A method, a classification method, a device and a device for establishing an insurance user classification model

Abstract

The present invention provides a method, a classification method, a device and a device for establishing an insurance user classification model. The method includes: obtaining the user insurance data and user attributes of historical insurance users, and the regional insurance data of the regions to which the historical insurance users belong; determining the type characteristics of different data in the user insurance data and the regional insurance data; performing binning encoding processing on the continuous data in the user insurance data and the regional insurance data to obtain the converted continuous data, and integrating the categorical data, the encoded continuous data in the user insurance data and the regional insurance data, and the user attributes into training set data; using the training set data to train a preset initial classification model to obtain a trained insurance user classification model. Through the insurance user classification model, the accuracy of classifying existing insurance users is improved, and thus reliable guidance is provided for insurance services.

Description

Technical Field

[0001] This article belongs to the technical field of insurance, and specifically relates to a method for establishing an insurance user classification model, a classification method, a device, and a device. Background Art

[0002] In the process of insurance marketing, existing customers whose policies are about to expire are generally selected for marketing. The current marketing strategy is to simply screen out a group of customers based on age and purchased products or randomly select customers for message delivery. This marketing effect has a poor recall rate, and recalling customers one by one incurs certain operating costs. Insurance repurchase prediction is an important task for the insurance industry's management department. Therefore, accurately classifying users helps insurance companies better understand customer demands and behavioral characteristics, conduct targeted marketing for existing customers, and accurately recommend products, which can improve the secondary conversion efficiency of customers, solve the problem of enterprise customer growth, and meet customers' insurance needs. Therefore, how to improve the accuracy of insurance user classification and reduce the recall cost has become a technical problem that urgently needs to be solved. Summary of the Invention

[0003] Aiming at the above problems of the prior art, the purpose of this article is to provide a method for establishing an insurance user classification model, a classification method, a device, and a device, which can improve the classification accuracy of insurance users.

[0004] To solve the above technical problems, the specific technical solutions of this article are as follows:

[0005] On the one hand, this article provides a method for establishing an insurance user classification model, and the method includes:

[0006] Obtain the user insurance data and user attributes of historical insurance users, as well as the regional insurance data of the regions where the historical insurance users are located. The user attributes include existing users who have repurchased insurance and those who have not.

[0007] Determine the type characteristics of different data in the user insurance data and the regional insurance data. The type characteristics include categorical data and continuous data.

[0008] Perform binning encoding processing on the continuous data in the user insurance data and the regional insurance data to obtain the converted continuous data, and integrate the categorical data, the encoded continuous data, and the user attributes in the user insurance data and the regional insurance data into training set data.

[0009] Use the training set data to train a preset initial classification model to obtain a trained insurance user classification model.

[0010] Further, the regional insurance data at least includes insurance density, insurance depth, and insurance coverage.

[0011] The insurance penetration rate is the ratio of the annual premium income of a region to the annual gross domestic product of the region;

[0012] The insurance density is the ratio of the annual premium income of a region to the annual total population of the region;

[0013] The insurance coverage rate is the ratio of the number of people who purchase insurance annually in a region to the annual total population of the region.

[0014] Further, the regional insurance data of the region to which the historical insurance users belong further includes:

[0015] According to the insurance density and insurance penetration rate of the region to which the historical insurance users belong, an insurance consumption coefficient of the region is calculated, and the insurance consumption coefficient represents the proportion of users in the region for insurance consumption;

[0016] According to the range interval where the insurance consumption coefficient is located and the preset correction rule corresponding to the range interval, the insurance consumption coefficient is corrected;

[0017] The corrected insurance consumption coefficient and the insurance coverage rate of the historical insurance users are used as the regional insurance data of the region to which the historical insurance users belong.

[0018] Further, the continuous data in the user insurance data and the regional insurance data is processed by bin coding to obtain the coded continuous data, including:

[0019] For each type of continuous data, the following steps are performed:

[0020] The sample values corresponding to the continuous data are sorted in ascending order to obtain a sample value queue;

[0021] The average value of two adjacent sample values in the sample value queue is used as a leaf division point to obtain a set of leaf division points;

[0022] The Gini coefficient when each leaf division point in the set of leaf division points is used as a binary classification point is calculated in turn, and the leaf division point with the smallest Gini coefficient is used as the optimal division point;

[0023] Using the optimal division point, the sample values corresponding to the continuous data are divided into two subsets, and the two subsets are coded according to a preset coding method to obtain the coded data.

