Customer group classification method and device, equipment, medium and program product

Abstract

The invention provides a customer group classification method which can be applied to the technical field of big data and the technical field of financial science and technology. The customer group classification method comprises the steps that customer group information is acquired, at least one evaluation index of each user feature is determined based on multiple user features included in the customer group information, and the evaluation indexes are used for representing the distinguishing ability of the user features relative to sample classification and / or the distribution stability of the user features on different sample sets; screening the user features based on the evaluation indexes, and determining a target feature set; based on the target feature set, determining a corresponding target classification model from a plurality of preset classification models, and processing the target feature set by using the target classification model to obtain a target rule, the target rule comprising a single rule or a combined rule; and classifying the plurality of users in the customer group information by using the target rule to obtain a classification result. The invention also provides a customer group classification device, equipment, a medium and a program product.

Description

Technical Field

[0001] This application relates to the fields of big data technology and financial technology technology, and more specifically to a customer group classification method, apparatus, equipment, medium and program product. Background Technology

[0002] Currently, in the financial system, customer groups need to be classified in scenarios such as identifying users' repayment ability. Customer group classification usually relies on the personal experience of business personnel to study and classify customers' business nature, revenue capacity, profitability, and historical operating conditions, or to use classification models trained on historical sample data to classify users. However, customer attribute features are usually numerous, historical data is large, and there are many features that are irrelevant or relatively unimportant to the classification results, resulting in low accuracy of the classification model and a lot of time consumption. Summary of the Invention

[0003] In view of the above problems, this application provides a customer group classification method, apparatus, equipment, medium and program product.

[0004] According to a first aspect of this application, a customer group classification method is provided, comprising: acquiring customer group information; determining at least one evaluation index for each user feature based on multiple user features included in the customer group information, wherein the evaluation index is used to represent the discriminative power of the user feature relative to sample classification and / or the distribution stability on different sample sets; filtering user features based on the evaluation index to determine a target feature set; determining a corresponding target classification model from multiple preset classification models based on the target feature set; processing the target feature set using the target classification model to obtain target rules, wherein the target rules include single rules or combined rules; and classifying multiple users in the customer group information using the target rules to obtain classification results.

[0005] According to an embodiment of this application, the process of filtering user features based on evaluation metrics to determine a target feature set includes: filtering user features based on evaluation metrics and corresponding threshold values ​​to obtain first filtered features; performing collinearity judgment on the first filtered features to obtain collinearity judgment results, the collinearity judgment results including whether there is a linear relationship between the first filtered features; filtering the first filtered features based on the collinearity judgment results to obtain second filtered features; and performing logistic regression processing on the second filtered features to obtain a target feature set.

[0006] According to embodiments of this application, the multiple preset classification models include at least a first classification model and a second classification model. The multiple preset classification models are pre-constructed based on the following method: obtaining a training set including historical sample data; when a target feature is selected or an instruction to classify using a single rule is received, generating a single rule corresponding to the target feature based on the value distribution of the target feature in the training set and the corresponding classification label, and storing it in a rule base to construct the first classification model; when multiple target features are selected, using the training set, constructing a decision tree model with the multiple target features as input features to construct the second classification model.

[0007] According to an embodiment of this application, the method further includes: training a first decision tree model based on all training data in the training set, using multiple target features as input features; dividing the training set based on user type to obtain multiple sub-training sets, wherein each sub-training set corresponds to a user type; and training a second decision tree model using the multiple sub-training sets, using multiple target features as input features.

[0008] According to an embodiment of this application, at least one evaluation index of a user feature is determined based on the following method: obtaining dynamic binning information pre-associated with the user feature and preset statistical parameters corresponding to each bin, wherein the dynamic binning information represents multiple binning boundaries adaptively determined under the condition of satisfying constraints; and calculating at least one evaluation index for each user feature based on the dynamic binning information and the preset statistical parameters.

[0009] According to an embodiment of this application, when the target rule includes a combination rule, classifying multiple users in the customer group information using the target rule to obtain the classification result includes: filtering the combination rules based on their respective evaluation values ​​to determine the target rule; classifying the multiple users included in the customer group using the target rule to obtain the classification result; wherein, the evaluation values ​​of each combination rule are obtained based on the following method: testing at least one combination rule to determine the effectiveness index of the combination rule, the effectiveness index including at least recall and precision; calculating the evaluation value of each combination rule using the weights of each effectiveness index and a pre-constructed weighting function, wherein the weights are used to characterize the preference for different effectiveness indexes.

[0010] According to an embodiment of this application, the method further includes: in response to receiving a task to update a preset classification model, acquiring new sample data; based on the new sample data, determining the volatility of a single rule or combined rule effectiveness index; and in response to the volatility being greater than a preset threshold, updating the preset classification model using the new sample data and historical sample data.

[0011] A second aspect of this application provides a customer group classification device, comprising: a determination module, configured to acquire customer group information, and based on multiple user features included in the customer group information, determine at least one evaluation index for each user feature, wherein the evaluation index is used to represent the discriminative power of the user feature relative to sample classification and / or the distribution stability on different sample sets; a first processing module, configured to filter user features based on the evaluation index to determine a target feature set; a second processing module, configured to determine a corresponding target classification model from multiple preset classification models based on the target feature set, and process the target feature set using the target classification model to obtain target rules, wherein the target rules include single rules or combined rules; and a classification module, configured to classify multiple users in the customer group information using the target rules to obtain classification results.

[0012] A third aspect of this application provides an electronic device comprising: one or more processors; and a memory for storing one or more computer programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the method described above.

[0013] A fourth aspect of this application also provides a computer-readable storage medium having a computer program or instructions stored thereon, which, when executed by a processor, implement the steps of the above-described method.

[0014] The fifth aspect of this application also provides a computer program product, including a computer program or instructions that, when executed by a processor, implement the steps of the above-described method.

[0015] According to embodiments of this application, user features are quantitatively screened based on evaluation metrics of discriminative ability and / or distribution stability to obtain a concise and stable set of target features. A matching target classification model is automatically determined from multiple preset classification models, and corresponding target rules are generated for customer group classification. This method effectively eliminates irrelevant or unstable features, improves the accuracy and robustness of the classification model, reduces modeling and computational complexity, shortens processing time, and saves CPU and memory requirements. Attached Figure Description

[0016] The above-mentioned contents, other objects, features and advantages of this application will become clearer from the following description of embodiments with reference to the accompanying drawings, in which:

[0017] Figure 1 The illustrations depict application scenarios of customer classification methods, apparatus, devices, media, and program products according to embodiments of this application.

[0018] Figure 2 A flowchart illustrating a customer group classification method according to an embodiment of this application is shown schematically.

[0019] Figure 3 A schematic diagram of a decision tree model according to an embodiment of this application is shown.

