Customer category determination method and device, electronic equipment and storage medium

Abstract

The invention discloses a customer category determination method and device, electronic equipment and a storage medium, and relates to the technical field of data processing. The customer category determination method comprises the steps of obtaining original customer feature data of a to-be-identified customer, and performing data preprocessing on the original customer feature data to obtain target customer feature data; determining a similarity score corresponding to the target customer feature data based on a preset similarity model, and generating a classification score corresponding to the target customer feature data based on a preset classification model; determining a target score according to a pre-trained logistic regression element learner, the similarity score and the classification score, and determining a customer category of the to-be-identified customer according to the target score; wherein the preset similarity model, the preset classification model and the logistic regression meta learner adopt stacked ensemble learning, the preset similarity model and the preset classification model are base learning layers, and the logistic regression meta learner is a meta learning layer, so that the accuracy of determining the customer category is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, and in particular to a customer category determination method and device, electronic equipment and a storage medium. BACKGROUND

[0002] With the continuous development of the financial industry, credit cards, as an indispensable payment tool in daily life, are increasingly attracting the attention of customers, especially young customers. In order to attract more customers, major banks have launched various marketing activities. These activities have attracted more customers, but also increased marketing and operating costs. These increased costs need to be reduced by attracting more customers, but attracting more customers requires increasing marketing and operating costs. This dilemma is particularly prominent in the context of increasing market competition and slowing growth. Therefore, it is particularly important for credit card business operations to acquire incremental customers at low cost.

[0003] Marketing and operating costs can be divided into two categories according to whether the customer has a card after marketing. One category is customers who have a card after marketing, which is called effective cost. The other category is customers who still have no card after marketing, which is called ineffective cost. Therefore, how to define the scope of potential customers before marketing, improve the card ratio of customers after marketing, and reduce ineffective costs is crucial. In the prior art, a recommendation algorithm based on customer similarity is generally used to find customers similar to existing card customers for marketing. However, only the correlation between customers is considered, and the willingness result is not comprehensively considered. How to comprehensively consider the category of customers and improve the accuracy of determining the category of customers has become a problem to be solved. SUMMARY

[0004] The present application provides a customer category determination method and device, electronic equipment and a storage medium to solve the problem of low accuracy of determining the category of customers by the recommendation algorithm based on customer similarity in the prior art.

[0005] According to an aspect of the present application, a customer category determination method is provided, wherein the method comprises:

[0006] Obtaining original customer feature data of a to-be-identified customer, and obtaining target customer feature data by data preprocessing the original customer feature data;

[0007] Determining a similarity score corresponding to the target customer feature data based on a preset similarity model, and generating a classification score corresponding to the target customer feature data based on a preset classification model;

[0008] determine a target score according to a pre-trained logistic regression meta-learner, the similarity score and the classification score, and determine the customer category of the to-be-identified customer according to the target score; wherein the preset similarity model, the preset classification model and the logistic regression meta-learner adopt stacked ensemble learning, the preset similarity model and the preset classification model are base learning layers, and the logistic regression meta-learner is a meta-learning layer.

[0009] According to another aspect of the present application, a customer category determination apparatus is provided, wherein the apparatus comprises:

[0010] a data acquisition module configured to acquire original customer feature data of a to-be-identified customer, and perform data preprocessing on the original customer feature data to obtain target customer feature data;

[0011] a score determination module configured to determine a similarity score corresponding to the target customer feature data based on a preset similarity model, and generate a classification score corresponding to the target customer feature data based on a preset classification model;

[0012] a category determination module configured to determine a target score according to a pre-trained logistic regression meta-learner, the similarity score and the classification score, and determine the customer category of the to-be-identified customer according to the target score; wherein the preset similarity model, the preset classification model and the logistic regression meta-learner adopt stacked ensemble learning, the preset similarity model and the preset classification model are base learning layers, and the logistic regression meta-learner is a meta-learning layer.

[0013] According to another aspect of the present application, an electronic device is provided, comprising:

[0014] at least one processor; and

[0015] a memory communicatively connected to the at least one processor; wherein

[0016] the memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor to enable the at least one processor to perform a customer category determination method according to any one of the embodiments of the present application.

[0017] According to another aspect of the present application, a computer readable storage medium is provided, which stores computer instructions for enabling a processor to perform a customer category determination method according to any one of the embodiments of the present application when executed by the processor.

[0018] The technical scheme of the embodiment of the application comprises the following steps: obtaining original customer feature data of a to-be-identified customer, performing data preprocessing on the original customer feature data to obtain target customer feature data, determining a similarity score corresponding to the target customer feature data based on a preset similarity model, generating a classification score corresponding to the target customer feature data based on a preset classification model, determining a target score according to the similarity score and the classification score, and determining the customer category of the to-be-identified customer according to the target score, so as to realize the identification of the customer category by combining the correlation between customers and the customer feature information of the customer itself, and improve the accuracy of determining the customer category of the to-be-identified customer. Meanwhile, the advantages and application ranges of the two technologies of the preset similarity model and the preset classification model are combined in a stacked ensemble learning manner, so as to improve the generalization ability of the model.

[0019] It should be understood that the content described in this part is not intended to identify key or important features of the embodiments of the application, nor is it used to limit the scope of the application. Other features of the application will become apparent from the following description. BRIEF DESCRIPTION OF DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the application, the drawings needed in the embodiment description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor.

[0021] Figure 1 is a flowchart of a customer category determination method according to an embodiment of the application;

[0022] Figure 2 is a flowchart of a customer category determination method according to an embodiment of the application;

[0023] Figure 3 is a structural schematic diagram of a customer category determination system according to an embodiment of the application;

[0024] Figure 4 is a flowchart of a data preprocessing module according to an embodiment of the application;

[0025] Figure 5 is a flowchart of a missing value processing according to an embodiment of the application;

[0026] Figure 6 is a flowchart of a repeated sample according to an embodiment of the application;

[0027] Figure 7 is a structural schematic diagram of a customer category determination device according to an embodiment of the application;

[0028] Figure 8 is a structural schematic diagram of an electronic device for implementing a customer category determination method according to an embodiment of the present application. DETAILED DESCRIPTION

[0029] In order to enable persons skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by persons skilled in the art without creative efforts should fall within the protection scope of the present application.

[0030] It should be noted that the terms "first", "second", and the like in the specification and claims of the present application and the above-described drawings are used to distinguish similar objects, and do not necessarily indicate a specific order or a chronological sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product, or device that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but can include other steps or units that are not clearly listed or inherent to the process, method, product, or device.