[0024] Further, the process of binning and encoding the continuous data in the user insurance data and the regional insurance data to obtain the transformed continuous data, and integrating the categorical data, the encoded continuous data, and the user attributes in the user insurance data and the regional insurance data into training set data includes:

[0025] Regarding different data types in the user insurance data and the regional insurance data as different binary tree feature nodes;

[0026] Generating multiple binary trees based on the binary tree feature nodes, where each binary tree includes different combinations of binary tree feature node partitions;

[0027] Calculating the Gini coefficient of each binary tree in sequence according to the data corresponding to the binary tree feature nodes in each binary tree;

[0028] Taking the combination of binary tree feature node partitions corresponding to the binary tree with the minimum Gini coefficient as the target node combination, so that the binary tree generated according to the target node combination and the user attributes serve as the training set data of the preset initial classification model.

[0029] Further, the process of training the preset initial classification model with the training set data to obtain the trained insurance user classification model includes:

[0030] Determining the region to which the preset initial classification model to be trained belongs and the training set data corresponding to the region;

[0031] Inputting the training set data into the preset initial classification model and training it using the preset loss function to obtain the pre-trained first classification model;

[0032] Determining the gain scores of the user insurance data of different data types during the training process according to the pre-trained first classification model;

[0033] Sorting the gain scores of the user insurance data of different data types;

[0034] Taking the user insurance data of the preset type quantity as the target user insurance data combination in descending order of the sorting;

[0035] Extracting the target training set data from the training set data according to the target user insurance data combination;

[0036] Fine-tuning the first classification model with the target training set data to obtain the trained insurance user classification model.

[0037] On the other hand, this article also provides an insurance user classification model establishment device, and the device includes:

[0038] A data acquisition module, configured to acquire user insurance data and user attributes of historical insurance users, as well as regional insurance data of the regions to which the historical insurance users belong, where the user attributes include existing users who have repurchased insurance and those who have not repurchased insurance;

[0039] A type feature determination module, configured to determine the type features of different data in the user insurance data and the regional insurance data, where the type features include categorical data and continuous data;

[0040] A processing module, configured to perform binning encoding processing on the continuous data in the user insurance data and the regional insurance data to obtain the converted continuous data, and integrate the categorical data, the encoded continuous data in the user insurance data and the regional insurance data, and the user attributes into training set data;

[0041] A training module, configured to train a preset initial classification model using the training set data to obtain a trained insurance user classification model.

[0042] On the other hand, this article also provides an insurance user classification method, and the method includes:

[0043] Acquire user insurance data of the user to be classified and regional insurance data of the region to which the user to be classified belongs;

[0044] Substitute the user insurance data and the regional insurance data into the insurance user classification model established by the method described above to obtain the user attributes of the user to be classified.

[0045] On the other hand, this article also provides an insurance user classification device, and the device includes:

[0046] A user data acquisition module to be classified, configured to acquire user insurance data of the user to be classified and regional insurance data of the region to which the user to be classified belongs;

[0047] A classification module, configured to substitute the user insurance data and the regional insurance data into the insurance user classification model established by the method described above to obtain the user attributes of the user to be classified.

[0048] On the other hand, this article also provides a computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor implements the method described above when executing the computer program.

[0049] Finally, this article also provides a computer-readable storage medium, where the computer-readable storage medium stores a computer program, and the computer program implements the method described above when executed by a processor.

[0050] With the above technical solution, a method, a classification method, a device and a device for establishing an insurance user classification model described in this article obtain the user insurance data and regional insurance data of historical insurance users, and perform binning transcoding processing on the continuous data in the user insurance data and the regional insurance data to obtain training set data for training, where user attributes are used as the target output for training, so as to obtain an insurance user classification model for classifying users, improving the classification accuracy of existing insurance users, and thus providing reliable guidance for insurance services.