[0020] Figure 4 A flowchart illustrating the determination of target rules according to an embodiment of this application is shown schematically;

[0021] Figure 5 A flowchart illustrating the determination of a target feature set according to an embodiment of this application is shown schematically;

[0022] Figure 6 A schematic diagram illustrating the structure of a customer classification device according to an embodiment of this application is shown; and

[0023] Figure 7 A block diagram schematically illustrates an electronic device suitable for implementing a customer group classification method according to an embodiment of this application. Detailed Implementation

[0024] The embodiments of this application will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of this application. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of this application for ease of explanation. However, it will be apparent that one or more embodiments may be implemented without these specific details. Furthermore, descriptions of well-known structures and technologies are omitted in the following description to avoid unnecessarily obscuring the concepts of this application.

[0025] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0026] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0027] When using expressions such as "at least one of A, B and C", they should generally be interpreted in accordance with the meaning that is commonly understood by those skilled in the art (e.g., "a system having at least one of A, B and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B and C, etc.).

[0028] It should be noted that the customer classification methods, apparatus, equipment, media, and program products defined in this application can be used in the fields of big data technology and fintech technology, and can also be used in a variety of other fields besides big data technology and fintech technology. The application fields of the customer classification methods, apparatus, equipment, media, and program products provided in the embodiments of this application are not limited.

[0029] In the technical solution of this application, the user information (including but not limited to user personal information, user image information, user device information, such as location information) and data (including but not limited to data used for analysis, stored data, and displayed data) involved are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of related data all comply with relevant laws, regulations, and standards, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entry points for users to choose to authorize or refuse.

[0030] In scenarios where personal information is used for automated decision-making, the customer classification method, device, and system provided in this application all offer users corresponding operation entry points, allowing them to choose to agree to or reject the automated decision results; if the user chooses to reject, the process proceeds to the expert decision-making process. Here, "expert decision-making" refers to the decision-making activities of personnel who specialize in a particular field, possess specialized experience, knowledge, and skills, and have reached a certain level of professional expertise.

[0031] Figure 1 The illustration shows an application scenario diagram of the customer group classification method, apparatus, device, medium, and program product according to embodiments of this application.

[0032] like Figure 1 As shown, application scenario 100 according to this embodiment may include a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104, and a server 105. The network 104 serves as a medium for providing a communication link between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. The network 104 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0033] Users can use the first terminal device 101, the second terminal device 102, and the third terminal device 103 to interact with the server 105 via the network 104 to receive or send messages, etc. Various communication client applications can be installed on the first terminal device 101, the second terminal device 102, and the third terminal device 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social media platform software, etc. (for example only).

[0034] The first terminal device 101, the second terminal device 102, and the third terminal device 103 can be various electronic devices with displays and support web browsing, including but not limited to smartphones, tablets, laptops, and desktop computers.

[0035] Server 105 can be a server that provides various services, such as a backend management server that supports websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103 (this is just an example). The backend management server can analyze and process data such as received user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal devices.

[0036] It should be noted that the customer classification method provided in this application embodiment can generally be executed by server 105. Correspondingly, the customer classification device provided in this application embodiment can generally be located in server 105. The customer classification method provided in this application embodiment can also be executed by a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105. Correspondingly, the customer classification device provided in this application embodiment can also be located in a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105.

[0037] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0038] The following will be based on Figure 1 The described scene, through Figures 2-5 The customer group classification method according to the embodiments of this application will be described in detail.

[0039] Figure 2 A flowchart illustrating a customer group classification method according to an embodiment of this application is shown.

[0040] like Figure 2 As shown, the customer group classification method in this embodiment includes operations S210 to S240, and the customer group classification method can be executed by the server 105.

[0041] In operation S210, customer group information is obtained, and based on the multiple user characteristics included in the customer group information, at least one evaluation index is determined for each user characteristic.

[0042] In the embodiments of this application, the evaluation index is used to represent the discriminative power of user features relative to sample classification and / or the distribution stability on different sample sets.

[0043] Customer segment information refers to a set of data collected for a specific financial business that reflects customer characteristics and behavior, such as basic customer attributes, operational data, transaction records, and repayment records. User characteristics are specific fields or indicators extracted from customer segment information for individual customers that can be used for modeling and analysis, such as company type, operating income, debt ratio, and historical overdue payment frequency.

[0044] Specifically, customer basic information, transaction records, asset status, repayment performance, and other data can be collected and integrated through business systems, credit reporting systems, and external data interfaces to form structured customer group information. Then, user features that can be used for classification analysis are extracted from these features. Based on historical sample data, statistical or machine learning methods are used to calculate evaluation indicators that reflect the distinguishing ability and / or distribution stability of each feature, measuring the difference between different categories of samples and the consistency of distribution in different time periods or batches of samples.

[0045] In operation S220, user characteristics are filtered based on evaluation indicators to determine the target feature set.

[0046] In this application embodiment, the target feature set can refer to the subset of features that have strong discriminative power for classification results and have a relatively stable distribution after evaluating and screening all user features. It can be understood as the concise, effective and suitable feature combination selected from all features for modeling.

[0047] For example, one or more evaluation indicators can be calculated for each feature, and then corresponding thresholds or sorting rules can be set. For example, threshold filtering can be to set a threshold for each evaluation indicator separately, and sorting can be to first filter out features with low discrimination ability scores, and then remove features with stability indicators greater than a certain threshold from the remaining features, retaining only features that have both high discrimination ability and stability, and finally forming the target feature set.

[0048] In operation S230, based on the target feature set, the corresponding target classification model is determined from multiple preset classification models, and the target feature set is processed using the target classification model to obtain the target rules.

[0049] In the embodiments of this application, the target rule includes a single rule or a combination of rules.

[0050] The pre-defined classification model can refer to a set of pre-configured classification models of different types, structures, or parameter settings. For example, it could be a rule engine that classifies based on judgment rules, threshold judgments, or decision tables, or machine learning models such as logistic regression, decision tree, and random forest. The target classification model can refer to the model selected from the above pre-defined models after comprehensively comparing their classification performance for a given set of target features, and used for actual modeling and rule generation. The target rule can refer to a set of specific rules that can be directly used to classify users, obtained by training or inferring the target classification model based on the target feature set. A single rule can refer to classifying a user using a single user feature, while a combined rule can refer to classifying a user using multiple user features.

[0051] Specifically, the target feature set and its corresponding labeled historical sample data can be input into various preset classification models for training and validation. Performance metrics such as classification accuracy and recall for each model on the validation set can be calculated. Alternatively, a comprehensive score can be set according to business requirements, and the performance of different models can be compared to select the model with the best performance or the optimal balance between performance and complexity as the target classification model. Then, this target classification model is used to further model and process the target feature set. For example, feature partitioning paths corresponding to each leaf node can be extracted from a decision tree model, and feature coefficients can be converted into binning and scoring rules for a logistic regression model. The partitioning paths of the decision tree or the scoring rules of the logistic regression model can be transformed into executable target rules for customer group classification.