[0031] In the technical solutions of the present application, the acquisition, storage, use, processing, etc. of data comply with the relevant provisions of national laws and regulations.

[0032] It should be noted that in the embodiments of the present application, some industry existing solutions, components, models, etc. may be mentioned, which should be considered as exemplary, and the purpose is only to illustrate the feasibility of the implementation of the technical solutions of the present application, but it does not mean that the applicant has or will necessarily use the solutions.

[0033] Embodiment one

[0034] Figure 1 is a flowchart of a customer category determination method according to Embodiment One of the present application. The present embodiment can be applied to the case of determining the customer category of a customer to be identified. The method can be performed by a customer category determination device, which can be implemented in the form of hardware and / or software, and can be configured in an electronic device. As shown in the figure, the method comprises: Figure 1

[0035] S110, obtaining original customer feature data of a customer to be identified, and performing data preprocessing on the original customer feature data to obtain target customer feature data.​

[0036] The to-be-identified customer can be understood as a customer who needs to be identified in a customer category. Generally, the to-be-identified customer can be a new customer of a financial institution (such as a bank) or an old customer who has not handled a financial-related business. The original customer feature data can be understood as information for representing attributes, behaviors, preferences, and the like of the to-be-identified customer without being processed, directly collected or recorded. Exemplarily, the original customer feature data can include, but is not limited to, basic information and financial information, such as a customer unique identifier, gender, age, region, occupation type, and reception situation of the to-be-identified customer. In actual operation, the original customer feature data can be bank account information and the like of legal information, and is subjected to necessary desensitization processing. The data preprocessing refers to a process of text cleaning, standardization, coding, integration and the like of the original customer feature data before data analysis, and aims to solve problems such as missing and format disorder in the original data.

[0037] In an embodiment, the basic information and the financial information of the to-be-identified customer can be collected as the original customer feature data. Generally, the original customer feature data can be provided by the customer, or can be collected by a legal channel of the financial institution. The original customer feature data is subjected to data cleaning, standardization processing, coding and the like to obtain target customer feature data. In actual operation, the original customer feature data can include numerical features and text features, and the numerical features and the text features can be subjected to data preprocessing respectively to obtain an input format of a preset similarity model and a preset classification model. The original customer feature data subjected to data preprocessing is taken as the target customer feature data.

[0038] S120, determining a similarity score corresponding to the target customer feature data based on the preset similarity model, and generating a classification score corresponding to the target customer feature data based on the preset classification model.

[0039] The preset similarity model refers to a binary classification model used to determine the similarity between the to-be-identified customer and the target customer according to the target customer feature data. In actual application, historical customer feature data can be obtained as a training set to construct the preset similarity model. The preset classification model can be constructed based on an artificial neural network (ANN). The preset classification model is a model that abstracts neurons in the human brain from the perspective of information processing, establishes a model according to different connection modes to form different networks, and simulates the brain neural network to process information. The similarity score is used to represent the similarity between the to-be-identified customer and the historical business customer. Generally, the similarity score can be understood as the probability that the to-be-identified customer is the target customer. The classification score can be used to represent the probability that the to-be-identified customer is the target customer, that is, the probability that the to-be-identified customer can handle the financial related business, that is, the intention score of the to-be-identified customer to handle the financial related business. The historical business customer can be understood as a customer who has handled the financial institution related business, for example, a customer who has handled a credit card.

[0040] In the embodiment, the target customer feature data can be input into the preset similarity model, and the similarity between the to-be-identified customer and the target customer is determined according to the target customer feature data by the preset similarity model to generate a corresponding similarity score. At the same time, the target customer feature data is input into the preset classification model, and the score of the target customer feature data belonging to the target customer by the preset classification model is taken as the classification score.

[0041] S130, determining a target score according to the pre-trained logistic regression meta-learner, the similarity score and the classification score, and determining the customer category of the to-be-identified customer according to the target score.

[0042] The preset similarity model, the preset classification model and the logistic regression meta-learner adopt stacked ensemble learning, the preset similarity model and the preset classification model are base learning layers, and the logistic regression meta-learner is a meta-learning layer. Stacked ensemble learning is an ensemble learning technique of machine learning, which trains a meta-model by taking the prediction results of multiple base models as input to obtain the final prediction. The base learning layer is used for preliminary learning of the original data, and outputs the prediction result. The meta-learning layer is used to learn the mapping relationship between the meta-features and the real labels by taking the prediction results of the base learning layer as the meta-features, and outputs the final prediction. The logistic regression meta-learner is a commonly used second layer model selection, which learns the prediction results of the first layer base model to achieve better ensemble effect.

[0043] The target score can be understood as a sum score combined according to the similarity score and the classification score, and is used to determine the customer category of the to-be-identified customer. The customer category is used to indicate whether the to-be-identified customer is a potential customer of the financial institution related business, and the customer category can include target customers and non-target customers, i.e., potential customers and non-potential customers.

[0044] In an embodiment, the similarity score and the classification score can be input into a pre-trained logistic regression meta-learner, and the logistic regression meta-learner is used to generate the target score by weighting the similarity score and the classification score. In this embodiment, the logistic regression meta-learner adopts logistic regression, and the loss function adopts cross-entropy loss function. When determining the customer category of the to-be-identified customer according to the size of the target score, for example, a target threshold can be set in advance. When the target score is greater than the target threshold, the to-be-identified customer is considered as a target customer; and when the target score is less than or equal to the target threshold, the to-be-identified customer is considered as a non-target customer.

[0045] In the embodiment of the application, the original customer feature data of the to-be-identified customer is obtained, the original customer feature data is preprocessed to obtain target customer feature data, the similarity score corresponding to the target customer feature data is determined based on a pre-set similarity model, the classification score corresponding to the target customer feature data is generated based on a pre-set classification model, the target score is determined according to the similarity score and the classification score, and the customer category of the to-be-identified customer is determined according to the target score. The customer category is identified by combining the correlation between customers and the customer's own feature information, and the accuracy of determining the customer category of the to-be-identified customer is improved. At the same time, the advantages and application ranges of the two technologies of the pre-set similarity model and the pre-set classification model are fused in a stacked ensemble learning manner, and the generalization ability of the model is improved.