[0051] To make the above and other purposes, features and advantages of this article more obvious and understandable, the following specifically gives preferred embodiments and, in conjunction with the accompanying drawings, makes a detailed description as follows. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] To more clearly illustrate the technical solutions in the embodiments of this article or in the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of this article. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0053] Figure 1 Shows a schematic diagram of the implementation environment of the method provided in the embodiments of this article;

[0054] Figure 2 Shows a schematic diagram of the steps of the method for establishing an insurance user classification model provided in the embodiments of this article;

[0055] Figure 3 Shows a schematic diagram of the steps for processing the insurance consumption coefficient in the embodiments of this article;

[0056] Figure 4 Shows a schematic diagram of the steps for processing continuous data in the embodiments of this article;

[0057] Figure 5 Shows a schematic diagram of the steps for processing training set data in the embodiments of this article;

[0058] Figure 6 Shows a schematic diagram of the steps for training an insurance user classification model in the embodiments of this article;

[0059] Figure 7 Shows a schematic diagram of the structure of the device for establishing an insurance user classification model provided in the embodiments of this article;

[0060] Figure 8 Shows a schematic diagram of the steps of the insurance user classification method provided in the embodiments of this article;

[0061] Figure 9Shows the structural schematic diagram of the insurance user classification device provided by the embodiments of the present invention;

[0062] Figure 10 Shows the structural schematic diagram of the device provided by the embodiments of the present invention.

[0063] Explanation of the reference numerals in the drawings:

[0064] 10. Server;

[0065] 20. Database;

[0066] 100. Data acquisition module;

[0067] 200. Type feature determination module;

[0068] 300. Processing module;

[0069] 400. Training module;

[0070] 500. User data to be classified acquisition module;

[0071] 600. Classification module;

[0072] 1002. Computer device;

[0073] 1004. Processor;

[0074] 1006. Memory;

[0075] 1008. Driving mechanism;

[0076] 1010. Input / output module;

[0077] 1012. Input device;

[0078] 1014. Output device;

[0079] 1016. Presentation device;

[0080] 1018. Graphical user interface;

[0081] 1020. Network interface;

[0082] 1022. Communication link;

[0083] 1024. Communication bus. Detailed implementation manners

[0084] The following will clearly and completely describe the technical solutions in the embodiments of this article in conjunction with the accompanying drawings in the embodiments of this article. Obviously, the described embodiments are only a part of the embodiments of this article, rather than all the embodiments. Based on the embodiments in this article, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of this article.

[0085] It should be noted that the terms "first", "second", etc. in the specification and claims of this article and the above accompanying drawings are used to distinguish similar objects, and do not necessarily need to be used to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments of this article described here can be implemented in an order other than those illustrated or described here. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, device, product, or equipment that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or are inherent to these processes, methods, products, or equipment.

[0086] In the prior art, in the process of insurance marketing, the marketing for existing customers only triggers messages by screening out some customers according to age and purchased products or randomly screening, but this marketing method has poor recall effect, low accuracy, and also brings relatively high operating costs. Therefore, how to improve the accuracy of classifying existing insurance users, thereby improving the reliability of insurance services and reducing the cost of insurance services has become a technical problem that urgently needs to be solved at present.

[0087] To solve the above problems, the embodiments of this specification provide a method for establishing an insurance user classification model. By using the insurance user classification model established by this method, the user attributes of existing insurance users can be determined quickly and accurately, the efficiency of user insurance services is improved, and the cost is reduced. As Figure 1As shown in the figure, it is a schematic diagram of the implementation environment of the method, which may include a server 10 and a database 20. The server 10 is connected to the database 20 to enable information interaction. The database 20 can be a database in the insurance field, used to store relevant insurance data, such as users' insurance data and insurance data in different regions. Among them, the insurance data in different regions can be obtained from relevant information repositories, and the users' insurance data is the insurance data purchased by users and the data generated during the interaction between users and insurance institutions. It should be noted that all the data involving users' privacy in the users' insurance data are obtained with the authorization of the users. The server 10 is configured with a patrol for training an insurance user classification model. By obtaining the users' insurance data and user attributes of historical insurance users in the database 20, as well as the regional insurance data of the regions where the historical insurance users belong, then performing binning encoding processing on the continuous data in the above data to convert the continuous data into categorical data, and then using the converted data combined with the original categorical data to train a preset initial classification model to obtain a trained insurance user classification model. By classifying and identifying existing insurance users through the trained insurance user classification model, the classification efficiency of user attributes can be improved, thereby providing a reliable basis for the insurance services of insurance institutions and reducing costs.

[0088] Optionally, the server 10 can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms.

[0089] In an optional embodiment, the implementation environment may also include a terminal, and the terminal can cooperate with the server 10 to train an insurance user classification model. Specifically, the terminal can include, but is not limited to, electronic devices such as smartphones, desktop computers, tablet computers, laptop computers, smart speakers, digital assistants, augmented reality (AR) / virtual reality (VR) devices, and smart wearable devices. Optionally, the operating systems running on the electronic devices can include, but are not limited to, Android systems, IOS systems, Linux, Windows, etc.