[0052] For example, when the preset classification model includes two types, logistic regression model and decision tree model, the same batch of target features and sample data can be input into the two types of models for training. If the verification results show that the decision tree model has a higher recall rate in customer identification and the risk of overfitting is controllable, then the decision tree model is determined as the target classification model. Then, the rules corresponding to the path are extracted from the decision tree and organized into a set of configurable and executable target rules.

[0053] In operation S240, multiple users in the customer group information are classified using the target rules to obtain the classification results.

[0054] In this embodiment of the application, after obtaining the target rule, the target rule is applied to the target feature data corresponding to each user in the customer group information, and each user is judged and grouped one by one. For example, the system matches and calculates the various features of the user according to the target rule and outputs the corresponding classification label or category number, such as classifying the user into different credit levels, marketing customer groups or management strategy layers.

[0055] It is understood that the embodiments of this application quantitatively screen user features based on evaluation metrics of discriminative ability and / or distribution stability to obtain a concise and stable set of target features. Then, it automatically determines the matching target classification model from multiple preset classification models and generates corresponding target rules for customer group classification. This method can effectively eliminate irrelevant or unstable features, improve the accuracy and robustness of the classification model, reduce modeling and computational complexity, shorten processing time, and save CPU and memory requirements.

[0056] According to an embodiment of this application, the process of filtering user features based on evaluation metrics to determine a target feature set includes: filtering user features based on evaluation metrics and corresponding threshold values ​​to obtain first filtered features; performing collinearity judgment on the first filtered features to obtain collinearity judgment results, the collinearity judgment results including whether there is a linear relationship between the first filtered features; filtering the first filtered features based on the collinearity judgment results to obtain second filtered features; and performing logistic regression processing on the second filtered features to obtain a target feature set.

[0057] In this embodiment, pre-selected evaluation indicators, such as discrimination ability scores and stability indicators, are calculated for each user characteristic, and corresponding thresholds are set for each type of indicator. Then, the evaluation indicators are compared with their thresholds, and features whose evaluation indicators are below the thresholds or do not meet the requirements are eliminated, leaving only the features that meet the evaluation results as the first set of selected features.

[0058] For example, evaluation metrics include Information Value (IV) and Population Stability Index (PSI) corresponding to each user feature. The threshold values ​​for these metrics can be preset. For instance, IV < 0.02 indicates extremely low or no discriminative power for the feature, which can be discarded; 0.02 ≤ IV < 0.1 indicates weak discriminative power, generally used with caution; 0.1 ≤ IV < 0.3 indicates moderate discriminative power, commonly used in modeling; IV ≥ 0.3 indicates strong discriminative power, a very good feature; and IV ≥ 0.5 indicates extremely strong pre-discriminative power, but prone to overfitting. As another example, PSI < 0.1 indicates a very stable distribution with no significant changes. This feature is stable at the current stage. 0.1 ≤ PSI < 0.2 indicates a slight change in distribution, requiring attention, possibly an early signal of normal fluctuations or changes. PSI ≥ 0.2 indicates a significant change in distribution, requiring high vigilance and consideration of parameter readjustment or retraining. A PSI ≥ 0.3 indicates a drastic change in distribution, requiring rebinding. Therefore, if the evaluation metrics are IV and PSI, the corresponding thresholds can be 0.1 and 0.25, respectively. That is, features with IV > 0.1 and PSI < 0.25 are selected as the first screening features.

[0059] After obtaining the first set of screening features, a collinearity analysis is performed on the feature set. For example, by calculating statistics such as the correlation coefficient and variance inflation factor between features, it is determined whether there is a strong linear correlation between different features. If the linear relationship between two or more features is significant, it is recorded as collinearity; otherwise, it is recorded as non-collinearity, thus forming a collinearity judgment result regarding the collinearity status between each feature pair.

[0060] Based on the collinearity assessment results, the first set of features exhibiting significant linear correlation undergoes further filtering. For example, in a set of highly collinear features, only those with higher discriminative power, better stability, or clearer business meaning are retained, while the remaining collinear features are eliminated; features without collinearity are directly retained. After the above decollinearity removal process, a second set of features with higher information retention and lower redundancy is obtained.

[0061] After obtaining the second set of screening features, a logistic regression model can be constructed using sample data with known classification labels to estimate and test the regression coefficients and significance of each feature. Through stepwise regression, regularization constraints, and other methods, features with low statistical significance or small contribution are eliminated, and finally, features that have a significant impact on the model output and meet business requirements are retained as the target feature set.

[0062] It is understood that the embodiments of this application can eliminate invalid and redundant features, reduce feature dimensionality, and reduce the computational burden of the model by selecting evaluation indicators, judging collinearity, and optimizing the feature set step by step through logistic regression. At the same time, it reduces the interference of collinearity on parameter estimation, making the model parameters more stable and reliable, thereby improving the accuracy, robustness and interpretability of the customer group classification model, and improving the overall modeling efficiency and classification effect.

[0063] According to embodiments of this application, the multiple preset classification models include at least a first classification model and a second classification model. The multiple preset classification models are pre-constructed based on the following method: obtaining a training set including historical sample data; when a target feature is selected or an instruction to classify using a single rule is received, generating a single rule corresponding to the target feature based on the value distribution of the target feature in the training set and the corresponding classification label, and storing it in a rule base to construct the first classification model; when multiple target features are selected, using the training set, constructing a decision tree model with the multiple target features as input features to construct the second classification model.

[0064] In this embodiment, the training set is a collection of labeled historical sample data used to build and train a classification model. Each sample contains several feature values ​​and corresponding true classification labels. Specifically, historical customer data can be extracted from data sources such as business systems and historical transaction databases. The data is then cleaned, missing values ​​are handled, and outliers are removed. Based on existing business judgment results (such as risk level, credit granting results, and whether a default has occurred), each sample is labeled with a classification label, thereby forming a training dataset that can be directly used for modeling.

[0065] Selecting a target feature can refer to a modeler or business user manually choosing a specific feature from a feature list. The instruction to use single-rule classification specifies the need to construct a simple and interpretable classification rule based on a single feature. Upon selecting a target feature or receiving an instruction to use single-rule classification, the distribution of various labels across different value ranges (or different categories) of the target feature can be statistically analyzed based on the training set, such as calculating the proportion of "high-risk samples" within a certain value range. Then, based on pre-defined criteria (such as risk percentage thresholds, precision or recall requirements), one or a small number of judgment conditions that meet the conditions are automatically searched or enumerated, for example, "if the target feature is greater than a certain threshold, it is judged as high-risk; otherwise, it is judged as low-risk." The selected judgment conditions are solidified into formal rule expressions, serving as the single rule corresponding to the target feature. The rule base stores classification rules, which may include rule number, applicable features, threshold or range settings, output category, and effective conditions. The first classification model is a single-rule classification model built based on a single target feature, suitable for scenarios requiring rapid judgment or where transparent rules are emphasized in business operations.