[0046] In an embodiment, the training process of the pre-set classification model includes:

[0047] The customer feature data and the customer business category of each customer at a historical moment are collected respectively, the numerical features in the customer feature data are determined as first numerical features, and the text features in the customer feature data are determined as first text features;

[0048] The first numerical features and the second text features are subjected to data cleaning to obtain second numerical features and second text features;

[0049] The second numerical features and the second text features are subjected to category balancing according to the customer business category, the numerical features after the category balancing are subjected to standardization processing to obtain target numerical features, and the text features after the category balancing are subjected to encoding to obtain target text features;

[0050] The target numerical feature and the corresponding target text feature are taken as first customer feature data, and the first customer feature data is divided into a training set and a test set in a random sampling manner.

[0051] The training set is input into a preset classification model to perform training, and a loss function of the preset classification model is a cross-entropy loss function. The preset classification model is based on an artificial neural network.

[0052] The customer business category is used to indicate whether the customer has handled a financial related business. For example, the customer business category can include a historical business customer and a historical non-business customer. The historical business customer can be understood as a customer who has handled a financial institution related business, for example, a customer who has handled a credit card. The historical non-business customer can be understood as a customer who has not handled a financial institution related business, for example, a customer who has not handled a credit card.

[0053] In an embodiment, the customer feature data and the customer business category of each customer in a historical moment can be collected respectively, that is, the customer feature data of the historical business customer and the customer feature data of the historical non-business customer in the historical moment are collected respectively. The historical moment can be the past 6 months, the past 1 year, etc. The numerical feature in the customer feature data is determined as a first numerical feature, and the text feature in the customer feature data is determined as a first text feature. Data cleaning is performed on the first numerical feature and the first text feature to obtain a second numerical feature and a second text feature. Specifically, for the average index value in the past period of time, the original data needs to be collected and the corresponding index needs to be calculated, such as the average account balance in the past 6 months, the account balance in the past 6 months is collected and the monthly average balance is calculated; for other data, samples with missing values are deleted. The calculated average index value feature needs to be marked as an accurate value to prevent subsequent operations from being processed. After the missing value processing, the data is processed for repeated samples. Since the customer unique identifier can represent the information of the customer, the repeated samples are deleted according to the identifier. First, samples belonging to the same user are grouped according to the identifier, and then each column feature of each group of samples is processed. Only one sample will be retained for each group.

[0054] According to the customer business category, the second numerical feature and the second text feature are balanced in category, a majority of samples are deleted by under-sampling, and the number of samples of the two categories is basically consistent. If the number of samples is too small after deletion, the deletion is cancelled, and a high weight is set for a minority of samples. The numerical feature after category balancing is standardized to obtain a target numerical feature, and the text feature after category balancing is encoded to obtain a target text feature. The target numerical feature and the corresponding target text feature are taken as first customer feature data, and the first customer feature data is divided into a training set and a test set in a random sampling manner. For example, 80% of positive samples and 80% of negative samples are randomly selected as the training set, and the remaining 20% of positive samples and 20% of negative samples are taken as the test set. The training set is input into a pre-constructed preset classification model for training, and the loss function of the preset classification model is a cross-entropy loss function Wherein, y is the true value of a single sample, y The predicted value of a single sample is shown. The number of layers, the number of nodes in each layer, the learning rate, the regularization coefficient and other hyperparameters of the preset classification model are set, the model weight and other parameters are initialized, the adaptive learning rate optimization algorithm (Adam) is selected, the initial learner is obtained, and then the hyperparameters are optimized by the Bayesian optimization algorithm to obtain the final preset classification model.

[0055] In an embodiment, after determining the target score according to the similarity score and the classification score, and determining the customer category of the to-be-identified customer according to the target score, the method further comprises:

[0056] The number of business customers in the customer category at the current time is determined as a first number, the number of business customers at a first historical time is determined as a second number, and the number of business customers at a second historical time is determined as a third number; wherein the second historical time is located in a previous acquisition period of the first historical time;

[0057] The difference between the first number and the second number is determined as a first difference, and the difference between the second number and the third number is determined as a second difference;

[0058] The ratio of the first difference to the first number is determined as a first ratio, and the ratio of the second difference to the second number is determined as a second ratio;

[0059] The target threshold is updated according to the first ratio and the second ratio.

[0060] Wherein, the number of business customers can be acquired according to the acquisition period, and the second historical time is located in a previous acquisition period of the first historical time. For example, when the acquisition period is 1 month, the first historical time is 1 month ago, and the second historical time is 2 months ago. The business customer is a customer who has handled a financial related business.

[0061] In this embodiment, the number of business customers in the customer category at the current moment can be determined as a first quantity, the number of business customers at a first historical moment can be determined as a second quantity, the number of business customers at a second historical moment can be determined as a third quantity, the difference between the first quantity and the second quantity can be determined as a first difference, the difference between the second quantity and the third quantity can be determined as a second difference, and the ratio of the first difference to the first quantity can be determined as a first ratio, which is the proportion of business customers who have identified the target customer to conduct financial-related business; the ratio of the second difference to the second quantity can be determined as a second ratio, which is the proportion of business customers who have identified the target customer to conduct financial-related business at a historical moment, and the target threshold can be updated according to the first ratio and the second ratio. Specifically, when the first ratio is α... t The second ratio is α1, α2, ..., α t-1 In this case, let the weighted moving average of the card issuance ratio be Δ, and the recursive calculation formula is as follows: Where Δ t The weighted moving average of the current proportion of financial-related transactions, Δ t -1 represents the weighted moving average of financial-related transactions processed in the previous statistical period, and β is the weighting coefficient, where 0 <= β <= 1. To increase the influence of the current proportion, β can be increased; conversely, β can be decreased. Financial institutions need to set upper and lower limits for Δ based on their own business situation. m in and Δ m ax indicates that the target threshold does not need to be adjusted within this upper and lower limit range, therefore the correction formula is as follows: Where Δ is Δ t γ represents the adjustment coefficient, with a value range of (0,1), which can be set according to business needs. This avoids cost and benefit issues caused by fixed thresholds being too large or too small.

[0062] Example 2

[0063] Figure 2 This is a flowchart of a customer category determination method according to Embodiment 2 of the present invention. This embodiment is a further optimization and extension based on the above-described embodiments, and can be combined with various optional technical solutions in the above-described embodiments. Figure 2 As shown, the method includes:

[0064] S210. Collect basic customer information and financial information, and use the basic information and financial information as raw customer characteristic data.

[0065] The basic information refers to information collected and recorded by the financial institution in the process of establishing a connection with the customer and carrying out business, and used to identify the identity of the customer. The financial information can be understood as information associated with the financial business. For example, the basic information can include, but is not limited to, a unique customer identifier, gender, age, region, occupation type, etc. The financial information can include, but is not limited to, the customer's level in the financial institution, loan situation, average credit score in a preset historical period, average account balance, average account transaction frequency, average transaction amount, and account subscription days, etc. For example, the preset historical period can be the past 1 year, the past 6 months, etc.

