[0117] The systems, devices, modules or units described in the above embodiments may be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0118] It can be known from the above description of the implementation mode that the technicians in this field can clearly understand that the present specification can be implemented by means of software plus the necessary general hardware platform. Based on such an understanding, the technical solution of the present specification can be essentially or the part that contributes to the prior art can be embodied in the form of a software product, which can be stored in a storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present specification or some parts of the embodiments.

[0119] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0120] This specification can be used in many general or special computer system environments or configurations, such as personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronic devices, network PCs, minicomputers, mainframe computers, and distributed computing environments that include any of the above systems or devices.

[0121] This specification may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. This specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules may be located in local and remote computer storage media, including storage devices.

[0122] Although the present specification is described through embodiments, those skilled in the art will appreciate that there are many modifications and changes to the present specification without departing from the spirit of the present specification, and it is intended that the appended claims include these modifications and changes without departing from the spirit of the present specification.

Claims

1. A method for determining a customer category, characterized in that: The method comprises: Obtain a preset number of customer samples; the customer samples include positive samples and negative samples; the positive samples indicate that the customer category is a bad customer, and the negative samples indicate that the customer category is an ordinary customer; Dividing the negative samples into multiple groups of negative samples, each group of negative samples is combined with the positive samples to form a sample subset, to obtain multiple sample subsets; For different sample subsets, feature vectors of customer samples in each sample subset are extracted from different dimensions; wherein the feature vectors include time series aggregation features and time series historical features; the time series aggregation features represent the transaction data of customers in a historical time period, and the time series historical features represent the transaction data of customers at historical time points; Based on a preset objective function, the characteristic vectors of the customer samples in each sample subset and the customer categories represented by the customer samples are fitted to obtain classification conditions of the customer categories under each dimension; wherein the preset objective function at least includes a dimension equivalence constraint item, and the dimension equivalence constraint item is used to make the classification conditions of the customer categories under different dimensions consistent with the classification results of the positive samples; Determine the category of target customers based on the classification criteria of customer categories under each dimension.

2. The method according to claim 1, characterized in that The bad customers meet at least one of the following conditions: business quick loan overdue, personal credit default, personal loan credit card default, corporate loan default; The ordinary customers are customers who do not have the above-mentioned situations.

3. The method according to claim 1, characterized in that: The ratio of the number of positive samples to the number of negative samples is 3:1; Accordingly, the negative samples are equally divided into multiple groups, and each group of negative samples is combined with the positive samples to form a sample subset, thereby obtaining multiple sample subsets; in each sample subset, the ratio of negative samples to positive samples is 1:

1.

4. The method according to claim 1, characterized in that: The number of sample subsets is three; Correspondingly, for different sample subsets, the feature vectors of customer samples in each sample subset are extracted from the international trade dimension, enterprise information dimension and business owner information dimension respectively.

5. The method according to claim 1, characterized in that The time series aggregation feature extraction method includes: F agg =[f(feature) time ,time=1,2,3,4,5,6,1-2,1-3,1-4,1-5,1-6] Among them, F agg Represents the time series aggregation feature, feature data f(feature) time Take the average, maximum, minimum and standard deviation of the transaction data respectively, and the time periods are the previous month, previous two months, previous three months, previous four months, previous five months, previous six months, previous second month, previous third month, previous fourth month, previous fifth month and previous sixth month respectively; The time series historical feature extraction method includes: F his =[feature time ,time=1,2,3,4,5,6] Among them, F his Represents the historical characteristics of the time series, feature data f(feature) time The average value, maximum value, minimum value and standard deviation of the transaction data are taken respectively, and the time periods are the first month before, the second month before, the third month before, the fourth month before, the fifth month before, and the sixth month before.

6. The method according to claim 1, characterized in that The objective function includes at least a dimension equivalence constraint: Where R mv (X + ) represents the dimension equivalence constraint; X + represents a positive sample, represents the i-th dimension of the positive sample, Indicates the classification condition under the i-th dimension.