[0066] When multiple target features are selected, the feature vectors corresponding to these features in the training set can be used as model input, and the classification labels of the samples can be used as output. Then, a decision tree algorithm (such as one based on information gain, Gini coefficient, or other splitting criteria) is used to recursively select the optimal splitting feature and its splitting threshold, starting from the root node, to progressively divide the sample space into several child nodes until a stopping condition is met (such as maximum depth, minimum number of samples, or purity requirements). Finally, a tree structure consisting of internal nodes, splitting conditions, and leaf nodes is obtained. This decision tree model is a secondary classification model that can simultaneously utilize multiple target features for joint judgment.

[0067] Specifically, taking the classification of sample data into first sample data and second sample data as an example, a decision tree can be constructed through the following steps:

[0068] Step 1: Calculate the Gini coefficient of the root node using formula (1).

[0069] Formula (1)

[0070] in, This indicates the number of all first sample data in the current node. This indicates the number of all second sample data in the current node. This represents the total number of sample data in the current node.

[0071] Step 2: Select one of the features Sort the values ​​in ascending order and use the average of two adjacent values ​​as a backup split point.

[0072] Step 3: Select a dividing point in b. Split the current node, less than or equal to The left child node is greater than 0. The node is the right child node.

[0073] Step 4: Calculate the Gini coefficients of the left and right nodes respectively according to formula (1), and denot them as follows. , .

[0074] Step 5: Calculate the Gini gradient descent amount, as shown in formula (2).

[0075] Formula (2)

[0076] in, This represents the gradient descent amount of the Gini coefficient. This represents the Gini coefficient of the node to be segmented. This represents the Gini coefficient of the left node. This represents the Gini coefficient of the right node. This indicates the amount of sample data included in the left node. This indicates the amount of sample data included in the right node. This indicates the amount of sample data for the node to be segmented.

[0077] Step 6: Repeat steps 3-5 to extract the features. All dividing points Calculate and select Maximum This serves as the segmentation point for this feature.

[0078] Step 7: Repeat steps 2-6 to calculate other features. ,Pick The largest feature is used as the node for actual segmentation. corresponding As the actual dividing point, left and right child nodes are divided to construct the decision tree model.

[0079] It's important to note that the core objective of the decision tree model is to maximize the purity of its child nodes, meaning that the nodes contain as many samples as possible of the same type. The higher the proportion of samples of the same type within a node (i.e., the higher the purity), the closer the Gini coefficient will be to 0. Furthermore, the higher the purity of the left and right child nodes, the greater the gradient descent value of the Gini coefficient. Therefore, the optimal splitting features and points can be found based on these segmentation features.

[0080] Figure 3 A schematic diagram of a decision tree model according to an embodiment of this application is shown.

[0081] like Figure 3 As shown, at the root node 301, the Gini coefficient is 0.268. The number of institutions that have credit transactions is selected as the split point for segmenting the sample data through steps 3-6. The average value of this feature is 5.5. The number of sample data corresponding to the root node is 3200, of which the number of first sample data is 2689 and the number of second sample data is 511.

[0082] Two child nodes are branched from the root node: node 302 and node 303. Node 302 corresponds to sample data where the number of institutions with credit transactions is greater than 5.5, and node 303 corresponds to sample data where the number of institutions with credit transactions is less than or equal to 5.5. For node 302, the Gini coefficient is 0.183, including 1846 first sample data and 209 second sample data. Through steps 3-6, the cutoff point is determined to be that the total number of liability accounts is less than or equal to 24.0. Further processing yields the following results: Figure 3 The decision tree model shown.

[0083] It is understood that, in the embodiments of this application, by pre-constructing a first classification model and a second classification model based on the training set and providing them as multiple preset classification models for selection, efficient and interpretable fast classification can be achieved directly using a single rule in simple scenarios, or decision trees can be used to mine the joint effect of multiple features in complex scenarios to improve classification accuracy and coverage, and enhance the adaptability of the customer group classification method to different business needs and data characteristics.

[0084] According to an embodiment of this application, the method further includes: training a first decision tree model based on all training data in the training set, using multiple target features as input features; dividing the training set based on user type to obtain multiple sub-training sets, wherein each sub-training set corresponds to a user type; and training a second decision tree model using the multiple sub-training sets, using multiple target features as input features.

[0085] In this embodiment, a first decision tree model can be trained based on all training data in the training set, using multiple target features as input features. Specifically, the values ​​of each historical sample on multiple target features are used to form a feature vector, and the corresponding classification label of the sample is used as the output. The optimal splitting feature and splitting threshold are recursively selected in the overall sample space using a decision tree algorithm to generate a first decision tree model covering all user types.

[0086] Alternatively, the training set can be divided into multiple sub-training sets based on pre-labeled or identified user type information (e.g., individual users, enterprise users, or different customer segmentation). Each sub-training set contains only training data for the same user type. For each sub-training set, a corresponding second decision tree model can be independently trained using the same set or a set of homogeneous target features as input features.

[0087] It is understood that the embodiments of this application construct a first decision tree model on all training data to learn global commonalities, divide the training set according to user type and train multiple second decision tree models respectively to explore intra-type differences, so that the corresponding classification model can be selected according to needs. For example, for users of unknown type or when the overall data distribution needs to be considered, the first decision tree model can be selected, and for users of known type or when the classification accuracy requirement is high, the second decision tree model can be selected. This embodiment is conducive to refined management and gives business personnel a certain degree of autonomy.

[0088] According to an embodiment of this application, at least one evaluation index of a user feature is determined based on the following method: obtaining dynamic binning information pre-associated with the user feature and preset statistical parameters corresponding to each bin, wherein the dynamic binning information represents multiple binning boundaries adaptively determined under the condition of satisfying constraints; and calculating at least one evaluation index for each user feature based on the dynamic binning information and the preset statistical parameters.

[0089] In this embodiment, the pre-associated dynamic binning information refers to a set of binning boundary information automatically determined by an adaptive binning algorithm for each user feature, based on its values ​​in historical samples and under preset constraints. Specifically, for continuous features, the value range can be divided into several sub-intervals according to sample distribution, bad sample rate trends, etc., with each sub-interval corresponding to a bin; for discrete features, multiple categories can be merged according to statistical characteristics to form several bins. The dynamic binning information can take the form of binning boundary values ​​or a set of categories + binning sequence number, which is associated with and stored one-to-one with the corresponding user feature.