[0066] In an embodiment, the basic information of the customer, such as the unique customer identifier, gender, age, region, occupation type, and the financial information of the customer, such as the customer's level in the financial institution, loan situation, average credit score in a preset historical period, average account balance, average account transaction frequency, average transaction amount, and account subscription days, etc. can be collected, and the basic information and the financial information are used as original customer feature data. In an embodiment, in order to ensure the timeliness of the data, the information of the account active in the past half year is selected.

[0067] S220, determining the numerical features and the text features in the basic information and the financial information.

[0068] The text feature refers to feature information stored in the form of text and cannot be directly mathematically operated. The numerical feature refers to a quantifiable feature represented by a number, which can directly reflect the quantity, size, degree, and other attributes of things.

[0069] In an embodiment, the numerical features and the text features can be pre-set, or can be divided according to the attribute properties. In order to facilitate subsequent data processing, the unique customer identifier, gender, age, region, occupation type, average account balance, and average credit score can be used as text features, and the customer's level in the financial institution, loan situation, average credit score in a preset historical period, average account balance, average account transaction frequency, average transaction amount, and account subscription days, etc. can be used as numerical features. In an embodiment, the basic information can also be used as a text feature, and the financial information can be used as a numerical feature.

[0070] S230, performing standardization processing on the numerical features to obtain target numerical features, and encoding the text features to obtain target text features.

[0071] In this embodiment, numerical features can be standardized to obtain target numerical features, and the original features can be encoded to obtain target text features. In actual operation, standardizing numerical features so that the mean of each field is 0 and the variance is 1 can eliminate the influence of different indicators due to different attributes. For text features, one-hot encoding can be performed to prepare them for subsequent input into the model.

[0072] In one embodiment, before performing standardization on the numerical features to obtain the target numerical features and encoding the text features to obtain the target text features, missing value processing and duplicate sample processing can also be performed on the numerical features and the text features.

[0073] S240. Use the target numerical features and target textual features as target customer feature data.

[0074] In this embodiment, after determining the target numerical features and the target text features, the target numerical features and the target text features can be merged into target customer feature data.

[0075] S250. Input the target customer feature data into the preset similarity model, and determine the textual similarity score and numerical similarity score of the target customer feature data belonging to the target customer through the preset similarity model. Determine the similarity score according to the textual similarity score and numerical similarity score.

[0076] The textual similarity score can be understood as an indication of whether the target customer feature data, determined based on the target textual features, belongs to the target customer. In practice, for each target numerical feature, the scores can be summed separately and the average value taken as the numerical similarity score.

[0077] In this embodiment, target customer feature data can be input into a preset similarity model, and the similarity score between the target textual features and the textual features of historical business customers can be determined by the preset similarity model as the textual similarity score of the target customer. The similarity score between the target numerical features and the numerical features of historical business customers can be determined as the numerical similarity score of the target customer. The sum of the textual similarity score and the numerical similarity score can be determined as the similarity score. Alternatively, the weighted average of the textual similarity score and the corresponding numerical similarity score can be used as the similarity score.

[0078] In an embodiment, the text type similarity score and the numerical type similarity score of the target customer feature data belonging to the target customer are determined respectively by a preset similarity model, and the similarity score is determined according to the text type similarity score and the numerical type similarity score, comprising:

[0079] The similarity score of the target text type feature in the target customer feature data and the text type feature of the historical business customer is determined by the preset similarity model as the text type similarity score of the target customer;

[0080] The similarity score of the target numerical type feature in the target customer feature data and the numerical type feature of the historical business customer is determined by the preset similarity model as the numerical type similarity score of the target customer;

[0081] The weighted average result of each text type similarity score and the corresponding numerical type similarity score is taken as the similarity score by the preset similarity model.

[0082] The historical business customer can be understood as a customer who has already handled the related business of the financial institution, for example, a customer who has already handled a credit card. The number of historical business customers can be multiple, and the historical business customers can be used to train the preset similarity model, and the historical business customers can be a limited number.

[0083] In an embodiment, the preset similarity model can determine the similarity score of the target text type feature and the text type feature of each historical business customer as the text type similarity score of the target customer, and determine the similarity score of the target numerical type feature in the target customer feature data and the numerical type feature of the historical business customer as the numerical type similarity score of the target customer, determine the weighted average result of each text type similarity score and the corresponding numerical type similarity score as the initial similarity score, and take the maximum value in the initial similarity score as the similarity score. In an embodiment, the number of initial similarity scores is the same as the number of historical business customers. In an embodiment, the determination method of the text type similarity score can be Wherein S c (u,v) represents the text type similarity score of the user to be determined and the historical business customer in the text category feature, D1 is the target text type feature set, length(D1) represents the size of the target text type feature set, i.e. the number of target text type features, g(T ui ,T vi ) represents the similarity score of the i-th category feature of the two users. The calculation formula is: The determination method of the numerical type similarity score can be Wherein S n(u,v) represents the numerical similarity score of the to-be-determined user and the historical service customer on the numerical category feature, D2 is the target numerical feature set, length(D2) represents the size of the target numerical feature set, that is, the number of target numerical features, R ui and R vi respectively represent the i-th numerical feature of the to-be-determined user and the historical service customer. The final score is obtained by weighted average of the text similarity score and the corresponding numerical similarity score, and the calculation formula is as follows: S = λ c S c + λ n S n ; wherein S represents the similarity score, λ c and λ n respectively represent the weight coefficients of the category type and the numerical type, which can be set based on business requirements. In the training process of the preset similarity model, the similarity score of each historical service customer can be determined respectively, and the maximum similarity score is selected as the prediction score of the similarity model.

[0084] S260, input the target customer feature data into the preset classification model, and generate the score of the target customer feature data belonging to the target customer as the classification score through the preset classification model.

[0085] In the embodiment, the target customer feature data can be input into the preset classification model, and the score of the target customer feature data belonging to the target customer is determined through the preset classification model, and the score is taken as the classification score.

[0086] S270, input the similarity score and the classification score into the pre-trained logistic regression meta-learner, and obtain the target score of the to-be-identified customer as the target customer according to the weighted similarity score and the classification score of the logistic regression meta-learner.