[0090] In addition, it should be noted that Figure 1 The figure only shows an application environment provided by the present disclosure. In actual applications, there may also be other application environments. For example, the training of the insurance user classification model can also be implemented on the terminal.

[0091] Specifically, the embodiments of the present invention provide a method for establishing an insurance user classification model, which can provide an accurate and reliable user classification and recognition model, saving the insurance service cost. Figure 2 It is a schematic diagram of the steps of a method for establishing an insurance user classification model provided by the embodiments of the present invention. This specification provides the method operation steps as described in the embodiments or flowcharts, but based on routine or non-creative labor, more or fewer operation steps may be included. The step order listed in the embodiments is only one way among the execution orders of numerous steps and does not represent the only execution order. When the actual system or device product is executed, it can be executed in the order of the method shown in the embodiments or the drawings or executed in parallel. Specifically, as Figure 2 shown, the method may include:

[0092] S101: Obtain the user insurance data and user attributes of historical insurance users, and the regional insurance data of the regions where the historical insurance users belong. The user attributes include existing users who have repurchased insurance and those who have not.

[0093] S102: Determine the type characteristics of different data in the user insurance data and the regional insurance data. The type characteristics include categorical data and continuous data.

[0094] S103: Process the continuous data in the user insurance data and the regional insurance data through binning encoding to obtain the converted continuous data, and integrate the categorical data, the encoded continuous data in the user insurance data and the regional insurance data, and the user attributes into training set data.

[0095] S104: Use the training set data to train a preset initial classification model to obtain a trained insurance user classification model.

[0096] It can be understood that in the embodiments of this specification, the preset initial classification model is jointly trained through the user insurance data of historical insurance users and the regional insurance data of the regions where the users belong to obtain an insurance user classification model for user classification. Among them, the continuous data in the user insurance data and the regional insurance data is processed through binning encoding to realize that both the user insurance data and the regional insurance data exist in the form of categorical data (i.e., discrete data), and then the processed data is combined with user attributes to train the model to obtain an insurance user classification model for user classification. Thus, the existing insurance users in the insurance institution can be classified through this classification model, and different service strategies can be adopted according to the classification results, improving the accuracy of the service. Compared with the random or simple screening methods in the prior art, the present application can reduce the service cost and enhance the service experience of users.

[0097] Among them, the user attributes include existing users who have repurchased insurance and those who have not. Existing users who have repurchased insurance can be historical insurance users who purchase insurance again. The type of insurance purchased again can be the same as or highly relevant to the type of insurance purchased last time. In such two purchase processes with a high degree of relevance of insurance types, the user's insurance data must also have a certain degree of relevance. Therefore, using the user's insurance data as the input data of the classification model can improve the accuracy of user classification. Existing users who have not repurchased insurance can be historical insurance users who have not purchased highly relevant insurance again within a specified time point. The specified time point can be the expiration date of the last insurance purchase, or a preset duration exceeding the expiration date. For example, if the expiration date of the last insurance purchase is October 1st, the specified time point can be October 1st, or half a month (October 16th) after the expiration date, one month (November 1st) after the expiration date, etc. The setting of the specified time point is not limited in the embodiments of this specification.

[0098] In the embodiments of this specification, the regional insurance data at least includes insurance density, insurance depth, and insurance coverage.

[0099] The insurance depth is the ratio of the annual premium income of the region to the annual gross domestic product of the region.

[0100] The insurance density is the ratio of the annual premium income of the region to the annual total population of the region.

[0101] The insurance coverage is the ratio of the number of people who purchase insurance annually in the region to the annual total population of the region.

[0102] It can be understood that in the embodiments of this specification, by collecting the insurance characteristics of different regions, the unique insurance data of each region can be determined, thereby determining the development level of the insurance business in each region. Furthermore, it can reflect to a certain extent the insurance awareness of users in this region. Therefore, using the regional insurance data as an index for user classification can also improve the accuracy and reliability of user classification. The division of each region can be carried out according to the geographical location of the region. For example, it can be divided according to administrative regions (such as prefecture-level cities). The development levels of different prefecture-level cities are different, and the insurance awareness of users will also change with the different development levels. For example, the insurance awareness of users in developed regions is stronger than that of users in underdeveloped regions, and thus there is a greater possibility of purchasing insurance. Therefore, multiple divided regions can be obtained by dividing according to administrative regions.