[0090] The constraints are a set of rules that the dynamic binning algorithm must satisfy during adaptive binning to ensure the stability and statistical validity of the binning results. Constraints may include at least one of the following: the number of samples in a single bin is not less than a certain minimum threshold; the sample rate of each category changes monotonically or substantially monotonically with the bin number; the total number of bins does not exceed a preset upper limit; and adjacent bins have statistically significant differences. By introducing these constraints into the binning process, binning structures with too few samples, excessive fluctuations, or no real discrimination can be avoided.

[0091] The preset statistical parameters corresponding to each bin can refer to a set of statistics predefined and calculated for each bin obtained by dynamic binning, used to characterize the distribution characteristics of samples within that bin under different categories. For example, these statistical parameters may include: total number of samples in the bin, number of positive samples, number of negative samples, positive and negative sample ratio, default rate or non-performing rate, weighted evidence (WOE) value, bin information value contribution, etc.

[0092] Specifically, the historical sample data is first preprocessed. In order to reduce the influence of the magnitude of different feature values, continuous data can be normalized to the 0-1 interval, while discrete data is not normalized. Then, 20% of the sample data is randomly selected as the test set, and the remaining 80% of the sample data is used as the training set.

[0093] If equal-frequency binning is used, the continuous data in the training set can be binned into five equal-frequency bins by default, and each feature value of the discrete data will be the corresponding bin. Null values ​​in both continuous and discrete data will be treated as separate bins.

[0094] If dynamic binning is used, finer initial bins (such as equidistant or high-frequency subdivision) can be set for continuous features, while discrete features are initially binned using each value or a combination of merged values, with null values ​​remaining in separate bins. Based on this, according to preset constraints (such as minimum sample size, maximum number of bins, monotonicity, etc.) and optimization criteria (such as chi-square statistics, information gain, IV boosting, etc.), adjacent or statistically similar bins are iteratively merged until the constraints are met and convergence to a reasonable number of bins. Furthermore, the obtained bin boundaries, bin numbers, and the number of samples corresponding to each bin can be used as dynamic binning information to calculate feature evaluation metrics.

[0095] For example, in a binary classification application scenario, after binning, the WOE value of each bin is calculated according to the following formula (3):

[0096] Formula (3)

[0097] in, This represents the proportion of the first sample in the i-th bin to all first samples in the training set. This represents the proportion of the second sample in the i-th bin to all second samples in the training set.

[0098] To compare the impact of different features on the target recognition, it is necessary to calculate the IV value of each feature, and the calculation formula is shown in formula (4):

[0099] Formula (4)

[0100] The PSI calculation formula for each feature is shown in formula (5):

[0101] Formula (5)

[0102] in, To ensure the fraction is extremely small, avoid denominators of 0 and cases where the exponent is 0 when calculating the logarithm. This represents the proportion of the second sample data in the i-th bin of the test set to the total number of samples in the test set. This represents the proportion of the second sample data in the i-th bin of the training set to the total sample data in the training set.

[0103] It is understood that the embodiments of this application, through the dynamic binning strategy, can display the relationship between features and classification labels more precisely and stably. The evaluation indicators calculated on this basis can more accurately reflect the distinguishing ability and stability of features, thereby achieving objective quantitative evaluation and effective screening of user features and improving the accuracy of classification.

[0104] According to an embodiment of this application, when the target rule includes a combination rule, classifying multiple users in the customer group information using the target rule to obtain the classification result includes: filtering the combination rules based on their respective evaluation values ​​to determine the target rule; classifying the multiple users included in the customer group using the target rule to obtain the classification result; wherein, the evaluation values ​​of each combination rule are obtained based on the following method: testing at least one combination rule to determine the effectiveness index of the combination rule, the effectiveness index including at least recall and precision; calculating the evaluation value of each combination rule using the weights of each effectiveness index and a pre-constructed weighting function, wherein the weights are used to characterize the preference for different effectiveness indexes.

[0105] In this embodiment, when the target rule includes combined rules, a pre-generated set of combined rules can be tested to evaluate each combined rule. Specifically, by executing the combined rules on a sample set with real labels, the correspondence between the hit results and the actual labels is statistically analyzed, and the effectiveness index of each combined rule is calculated, which includes at least recall and precision. Then, by combining pre-set index weights and a constructed weighting function, the effectiveness indices such as recall and precision are weighted and summarized according to their weights to obtain the evaluation value corresponding to each combined rule. The higher the evaluation value, the better the overall performance of the combined rule under the current preference setting, thus intuitively reflecting the ranking of the advantages and disadvantages of each combined rule.

[0106] After obtaining the evaluation values ​​of each combination rule, the combination rules can be filtered based on the evaluation values, eliminating combination rules with low evaluation values ​​or those that do not meet the threshold requirements, and retaining only one or more rules with the highest evaluation values ​​as target rules.

[0107] Specifically, the formula for calculating precision is shown in formula (6).

[0108] Formula (6)

[0109] Wherein, TP represents the number of data points that are classified as first samples and are actually first samples, i.e., the total number of correct hits, and FP represents the number of data points that are classified as first samples and are actually second samples, i.e., the number of false alarms.

[0110] The formula for calculating recall is shown in formula (7).

[0111] Formula (7)

[0112] Where FN represents the number of sample data that are classified as second sample data and are actually second sample data, i.e., the number of correctly rejected samples.

[0113] Understandably, by introducing effectiveness metrics such as recall and precision, and combining them with adjustable weights and weighting functions to quantitatively evaluate and filter combined rules, the target rule can be automatically selected from multiple combined rules. This avoids relying on subjective rule selection based on human experience and allows for flexible trade-offs between recall and precision, thereby improving the accuracy and stability of customer segmentation.

[0114] According to an embodiment of this application, the method further includes: in response to receiving a task to update a preset classification model, acquiring new sample data; based on the new sample data, determining the volatility of a single rule or combined rule effectiveness index; and in response to the volatility being greater than a preset threshold, updating the preset classification model using the new sample data and historical sample data.

[0115] In this embodiment, a scheduled task can be set, such as retrieving the latest data from the previous month every month and using this data to update the preset classification model. When a task instruction to update the preset classification model is received, newly generated business records within a certain period can be extracted from the online business system or data warehouse to form new sample data. The new sample data carries user characteristics and true classification labels with the same structure as historical samples, reflecting the latest user behavior and environmental changes. Then, based on the new sample data, the effectiveness indicators of the single rules or combined rules involved in the current preset classification model are recalculated, such as precision, recall, and hit rate, and compared with the indicator values ​​of the rule on historical sample data to obtain the volatility of each effectiveness indicator. Specifically, the volatility of the effectiveness indicators of commonly used rules can be calculated. For example, when using target rules for customer group classification, the number of times each rule is used can be recorded, and commonly used rules can be determined based on this number of times.