[0087] In the embodiment, the similarity score and the classification score can be input into the pre-trained logistic regression meta-learner, and the result is obtained by weighting the similarity score and the classification score through the logistic regression meta-learner, and the result is taken as the target score of the target customer. The logistic regression meta-learner of the present application adopts logistic regression, and the loss function adopts cross-entropy loss. The determination formula is wherein w represents the weight coefficient, z represents the output of the two base learners, which needs to be standardized, and b represents the bias term.

[0088] S280, when it is determined that the target score is greater than the target threshold, the to-be-identified customer is determined as the target customer.

[0089] In the embodiment, when it is determined that the target score is greater than the target threshold, the to-be-identified customer can be taken as the target customer.

[0090] S290. When the target score is determined to be less than or equal to the target threshold, the customer to be identified is determined to be a non-target customer.

[0091] In an embodiment, when the target score is determined to be less than or equal to the target threshold, the customer to be identified can be regarded as a non-target customer.

[0092] In this embodiment of the invention, basic customer information and financial information are collected and used as raw customer feature data. Numerical and textual features within the basic and financial information are determined. Numerical features are standardized to obtain target numerical features, and textual features are encoded to obtain target textual features. These target numerical and textual features are used as target customer feature data, achieving accurate data processing. The target customer feature data is input into a preset similarity model. The preset similarity model determines the textual similarity score and numerical similarity score of the target customer feature data belonging to the target customer. A similarity score is determined based on the textual and numerical similarity scores. The target customer feature data is then input into a preset classification model. The classification model assigns the score of the target customer's characteristic data to the target customer as the classification score. The similarity score and classification score are input into a pre-trained logistic regression meta-learner. The logistic regression meta-learner weights the similarity score and classification score to obtain the target score for identifying the customer as the target customer. If the target score is greater than the target threshold, the customer is identified as the target customer. If the target score is less than or equal to the target threshold, the customer is identified as a non-target customer. The correlation between different customer entities is identified to quantitatively represent the customer's demand for financial products. At the same time, based on the pre-set classification model, customer characteristic information is statistically analyzed to predict the customer's willingness to purchase financial products. Finally, the two techniques are combined through stacked ensemble learning to achieve the identification of the target customer.

[0093] Example 3

[0094] Figure 3 This is a schematic diagram of a customer category determination system according to Embodiment 3 of the present invention, as shown below. Figure 3 As shown, the customer category determination system includes a data acquisition module, a data preprocessing module, a model training module, and a prediction module. This embodiment takes credit card issuance as an example of a financial-related business, and uses potential customers as target customers and non-potential customers as non-target customers to further illustrate a customer category determination method.

[0095] Step 1, collect data through a data collection module, the data source is bank account information and other legal information, and the data is desensitized as necessary, in order to ensure the timeliness of the data, the information of the account active in the past six months is selected, the features include customer basic information and financial information, the basic information includes customer unique identification, gender, age, region, occupation type, etc., the financial information includes customer level in the bank, whether there is a loan, average credit score in the past 6 months, average account balance in the past 6 months, average account transaction frequency in the past 6 months, average account transaction amount in the past 6 months, account signing days, etc., and the label is whether the customer has a credit card in the institution, 1 if yes, otherwise 0.

[0096] Step 2, send the data obtained by the data collection module to the data preprocessing module, and the data preprocessing module is responsible for preprocessing the data. In an embodiment, Figure 4 is a flowchart of a data preprocessing module according to an embodiment of the application. It includes missing value processing, duplicate sample processing, class balancing, and data set division.

[0097] First, data cleaning is performed, including missing value processing and duplicate value processing.

[0098] The input raw data is processed for missing values, which needs to be processed separately for different features. In an embodiment, Figure 5 is a flowchart of missing value processing according to an embodiment of the application. As Figure 5 shown, specifically, the first numerical feature and the second text feature can be subjected to missing value processing and duplicate sample processing. When it is detected that the first numerical feature and the second text feature contain missing values (such as a user's 6-month "income" field, one of which is empty), it can be determined whether the average index value. If the sample is the first numerical feature, it is the average index value; if the sample is the first text feature, it is not the average index value. If it is not the average index value, discard the sample; if it is the average index value, calculate the average of the complete data (such as taking the average of the values corresponding to the "income" of the other 5 months), and fill the missing position with the calculated average value, while marking the sample "was missing value, has been filled with average value", to prevent subsequent operations from being processed. The raw data is the original customer feature data.

[0099] In an embodiment, Figure 6 is a flowchart of duplicate sample processing according to an embodiment of the application. As Figure 6As shown, the data after missing value processing is processed for repeated samples. Since the customer unique identifier can represent the information of the customer, the deletion of repeated samples is performed according to the identifier. First, samples belonging to the same user are grouped according to the identifier, and then each column feature of each group of samples is processed. Only one sample is finally retained for each group. Specifically, each row of samples (i.e. each category in the numerical features and the text features) of the data set is traversed to ensure that all samples are processed. If “no”, it is directly ended, and if “yes”, the feature column processing is entered. Each column (feature, such as “age” and “income”) of the current sample is processed separately, and different features can follow different logics. If “no”, it returns to the sample loop, and if “yes”, the cell repair is entered. For a single cell of a column feature (such as the “income” in the third row), it is determined whether the value is abnormal (inconsistent / missing), and then repaired according to the rules. It is determined whether the feature value is consistent. If yes, the feature value is directly retained (the cell value is normal and does not need to be processed), and if no, the “feature value abnormality” processing flow is entered (the value is missing / inconsistent and needs to be repaired). It is determined whether to average the index value (numerical feature priority strategy). If yes, it is attempted to repair with “average value”, which is divided into two cases: if yes, the calculated accurate value is directly obtained (reuse of historical statistics to avoid repeated calculation). If no, the original data is obtained, the average value is calculated, and the average value is filled. If no, other strategies are attempted. If the average value is not used, the following priority is attempted to repair: it is determined whether there is a value range. If yes for the numerical feature, the average value is calculated and filled within the range. If no for the text feature, the feature value is selected and filled within the range. If none, a value is randomly selected and filled.

[0100] Secondly, category balancing is performed. Here, the majority samples are deleted by undersampling to ensure that the number of samples of the two categories is basically consistent. If the number of samples is too small after deletion, the deletion is cancelled, and a high weight is set for the minority samples.

[0101] Then, standardization and encoding processing are performed. The numerical features are standardized so that the mean value of each field is 0 and the variance is 1, thereby eliminating the influence of different attributes between different indicators. The text features are onehot encoded to prepare for subsequent similarity calculation and ANN training.