[0103] In the embodiments of this specification, the insurance depth can be the ratio of the annual premium income of a region to the annual gross domestic product (GDP) of the region, which represents the proportion of the user insurance service in the national economy of the region; the insurance density is the ratio of the annual premium income of the region to the annual total population of the region, which represents the per capita insurance consumption level in the region and further reflects the level of the insurance consumption ability of the users in the region; the insurance coverage rate is the ratio of the number of people who purchase insurance annually in the region to the annual total population of the region, which represents the proportion of the number of people consuming insurance among the residents in the region and further reflects the level of the insurance awareness of the users in the region.

[0104] The annual premium income of the region can be obtained by querying the premium data of insurance institutions. Correspondingly, in order to improve the accuracy and comprehensiveness of the data, the annual premium income of the region can be the premium data of multiple insurance institutions, which can be obtained by querying the insurance data published by each insurance institution. The annual GDP of the region can be the annual production data (GDP) published by each region, and the annual total population of the region can be the population data published by each region. The number of people who purchase insurance annually in the region can be obtained by querying the annual insurance data published by insurance institutions or the China Banking and Insurance Regulatory Commission.

[0105] In the embodiments of this specification, as Figure 3 shown, after the regional insurance data of the historical insurance users' affiliated regions further includes:

[0106] S201: Calculate the insurance consumption coefficient of the region according to the insurance density and insurance depth of the region where the historical insurance user is located. The insurance consumption coefficient represents the proportion of users' insurance consumption in the region;

[0107] S202: Modify the insurance consumption coefficient according to the range interval where the insurance consumption coefficient is located and the preset correction rules corresponding to the range interval;

[0108] S203: Use the modified insurance consumption coefficient and the insurance coverage rate of the historical insurance user as the regional insurance data of the region where the historical insurance user is located.

[0109] The inventor can obtain through data correlation analysis of the user attributes of historical insurance users and the regional insurance data of their affiliated regions that the composite index of "insurance density / insurance depth" has a significant correlation with the insurance consumption decision of users, that is, the "insurance density / insurance depth" (i.e., the insurance consumption coefficient) can be used as an index for classifying insurance users. The specific analysis process is as follows:

[0110] Obtain the insurance density data and insurance depth data of different regions, calculate the corresponding insurance consumption coefficients (E values), and summarize them to obtain Table 1 below:

[0111] Table 1 Statistical Table of Historical Insurance Data

[0112]

[0113] Pearson correlation calculation is performed on a set of data using the repurchase decision (yes). Pearson correlation calculation is a commonly used means for correlation calculation, which will not be described in detail in the embodiments of this specification. The correlation between the insurance consumption coefficient and the insurance repurchase decision can be obtained, and the calculation results are shown in Table 2 below:

[0114] Table 2 Pearson Calculation Results

[0115]

[0116] It can be seen from Table 2 that the correlation between the insurance consumption coefficient and the insurance consumption decision (i.e., the number of people with the repurchase decision (yes)) is close (generally, a correlation coefficient above 0.7 indicates a very close relationship, between 0.4 and 0.7 indicates a close relationship, and between 0.2 and 0.4 indicates a general relationship), and it has passed the two-sided significance test (less than or equal to 0.01 is considered passed), that is, the insurance consumption coefficient and the repurchase decision (yes) are significantly correlated.

[0117] Similarly, through the correlation analysis of the insurance coverage of the regions where historical insurance users are located, the correlation between the insurance coverage and the insurance repurchase decision can also be obtained to be relatively significant.

[0118] Continuing to refer to Table 1, on the basis that there is a significant correlation between the insurance consumption coefficient (E value) and the repurchase decision (yes), a corresponding relationship between the insurance consumption coefficient and the repurchase decision (yes) can be established, that is, for different ranges of changes in the insurance consumption coefficient, the proportion of the repurchase decision (yes) will also be different. That is to say, the influence degree of the insurance consumption coefficient in different change ranges on the insurance user's repurchase decision (yes) is different. Therefore, when correcting the insurance consumption coefficient, the range intervals of the insurance consumption coefficients with different influence degrees can be determined in advance through historical data, and then the weight ratio of the insurance consumption coefficient during model training can be adjusted based on different range intervals. For example, for the insurance consumption coefficient with a higher influence degree on the user's repurchase decision, increase its weight ratio during model training; for the consumption coefficient with a lower influence on the user's repurchase decision, reduce its weight ratio during model training, thereby further ensuring the accuracy of model training.