[0116] After obtaining the volatility of the evaluation metrics, the volatility of each rule can be compared with a preset threshold. When the volatility of the effectiveness metric for a certain rule or a class of rules exceeds the threshold, it means that the rule's performance on new data differs significantly from its historical performance, potentially indicating a risk of rule failure. In this case, a model update process can be triggered, using both new sample data and existing historical sample data as training data to update the preset classification model. This can involve retraining the decision tree model, regenerating or adjusting single / combined rules, reestimating bins and parameters, etc. After successful validation, the original model is replaced, ensuring that the preset classification model can promptly absorb information from the latest data.

[0117] It should be noted that, in order to ensure the effectiveness of the model, the model can be updated only if the time between the last update and the current time exceeds a time threshold. For example, if the model has not been updated for more than six months, the task of updating the model can be executed automatically.

[0118] It is understood that the embodiments of this application introduce a volatility monitoring mechanism based on the effectiveness index of newly added sample data, and trigger model updates when the volatility exceeds a preset threshold. This can automatically identify situations where the performance of the preset classification model has significantly degraded in the actual environment, and promptly update the model using a combination of new and old samples, thereby effectively responding to changes in data distribution and user behavior and maintaining the accuracy of the classification model.

[0119] Figure 4 A flowchart illustrating the determination of target rules according to an embodiment of this application is shown, such as... Figure 4 As shown, the method for determining the target rule in this embodiment includes operations S401 to S417.

[0120] Using S401, customer information is obtained.

[0121] Specifically, this can involve reading basic information and related characteristic data of multiple users in the customer group to be analyzed from the business system or data warehouse.

[0122] In operation S402, determine whether it is necessary to select sample data based on user type.

[0123] Depending on configuration or business requirements, determine whether to uniformly model all users included in the sample data, or to select samples separately according to different user types such as individual users and enterprise users. If it is not necessary to select sample data according to user type, execute operation S403; if it is necessary to select sample data according to user type, execute operation S404.

[0124] In operation S403, retrieve all historical sample data.

[0125] When there is no need to distinguish user types, all historical sample data can be obtained, that is, the full amount or sample records covering all user types within a preset time range can be extracted from the historical database.

[0126] In operation S404, sample data within a specified range is obtained.

[0127] When it is necessary to differentiate by user type or other conditions, sample data within a specified range can be obtained, such as extracting historical samples that meet the requirements based on user type, time interval, product type, or other limiting conditions.

[0128] In operation S405, the features involved in the sample data are filtered.

[0129] Specifically, the features involved in the sample data are screened. Based on indicators such as missing rate, stability, discriminative power and business relevance, inapplicable features are removed from the original feature set to form a candidate feature set.

[0130] In operation S406, determine whether to perform custom processing on the features.

[0131] Determine whether the user or expert needs to make further manual intervention and adjustments to the features based on the automatic filtering results. If no custom processing of the features is required, proceed to operation S407; if custom processing of the features is required, proceed to operation S408.

[0132] In operation S407, select all filtered features.

[0133] When no custom processing is required, all features obtained through automatic filtering can be used directly as input features for rule calculation.

[0134] In operation S408, the filtered features are added or deleted.

[0135] When custom processing is required, users can manually add custom features or delete features they do not want to participate in the modeling, resulting in a manually adjusted feature set.

[0136] When operating S409, feature recommendation is performed.

[0137] Based on statistical analysis or experimental modeling results, the currently available features can be evaluated, and a subset of features that contribute significantly to the target classification performance can be automatically recommended.

[0138] In operation S410, determine whether to select all recommended features.

[0139] Based on user settings or system policies, it can be determined whether to directly adopt all feature recommendation results. If it is not necessary to select all recommended features, execute operation S411; if it is necessary to select all recommended features, execute operation S412.

[0140] In operation S411, select all recommended features.

[0141] When operating S412, customize the recommended features.

[0142] This operation allows users to select or delete certain features as needed, forming a custom feature subset.

[0143] In operation S413, determine whether to customize the selection rules.

[0144] If no custom selection rules are required, operation S414 can be executed; if custom selection rules are required, operation S415 can be executed.

[0145] In operation S414, the effectiveness indices of single rules and combined rules are calculated respectively, and the target rule is determined.

[0146] Without limiting the rule type, multiple single rules and combined rules can be generated and evaluated, and the target rule with the best overall performance can be selected based on the effectiveness index.

[0147] In operation S415, determine whether to select the combined rule as the target rule.

[0148] If it is not necessary to select a combined rule as the target rule, then execute operation S416; if it is necessary to select a combined rule as the target rule, then execute operation S417.

[0149] In operation S416, the validity index of each individual rule is calculated, and the target rule is determined.

[0150] If only single rules are used, the effectiveness index of each single rule is calculated, the target rule is determined, and the optimal single rule is selected as the target rule.

[0151] In operation S417, the effectiveness index of the combined rules is calculated and the target rule is determined.

[0152] When choosing to use combined rules, calculate the effectiveness index of each combined rule, determine the target rule, and select the optimal combined rule as the target rule.

[0153] Figure 5 A flowchart illustrating the determination of a target feature set according to an embodiment of this application is shown, such as... Figure 5 As shown, the method for determining the target feature set in this embodiment includes operations S501 to S510.

[0154] In operation of S501, user characteristics are obtained.

[0155] For example, read multi-dimensional feature data related to target users from business systems or data sources, such as behavioral features, credit features, product usage features, or process customer information to obtain user features.

[0156] In operation S502, user characteristics are preprocessed.

[0157] For example, operations such as missing value handling, outlier correction, standardization or normalization, and type conversion can be performed on the original features.

[0158] When operating S503, binning is performed based on user characteristics.

[0159] For continuous data, it can be divided into several intervals according to preset binning rules (such as equal frequency binning, equal interval binning, or dynamic binning), and discrete features can be categorized and merged or independently binned.

[0160] In operation S504, the evaluation index for each feature is calculated.

[0161] For example, quantitative indicators such as information value, discrimination, and stability index.

[0162] When operating S505, determine whether the evaluation metrics for each feature meet the requirements.

[0163] If the evaluation indicators meet the requirements, execute operation S506; if the evaluation indicators do not meet the requirements, execute operation S507.

[0164] Specifically, the evaluation indicators of each feature can be compared with preset thresholds or screening criteria to determine whether the feature meets the conditions for constructing the target feature set.

[0165] When operating S506, retain the features.

[0166] For features that meet the evaluation criteria, perform the feature retention operation.

[0167] When operating S507, discard features.