[0102] The last step of the data preprocessing module is to divide the data set. The data set is randomly sampled from the positive samples and the negative samples. 80% of the positive samples and 80% of the negative samples are randomly sampled as the training set, and the remaining 20% of the positive samples and 20% of the negative samples are randomly sampled as the test set.

[0103] Step 3, the prediction model is built and predicted, the present application uses similarity and ANN to build base learner, combines with Bayesian optimization algorithm to optimize the hyperparameters of ANN model, uses logistic regression meta-learner to weight the prediction results of both, so as to obtain the final prediction result. The preset similarity model is based on the similarity of the user and the card user (business user), and the similarity score is obtained, and the ANN model (preset classification model) is used to predict the potential card score of the user, and the similarity score and the potential card score are learned through the stacking integration algorithm, and the final prediction score of the user is obtained. Compared with the set threshold, whether it is a potential customer is determined.

[0104] The preset similarity model needs to calculate the similarity of the user and the card user, and the card user does not contain the current user. Since the data used is for the past six months, it has timeliness. The determination method of the text type similarity score can be Where S c (u,v) represents the text type similarity score of the user to be determined and the historical business customer in the text category feature, D1 is the target text type feature set, length(D1) represents the size of the target text type feature set, that is, the number of target text type features, g(T ui ,T vi ) represents the similarity score of the i-th category feature of the two users. The calculation formula is: The determination method of the numerical type similarity score can be Where S n (u,v) represents the numerical type similarity score of the user to be determined and the historical business customer in the numerical category feature, D2 is the target numerical feature set, length(D2) represents the size of the target numerical feature set, that is, the number of target numerical type features, R ui and R vi represent the i-th numerical feature of the user to be determined and the historical business customer respectively. The text type similarity score and the corresponding numerical type similarity score are obtained by weighted average to obtain the final score, and the calculation formula is as follows: S = λ c S c + λ n S n ; wherein S represents the similarity score, λ c and λ n represent the weight coefficients of the category type and the numerical type respectively, which can be set based on business requirements. The similarity score of each user with the historical business customer can be determined respectively, and the maximum similarity score is selected as the prediction score of the similarity model.

[0105] The preset classification model is built, the full training set sample is selected, the model learner is set as y=F(x), since it is a binary classification problem, cross entropy is used as the loss function, and the loss function calculation formula of a single sample is as follows: Wherein, y is the true value of a single sample, The predicted value of a single sample is represented. The total loss function calculation formula is as follows:

[0106] The number of layers of the preset classification model, the number of nodes of each layer, the learning rate, the regularization coefficient and other hyperparameters can be set, the model weight and other parameters are initialized, the Adam algorithm is selected as the optimizer algorithm, the initial learner is obtained, and then the hyperparameters are optimized through the Bayesian optimization algorithm, and the final ANN binary classification model is obtained as the preset classification model.

[0107] The logistic regression meta-learner of the application adopts logistic regression, and the loss function adopts cross entropy loss. Its determination formula is Wherein w represents the weight coefficient, z represents the output of two base learners, which needs to be standardized, and b represents the bias term.

[0108] The score of the classified potential user needs to be used during prediction, and the score is compared with the set adaptive threshold (target consistency) θ, if the threshold is exceeded, the user is considered as a potential user, otherwise, it is a non-potential user.

[0109] Due to the subsequent credit card handling of potential users and the addition of new bank users, in order to ensure the timeliness of the model, it is necessary to retrain after a period of time, the interval time is set by the bank according to its own business needs, and the threshold θ needs to be adjusted during retraining, and the threshold correction needs to consider the proportion of potential customers after the recommendation of potential customers to handle cards α t , the historical proportion of the past year α1, α2... α t-1 , the current proportion and the historical proportion are considered comprehensively, the card handling proportion weighted moving average is set as Δ, and the recursive calculation formula is as follows: Wherein Δ t is the weighted moving average of the current financial related business proportion, Δ t -1 is the weighted moving average of the financial related business in the last statistical period, β is the weight coefficient, and 0<=β<=1, if you want to increase the influence of the current proportion, you can increase β, and vice versa. The upper and lower limits of Δ of the financial institution need to be set according to the business situation of the financial institution Δ m in and Δ m ax, indicating that the target threshold does not need to be adjusted in the range, so the correction formula is as follows: Wherein, Δ is Δ tγ represents an adjustment coefficient, and the value range is (0, 1), which can be set according to business needs.

[0110] It can be concluded that if the predicted potential customer has a high card application rate after being recommended, the threshold can be appropriately lowered to expand the recommended range of potential customers, otherwise if the proportion is low, in order to reduce the cost of recommendation and marketing, the threshold can be appropriately increased to reduce the recommended range of potential customers.

[0111] In this embodiment, on the one hand, the similarity between the customers who have not applied for a card and the customers who have applied for a card is calculated, the correlation between different customer entities is found out, the demand degree of the customers for the credit card product is quantitatively represented, and based on the binary classification model of ANN, the customer feature information is statistically analyzed to quantitatively predict the card application willingness degree of the customers. Then, the potential customers are identified by combining the two technologies through stacked ensemble learning, and through this integrated manner, the similarity between the customers and the customer characteristics can be comprehensively and overall considered, and the generalization ability of the model is improved. On the other hand, in order to ensure the timeliness of the model, the strategy of retraining after a period of time is adopted, the dynamic adaptive threshold is used for the adjudication of the final result, the threshold is set by comprehensively considering the current card application rate and the historical card application rate, the threshold can be automatically adjusted according to the increase and decrease of the proportion, the cost and benefit problems caused by the fixed threshold being too large or too small are avoided, and the bank is given the power to set the upper and lower limits of the threshold according to its own business situation, which matches the business development needs.

[0112] Embodiment Four

[0113] Figure 7 is a structural schematic diagram of a customer category determination apparatus provided by Embodiment Four of the present application. As shown in Figure 7 , the apparatus comprises a data acquisition module 71, a score determination module 72 and a category determination module 73.

[0114] The data acquisition module 71 is configured to acquire original customer feature data of a to-be-identified customer, and obtain target customer feature data by performing data preprocessing on the original customer feature data.

[0115] The score determination module 72 is configured to determine a similarity score corresponding to the target customer feature data based on a preset similarity model, and generate a classification score corresponding to the target customer feature data based on a preset classification model.

[0116] The category determination module 73 is configured to determine a target score according to a pre-trained logistic regression meta-learner, the similarity score and the classification score, and determine a customer category of the to-be-identified customer according to the target score. The preset similarity model, the preset classification model and the logistic regression meta-learner adopt stacked ensemble learning, the preset similarity model and the preset classification model are base learning layers, and the logistic regression meta-learner is a meta-learning layer.