7. The method according to claim 1, characterized in that The objective function includes: L=R emp +αR mv +λR reg Among them, L represents the objective function, R emp represents the experience loss, R mv represents the dimension equivalence constraint, R reg represents the L2 regularization loss, and α and λ are hyperparameters.

8. The method according to claim 1, characterized in that Determining the categories of target customers according to the classification conditions of customer categories under various dimensions includes: Calculate the weight of the classification conditions of customer categories under each dimension according to the fitting degree of the positive samples under each dimension; Generate comprehensive classification conditions for customer categories according to the weights of the classification conditions for customer categories under various dimensions; Extract feature vectors of target customers in various dimensions; Substitute the characteristic vectors of the target customers in various dimensions into the comprehensive classification conditions to obtain the categories of the target customers.

9. A device for determining a customer category, characterized in that: The device comprises: An acquisition module is used to acquire a preset number of customer samples; the customer samples include positive samples and negative samples; the positive samples indicate that the customer category is a bad customer, and the negative samples indicate that the customer category is an ordinary customer; A division module, used for dividing the negative samples into multiple groups of negative samples, each group of negative samples is combined with the positive samples to form a sample subset, to obtain multiple sample subsets; An extraction module is used to extract feature vectors of customer samples in each sample subset from different dimensions for different sample subsets; wherein the feature vectors include time series aggregation features and time series historical features; the time series aggregation features represent the transaction data of customers in a historical time period, and the time series historical features represent the transaction data of customers at historical time points; A fitting module is used to fit the feature vectors of customer samples in each sample subset and the customer categories represented by the customer samples based on a preset objective function to obtain classification conditions of customer categories under each dimension; wherein the preset objective function at least includes a dimension equivalence constraint item, and the dimension equivalence constraint item is used to make the classification conditions of customer categories under different dimensions consistent with the classification results of positive samples; The determination module is used to determine the category of target customers according to the classification conditions of customer categories under various dimensions.

10. An electronic device, characterized in that: include: Memory for storing computer programs; A processor, configured to execute the computer program to achieve: obtaining a preset number of customer samples; the customer samples include positive samples and negative samples; the positive samples indicate that the customer category is a bad customer, and the negative samples indicate that the customer category is an ordinary customer; dividing the negative samples into multiple groups of negative samples, each group of negative samples is combined with the positive samples to form a sample subset, and obtaining multiple sample subsets; for different sample subsets, extracting feature vectors of the customer samples in each sample subset from different dimensions; wherein the feature vectors include time series aggregation features and time series historical features; the time series aggregation features indicate the transaction data of the customer in a historical time period, and the time series historical features indicate the transaction data of the customer at a historical time point; based on a preset objective function, fitting the feature vectors of the customer samples in each sample subset and the customer categories represented by the customer samples to obtain classification conditions of the customer categories in each dimension; wherein the preset objective function includes at least a dimension equivalence constraint item, and the dimension equivalence constraint item is used to make the classification conditions of the customer categories in different dimensions consistent with the classification results of the positive samples; determining the category of the target customer according to the classification conditions of the customer categories in each dimension.

11. A computer-readable storage medium, characterized in that: Computer instructions are stored thereon, and when the instructions are executed, the following are achieved: obtaining a preset number of customer samples; the customer samples include positive samples and negative samples; the positive samples indicate that the customer category is a bad customer, and the negative samples indicate that the customer category is an ordinary customer; dividing the negative samples into multiple groups of negative samples, each group of negative samples is combined with the positive samples to form a sample subset, and multiple sample subsets are obtained; for different sample subsets, extracting feature vectors of customer samples in each sample subset from different dimensions; wherein the feature vectors include time series aggregation features and time series historical features; the time series aggregation features indicate the transaction data of customers in a historical time period, and the time series historical features indicate the transaction data of customers at historical time points; based on a preset objective function, fitting the feature vectors of customer samples in each sample subset and the customer categories represented by the customer samples to obtain classification conditions of customer categories under each dimension; wherein the preset objective function includes at least a dimension equivalence constraint item, and the dimension equivalence constraint item is used to make the classification conditions of customer categories under different dimensions consistent with the classification results of positive samples; determining the category of the target customer according to the classification conditions of customer categories under each dimension.

Images