[0168] For features that do not meet the evaluation criteria, discard the features.

[0169] In operation S508, determine whether the feature is a discrete feature.

[0170] If the feature is discrete, execute operation S509; if the feature is continuous, execute operation S510.

[0171] In operation S509, discrete features are converted into dummy variables.

[0172] For example, one-hot encoding can be used to map different categories to multiple binary indicator variables.

[0173] In operation S510, the retained features are merged and used as the target feature set.

[0174] All retained features are aggregated to obtain the target feature set used to build the classification model and generate rules.

[0175] Figure 6 A schematic block diagram of a customer group classification device according to an embodiment of this application is shown.

[0176] like Figure 6As shown, the customer group classification device 600 of this embodiment includes a determination module 610, a first processing module 620, a second processing module 630, and a classification module 640.

[0177] The determination module 610 is used to acquire customer group information and, based on multiple user characteristics included in the customer group information, determine at least one evaluation index for each user characteristic. The evaluation index is used to represent the discriminative power of the user characteristic relative to sample classification and / or the distribution stability across different sample sets. In one embodiment, the determination module 610 can be used to perform the operation S210 described above, which will not be repeated here.

[0178] The first processing module 620 is used to filter user features based on evaluation metrics to determine a target feature set. In one embodiment, the first processing module 620 can be used to perform the operation S220 described above, which will not be repeated here.

[0179] The second processing module 630 is used to determine a corresponding target classification model from multiple preset classification models based on the target feature set, and to process the target feature set using the target classification model to obtain target rules, wherein the target rules include single rules or combined rules. In one embodiment, the second processing module 630 can be used to perform the operation S230 described above, which will not be repeated here.

[0180] The classification module 640 is used to classify multiple users in the customer group information using target rules to obtain classification results. In one embodiment, the classification module 640 can be used to perform the operation S240 described above, which will not be repeated here.

[0181] According to an embodiment of this application, the first processing module 620 includes a first filtering module, a judgment module, a second filtering module, and a logistic regression processing module. The first filtering module is used to filter user features based on evaluation indicators and corresponding indicator thresholds to obtain first filtered features. The judgment module is used to perform collinearity judgment on the first filtered features to obtain a collinearity judgment result, which includes whether there is a linear relationship between the first filtered features. The second filtering module filters the first filtered features based on the collinearity judgment result to obtain second filtered features. The logistic regression processing module performs logistic regression processing on the second filtered features to obtain a target feature set.

[0182] According to embodiments of this application, the multiple preset classification models include at least a first classification model and a second classification model. These multiple preset classification models are pre-constructed based on a first acquisition module, a third processing module, and a fourth processing module. The first acquisition module is used to acquire a training set including historical sample data. The third processing module, upon selecting a target feature or receiving an instruction to classify using a single rule, generates a single rule corresponding to the target feature based on the value distribution and corresponding classification label of the target feature in the training set and stores it in a rule base to construct the first classification model. The fourth processing module, upon selecting multiple target features, uses the training set and multiple target features as input features to construct a decision tree model to construct the second classification model.

[0183] According to an embodiment of this application, the customer group classification device 600 further includes a first training module, a partitioning module, and a second training module: the first training module is used to train a first decision tree model based on all training data in the training set, using multiple target features as input features; the partitioning module is used to partition the training set based on user type to obtain multiple sub-training sets, wherein each sub-training set corresponds to a user type; the second training module is used to train a second decision tree model using multiple sub-training sets, using multiple target features as input features.

[0184] According to an embodiment of this application, at least one evaluation index of a user feature is determined based on a second acquisition module and a first calculation module: the second acquisition module is used to acquire dynamic binning information pre-associated with the user feature and preset statistical parameters corresponding to each bin, wherein the dynamic binning information represents multiple binning boundaries adaptively determined under the condition of satisfying constraints; the first calculation module is used to calculate at least one evaluation index for each user feature based on the dynamic binning information and the preset statistical parameters.

[0185] According to an embodiment of this application, when the target rule includes a combination rule, the classification module 640 includes a target rule determination module and a classification submodule. The target rule determination module is used to filter the combination rules based on their respective evaluation values ​​to determine the target rule; the classification submodule is used to classify the multiple users included in the customer group using the target rule to obtain the classification result; wherein, the evaluation values ​​of each combination rule are obtained based on a fifth processing module and a second calculation module, the fifth processing module is used to test at least one combination rule to determine the effectiveness index of the combination rule, the effectiveness index including at least recall and precision; the second calculation module is used to calculate the evaluation value of each combination rule using the weights of each effectiveness index and a pre-constructed weighting function, wherein the weights are used to characterize the preference for different effectiveness indexes.

[0186] According to an embodiment of this application, the customer group classification device 600 further includes a third acquisition module, a volatility determination module, and an update module. The third acquisition module is used to acquire new sample data in response to receiving a task to update the preset classification model. The volatility determination module is used to determine the volatility of a single rule or a combination of rule validity indicators based on the new sample data. The update module is used to update the preset classification model using the new sample data and historical sample data in response to the volatility being greater than a preset threshold.

[0187] According to embodiments of this application, any multiple modules among the determining module 610, the first processing module 620, the second processing module 630, and the classification module 640 can be combined into one module, or any one of these modules can be split into multiple modules. Alternatively, at least some of the functions of one or more of these modules can be combined with at least some of the functions of other modules and implemented in one module. According to embodiments of this application, at least one of the determining module 610, the first processing module 620, the second processing module 630, and the classification module 640 can be at least partially implemented as hardware circuitry, such as a field-programmable gate array (FPGA), a programmable logic array (PLA), a system-on-a-chip, a system-on-a-substrate, a system-on-package, an application-specific integrated circuit (ASIC), or implemented in hardware or firmware by any other reasonable means of integrating or packaging circuitry, or implemented in software, hardware, or firmware, or in any suitable combination of any of these three implementation methods. Alternatively, at least one of the determining module 610, the first processing module 620, the second processing module 630, and the classification module 640 may be at least partially implemented as a computer program module, which can perform corresponding functions when the computer program module is run.

[0188] Figure 7 A block diagram schematically illustrates an electronic device suitable for implementing a customer group classification method according to an embodiment of this application.

[0189] like Figure 7 As shown, an electronic device 700 according to an embodiment of this application includes a processor 701, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 702 or a program loaded from a storage portion 708 into a random access memory (RAM) 703. The processor 701 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 701 may also include onboard memory for caching purposes. The processor 701 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of this application.

[0190] RAM 703 stores various programs and data required for the operation of electronic device 700. Processor 701, ROM 702, and RAM 703 are interconnected via bus 704. Processor 701 executes various operations of the method flow according to embodiments of this application by executing programs in ROM 702 and / or RAM 703. It should be noted that the programs may also be stored in one or more memories other than ROM 702 and RAM 703. Processor 701 may also execute various operations of the method flow according to embodiments of this application by executing programs stored in said one or more memories.