[0117] The technical scheme of the embodiment of the application obtains the original customer feature data of the to-be-identified customer through the data acquisition module, obtains target customer feature data by performing data preprocessing on the original customer feature data, determines a similarity score corresponding to the target customer feature data based on a preset similarity model, generates a classification score corresponding to the target customer feature data based on a preset classification model, determines a target score according to the similarity score and the classification score by the category determination module, and determines the customer category of the to-be-identified customer according to the target score, so as to realize the identification of the customer category in combination with the correlation between customers and the customer feature information, and improve the accuracy of determining the customer category of the to-be-identified customer. Meanwhile, the advantages and application ranges of the two technologies of the preset similarity model and the preset classification model are fused in a stacked ensemble learning manner, and the generalization ability of the model is improved.

[0118] In an embodiment, the data acquisition module 71 comprises:

[0119] The data acquisition unit is configured to acquire the basic information and the financial information of the customer, and take the basic information and the financial information as the original customer feature data.

[0120] The feature identification unit is configured to determine the numerical features and the text features in the basic information and the financial information.

[0121] The feature coding unit is configured to perform standardization processing on the numerical features to obtain target numerical features, and perform coding on the text features to obtain target text features.

[0122] The feature determination unit is configured to take the target numerical features and the target text features as the target customer feature data.

[0123] In an embodiment, the score determination module 72 comprises:

[0124] The first score determination unit is configured to input the target customer feature data into the preset similarity model, determine a text similarity score and a numerical similarity score of the target customer feature data belonging to the target customer through the preset similarity model, and determine the similarity score according to the text similarity score and the numerical similarity score.

[0125] The second score determination unit is configured to input the target customer feature data into the preset classification model, and generate a score of the target customer feature data belonging to the target customer as the classification score through the preset classification model.

[0126] In an embodiment, the first score determination unit is specifically configured to:

[0127] determine a similarity score of the target text features in the target customer feature data and the text features of the historical business customers as the text similarity score of the target customer through the preset similarity model.

[0128] determine, by the preset similarity model, a similarity score of a target numerical feature in the target customer feature data and a numerical feature of the historical business customer as a numerical similarity score of the target customer;

[0129] determine, by the preset similarity model, a weighted average result of each text similarity score and a corresponding numerical similarity score as the similarity score.

[0130] In an embodiment, the category determining module 73 comprises:

[0131] a target score determining unit configured to input the similarity score and the classification score into a pre-trained logistic regression meta-learner, and obtain a target score of the to-be-identified customer being the target customer by weighting the similarity score and the classification score by the logistic regression meta-learner;

[0132] a first category determining unit configured to determine that the to-be-identified customer is the target customer when the target score is greater than a target threshold;

[0133] a second category determining unit configured to determine that the to-be-identified customer is a non-target customer when the target score is less than or equal to the target threshold.

[0134] In an embodiment, the customer category determining apparatus further comprises:

[0135] a first feature determining module configured to respectively collect customer feature data and customer business categories of each customer at a historical time, determine a numerical feature in the customer feature data as a first numerical feature, and determine a text feature in the customer feature data as a first text feature;

[0136] a second feature determining module configured to perform data cleaning on the first numerical feature and the second text feature to obtain a second numerical feature and a second text feature;

[0137] a target feature determining module configured to perform category balancing on the second numerical feature and the second text feature according to the customer business categories, perform standardization processing on the category-balanced numerical feature to obtain a target numerical feature, and perform encoding on the category-balanced text feature to obtain a target text feature;

[0138] a test set determining module configured to take the target numerical feature and the corresponding target text feature as first customer feature data, and divide the first customer feature data into a training set and a test set in a random sampling manner;

[0139] The training module is configured to input a training set into a preset classification model to perform training, and a loss function of the preset classification model is a cross-entropy loss function.

[0140] In an embodiment, the customer category determination apparatus further comprises:

[0141] The first value determination module is configured to determine a quantity of business customers in the customer category at a current time as a first quantity, determine a quantity of business customers at a first historical time as a second quantity, and determine a quantity of business customers at a second historical time as a third quantity, wherein the second historical time is located in a previous acquisition period of the first historical time.

[0142] The second value determination module is configured to determine a difference between the first quantity and the second quantity as a first difference, and determine a difference between the second quantity and the third quantity as a second difference.

[0143] The third value determination module is configured to determine a ratio of the first difference to the first quantity as a first ratio, and determine a ratio of the second difference to the second quantity as a second ratio.

[0144] The threshold value updating module is configured to update the target threshold value according to the first ratio and the second ratio.

[0145] The customer category determination apparatus provided in the embodiments of the present application can perform the customer category determination method provided in any of the embodiments of the present application, and has the corresponding function modules and beneficial effects of the execution method.

[0146] Embodiment five

[0147] Figure 8 is a structural schematic diagram of an electronic device for implementing a customer category determination method according to an embodiment of the present application. The electronic device is intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular telephones, smart phones, wearable devices (such as helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions, are meant to be examples only, and are not intended to limit the implementations of the present application described and / or claimed in this document.

[0148] As Figure 8As shown, the electronic device 10 includes at least one processor 11, and a memory, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., communicatively connected to the at least one processor 11, where the memory stores a computer program executable by the at least one processor. The processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or loaded into the random access memory (RAM) 13 from the storage unit 18. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other through a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0149] Various components in the electronic device 10 are connected to the I / O interface 15, including an input unit 16, such as a keyboard, a mouse, etc., an output unit 17, such as various types of displays, a speaker, etc., a storage unit 18, such as a magnetic disk, an optical disk, etc., and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices through a computer network, such as the Internet, and / or various telecommunication networks.

[0150] The processor 11 can be various general and / or special purpose processing components with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 11 performs various methods and processes described above, such as a customer category determination method.

[0151] In some embodiments, a customer category determination method can be implemented as a computer program tangibly embodied in a computer readable storage medium, such as the storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed onto the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of a customer category determination method described above can be performed. Alternatively, in other embodiments, the processor 11 can be configured to perform a customer category determination method by any other appropriate means, such as by means of firmware.

[0152] The various embodiments of the systems and techniques described above can be implemented in digital electronic circuitry, integrated circuitry, a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), a system on a chip (SOC), a load programmable logic device (CPLD), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include implementation in one or more computer programs that are executable and / or interpretable on a programmable system including at least one programmable processor, which can be special or general purpose, coupled to receive data and instructions from, and to transmit data and instructions to, a storage system, at least one input device, and at least one output device.