[0191] According to embodiments of this application, the electronic device 700 may further include an input / output (I / O) interface 705, which is also connected to a bus 704. The electronic device 700 may also include one or more of the following components connected to the input / output (I / O) interface 705: an input section 706 including a keyboard, mouse, etc.; an output section 707 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 708 including a hard disk, etc.; and a communication section 709 including a network interface card such as a LAN card, modem, etc. The communication section 709 performs communication processing via a network such as the Internet. A drive 710 is also connected to the input / output (I / O) interface 705 as needed. A removable medium 711, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 710 as needed so that computer programs read from it can be installed into the storage section 708 as needed.

[0192] This application also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs, which, when executed, implement the method according to the embodiments of this application.

[0193] According to embodiments of this application, the computer-readable storage medium can be a non-volatile computer-readable storage medium, such as including but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to embodiments of this application, the computer-readable storage medium may include ROM 702 and / or RAM 703 and / or one or more memories other than ROM 702 and RAM 703 described above.

[0194] Embodiments of this application also include a computer program product comprising a computer program containing program code for performing the methods shown in the flowchart. When the computer program product is run on a computer system, the program code is used to enable the computer system to implement the customer classification method provided in the embodiments of this application.

[0195] When the computer program is executed by the processor 701, it performs the functions defined in the system / apparatus of this application embodiment. According to the embodiments of this application, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0196] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and may be downloaded and installed via the communication section 709, and / or installed from a removable medium 711. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0197] In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 709, and / or installed from the removable medium 711. When the computer program is executed by the processor 701, it performs the functions defined in the system of this application embodiment. According to the embodiments of this application, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0198] According to embodiments of this application, program code for executing the computer programs provided in the embodiments of this application can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C", or similar programming languages. The program code can be executed entirely on the user's computing device, partially on the user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0199] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0200] Those skilled in the art will understand that the features described in the various embodiments of this application can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in this application. In particular, the features described in the various embodiments of this application can be combined and / or combined in various ways without departing from the spirit and teachings of this application. All such combinations and / or combinations fall within the scope of this application.

Claims

1. A customer group classification method, characterized in that, The method comprises: obtaining customer group information, determining at least one evaluation index of each user feature based on a plurality of user features included in the customer group information, the evaluation index being used to represent the distinguishing ability of the user feature relative to sample classification and / or the distribution stability on different sample sets; screening the user features based on the evaluation indexes to determine a target feature set; determining a corresponding target classification model from a plurality of preset classification models based on the target feature set, processing the target feature set by using the target classification model to obtain a target rule, wherein the target rule includes a single rule or a combined rule; classifying a plurality of users in the customer group information by using the target rule to obtain a classification result.

2. The method of claim 1, wherein, The screening of the user features based on the evaluation indexes to determine a target feature set comprises: screening the user features based on the evaluation indexes and index thresholds corresponding to the evaluation indexes to obtain first screening features; performing collinearity judgment on the first screening features to obtain collinearity judgment results, the collinearity judgment results including whether there is a linear relationship between the first screening features; screening the first screening features based on the collinearity judgment results to obtain second screening features; performing logistic regression processing on the second screening features to obtain a target feature set.

3. The method of claim 1, wherein, The plurality of preset classification models at least includes a first classification model and a second classification model, and the plurality of preset classification models are pre-constructed based on the following method: obtaining a training set including historical sample data; in the case of selecting a target feature or receiving an instruction for classifying by using a single rule, generating a single rule corresponding to the target feature based on the value distribution of the target feature in the training set and the corresponding classification label, and storing the single rule in a rule library to construct a first classification model; in the case of selecting a plurality of target features, using the training set to construct a decision tree model with the plurality of target features as input features to construct a second classification model.

4. The method of claim 3, wherein, The method further comprises: training a first decision tree model with the plurality of target features as input features based on all training data in the training set; dividing the training set based on user types to obtain a plurality of sub-training sets, wherein each sub-training set corresponds to one user type; training a second decision tree model with the plurality of target features as input features based on the plurality of sub-training sets respectively.

5. The method of claim 1, wherein, The at least one evaluation index of the user feature is determined based on the following method: obtaining dynamic binning information pre-associated with the user feature and preset statistical parameters corresponding to each bin, the dynamic binning information representing a plurality of bin boundaries adaptively determined under the condition of satisfying a constraint condition; calculating at least one evaluation index of each user feature based on the dynamic binning information and the preset statistical parameters.

6. The method of claim 1, wherein, In the case that the target rule includes a combined rule, the classification of a plurality of users in the customer group information by using the target rule to obtain a classification result comprises: screening the combined rules based on respective evaluation values of the combined rules to determine a target rule; Classifying a plurality of users included in the customer group by using the target rule to obtain a classification result; wherein the evaluation value of each of the combined rules is obtained based on the following method: Testing the at least one combined rule to determine an effectiveness index of the combined rule, the effectiveness index at least including recall and precision; Using the respective weight of the effectiveness index and a pre-constructed weighting function to calculate the evaluation value of each of the combined rules, wherein the weight is used to represent the preference degree for different effectiveness indexes.

7. The method of claim 3, wherein, The method further comprises: In response to receiving a task of updating the preset classification model, obtaining new sample data; Based on the new sample data, determining the volatility of the single rule or combined rule effectiveness index; In response to the volatility being greater than a preset threshold, updating the preset classification model by using the new sample data and historical sample data.

8. A customer classification device, characterized by comprising: The device comprises: A determination module configured to obtain customer group information, determine at least one evaluation index of each user feature based on a plurality of user features included in the customer group information, and use the evaluation index to represent the distinguishing ability of the user feature with respect to sample classification and / or the distribution stability on different sample sets; A first processing module configured to filter the user features based on the evaluation index and determine a target feature set; A second processing module configured to determine a corresponding target classification model from a plurality of preset classification models based on the target feature set, process the target feature set by using the target classification model, and obtain a target rule, wherein the target rule includes a single rule or a combined rule; A classification module configured to classify a plurality of users in the customer group information by using the target rule to obtain a classification result. 9.An electronic device comprising: one or more processors; a memory for storing one or more computer programs, characterized in that the one or more processors execute the one or more computer programs to implement the steps of the method according to any one of claims 1-7.

10. A computer readable storage medium having stored thereon a computer program or instructions, characterized in that, The computer program or instructions are executed by the processor to implement the steps of the method according to any one of claims 1-7.

11. A computer program product comprising computer programs or instructions, characterized in that, The computer program or instructions are executed by the processor to implement the steps of the method according to any one of claims 1-7.

Images