[0153] Computer programs used to implement the processes of the application can be written in any combination of one or more programming languages. These computer programs can be provided to a processor of a general purpose computer, special purpose computer, or other programmable data processing apparatus, such that the computer program

[0154] In the context of the present application, a computer-readable storage medium can be a tangible medium that can contain or store computer programs for use by or in connection with an instruction execution system, apparatus, or device. Computer-readable storage media can include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium can be a machine-readable signal medium. More specific examples of the machine-readable storage medium will include one or more lines of a program of instructions in a transitory signal, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0155] To provide for interaction with a user, the systems and techniques described here can be implemented on an electronic device having a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the electronic device. Other kinds of devices can be used to provide for interaction with a user as well; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form, including acoustic, speech, or tactile input.

[0156] The systems and techniques described here can be implemented in a computing system that includes a back end component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front end component (e.g., a user computer having a graphical user interface or a Web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such back end, middleware, or front end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0157] The computing system can include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other. A server can be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system, to solve the defects of large management difficulty and weak business scalability in traditional physical host and VPS service.

[0158] It should be understood that the various forms of flow shown above can be re-ordered, added to, or deleted from without departing from the scope of the present disclosure. For example, the steps recited in the present disclosure can be executed in parallel, executed in sequence, or executed in a different order, as long as the desired results of the present disclosure are achieved, and the present disclosure is not limited herein.

[0159] The specific embodiments described above are not intended to be limiting, and persons skilled in the art will appreciate that various modifications, combinations, sub-combinations and alternatives can be made to the specific embodiments without departing from the spirit and principles of the disclosure. Accordingly, the disclosure is not limited to the specific embodiments described above, but only by the scope of the appended claims.

Claims

1. A customer category determining method characterized by, The method comprises the following steps: obtaining original customer feature data of a to-be-identified customer, and performing data preprocessing on the original customer feature data to obtain target customer feature data; determining a similarity score corresponding to the target customer feature data based on a preset similarity model, and generating a classification score corresponding to the target customer feature data based on a preset classification model; determining a target score according to a pre-trained logistic regression meta-learner, the similarity score and the classification score, and determining a customer category of the to-be-identified customer according to the target score; wherein the preset similarity model, the preset classification model and the logistic regression meta-learner adopt stacked ensemble learning, the preset similarity model and the preset classification model are base learning layers, and the logistic regression meta-learner is a meta-learning layer.

2. The method of claim 1, wherein, The method comprises the following steps: collecting basic information and financial information of a customer, and taking the basic information and the financial information as original customer feature data; determining numerical features and text features in the basic information and the financial information; performing standardization processing on the numerical features to obtain target numerical features, and performing encoding on the text features to obtain target text features; taking the target numerical features and the target text features as target customer feature data.

3. The method of claim 1, wherein, The method comprises the following steps: inputting the target customer feature data into a preset similarity model, determining a text similarity score and a numerical similarity score of the target customer feature data belonging to a target customer through the preset similarity model, and determining a similarity score according to the text similarity score and the numerical similarity score; inputting the target customer feature data into a preset classification model, and generating a score of the target customer feature data belonging to a target customer as a classification score through the preset classification model.

4. The method of claim 3, wherein, The method comprises the following steps: determining a similarity score of a target text feature in the target customer feature data and a text feature of a historical business customer as a text similarity score of the target customer through a preset similarity model; determining a similarity score of a target numerical feature in the target customer feature data and a numerical feature of a historical business customer as a numerical similarity score of the target customer through a preset similarity model; determining a weighted average result of each text similarity score and a corresponding numerical similarity score as a similarity score through a preset similarity model.

5. The method of claim 1, wherein, The method comprises the following steps: inputting the similarity score and the classification score into a pre-trained logistic regression meta-learner, weighting the similarity score and the classification score according to the logistic regression meta-learner to obtain a target score of the to-be-identified customer being a target customer; determining that the to-be-identified customer is a target customer when it is determined that the target score is greater than a target threshold; determining that the to-be-identified customer is a non-target customer when it is determined that the target score is less than or equal to the target threshold.

6. The method of claim 1, wherein, The training process of the preset classification model comprises: collecting customer feature data and customer business categories of each customer in a historical moment, determining numerical features in the customer feature data as first numerical features, and determining text features in the customer feature data as first text features; performing data cleaning on the first numerical features and the second text features to obtain second numerical features and second text features; performing category balancing on the second numerical features and the second text features according to the customer business categories, performing standardization processing on the category-balanced numerical features to obtain target numerical features, and performing encoding on the category-balanced text features to obtain target text features; dividing the target numerical features and the corresponding target text features as first customer feature data into a training set and a test set in a random sampling manner; inputting the training set into the preset classification model constructed in advance for training, and the loss function of the preset classification model is a cross-entropy loss function; wherein the preset classification model is based on an artificial neural network.

7. The method of claim 5, wherein, After determining the target score according to the similarity score and the classification score, and determining the customer category of the to-be-identified customer according to the target score, the method further comprises: determining the number of business customers in the customer category at the current moment as a first number, determining the number of business customers at a first historical moment as a second number, and determining the number of business customers at a second historical moment as a third number; wherein the second historical moment is located in a previous acquisition period of the first historical moment; determining a first difference value as the difference between the first number and the second number, and determining a second difference value as the difference between the second number and the third number; determining a first ratio value as the ratio of the first difference value to the first number, and determining a second ratio value as the ratio of the second difference value to the second number; updating the target threshold according to the first ratio value and the second ratio value.

8. A customer category determining apparatus characterized by comprising: The method comprises: a data acquisition module configured to acquire original customer feature data of a to-be-identified customer, and perform data preprocessing on the original customer feature data to obtain target customer feature data; a score determination module configured to determine a similarity score corresponding to the target customer feature data based on a preset similarity model, and generate a classification score corresponding to the target customer feature data based on a preset classification model; The category determining module is configured to determine a target score according to a pre-trained logistic regression meta-learner, the similarity score and the classification score, and determine the customer category of the customer to be identified according to the target score; wherein the preset similarity model, the preset classification model and the logistic regression meta-learner adopt stacked ensemble learning, the preset similarity model and the preset classification model are base learning layers, and the logistic regression meta-learner is a meta-learning layer.

9. An electronic device, comprising: The electronic device comprises: at least one processor; and a memory connected to the at least one processor in communication; wherein The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor to enable the at least one processor to execute the customer category determination method of any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer instructions for enabling the processor to implement the customer category determination method of any one of claims 1-7 when executed.

Images