[0119] In the embodiments of this specification, the preset correction rule can be the correction of the training weight of the consumption coefficient, which can be determined according to the performance of the historical data of the consumption coefficient. The specific determination method is not limited in the embodiments of this specification.

[0120] In the embodiments of this specification, the user insurance data may include user consumption insurance data and user service insurance data. The user consumption insurance data may be data on the user's past insurance consumption and the user's insurance behavior data, and the user service insurance data may be service data of the insurance institution for the user.

[0121] The user consumption insurance data may include: multiple purchases, number of purchases for oneself, number of product views, number of wealth management insurance policies, marital status, number of platform logins, whether active in 60 days, number of views in the past year, whether active in 30 days, whether WeChat is bound, whether active in 7 days, number of purchase channels, number of purchased products, number of claims, number of maturity policies, etc.

[0122] The user service insurance data may include: number of customer service receptions, number of claims, claim amount, indemnity amount, number of customer service seat receptions, call duration, number of complaints, etc.

[0123] By analyzing the user insurance data and the regional insurance data, the type characteristics of different data can be obtained. For example, the user's marital status data can be categorical data (i.e., discrete data), and the user's product view count can be continuous data, etc. To improve the reliability and accuracy of model training, optionally, as Figure 4 shown, the continuous data in the user insurance data and the regional insurance data are processed by binning encoding to obtain the encoded continuous data, including:

[0124] The following steps are performed for each type of continuous data:

[0125] S301: Sort the sample values corresponding to the continuous data in ascending order to obtain a sample value queue;

[0126] S302: Take the average value of two adjacent sample values in the sample value queue as the leaf division point to obtain a set of leaf division points;

[0127] S303: Calculate the Gini coefficient of each leaf division point in the set of leaf division points as a binary classification point in turn, and take the leaf division point with the smallest Gini coefficient as the optimal division point;

[0128] S304: Divide the sample values corresponding to the continuous data into two subsets with the optimal division point, and encode the two subsets according to a preset encoding method to obtain the encoded data.

[0129] It can be understood that for continuous data, in the embodiments of this specification, the Gini coefficient of the binary classification data set obtained by each leaf division point is calculated, and then the discretization effect of the discrete data interval obtained by different leaf division points is obtained. The smaller the Gini coefficient, the better the discretization effect. The leaf division point with the smallest Gini coefficient is used as the optimal division point, which can improve the feature discretization effect of continuous data. The calculation of the Gini coefficient is a well-known calculation formula for those skilled in the art and is not limited in the embodiments of this specification.

[0130] In some other embodiments, in order to ensure the reliability of the data after division and avoid the situation of excessive concentration of the sample size after data division, the sample number of the data set in each leaf node after division can also be calculated. When the sample number is lower than a preset sample ratio (such as 5% of the total sample number), it may be necessary to adjust the optimal division point, use the leaf division point with the second smallest Gini coefficient as the optimal division point, and repeatedly calculate the sample number of the data set in each leaf node until the sample number is lower than the preset sample ratio.

[0131] The preset encoding method can be one-hot encoding, which can realize the processing of data with practical significance (such as data with units), and can also play a role in expanding features.

[0132] In some other embodiments, for the encoded data, coarse-grained feature extraction can also be performed by the principal component analysis (PCA) method, and a certain number of data with higher importance can be selected from various data. The specific analysis process is not limited in the embodiments of this specification.

[0133] In the embodiments of this specification, as Figure 5 shown, the continuous data in the user insurance data and the regional insurance data is processed by binning encoding to obtain the converted continuous data, and the categorical data, the encoded continuous data, and the user attributes in the user insurance data and the regional insurance data are integrated into the training set data, including:

[0134] S401: Use different data types in the user insurance data and the regional insurance data as different binary tree feature nodes;

[0135] S402: Generate multiple binary trees according to the binary tree feature nodes, and each binary tree includes different combinations of binary tree feature node divisions;

[0136] S403: Calculate the Gini coefficient of each binary tree in turn according to the data corresponding to the binary tree feature nodes in each binary tree;

[0137] S404: Use the combination of binary tree feature nodes corresponding to the binary tree with the smallest Gini coefficient as the target node combination, so that the binary tree generated according to the target node combination and the user attributes are used as the training set data of the preset initial classification model.

[0138] It can be understood that after the above discretization process for continuous data, discretized user insurance data and regional insurance data can be obtained. Furthermore, the training set data can be integrated to obtain a data set that is more conducive to training. Since the user attributes include two types (i.e., existing users who have repurchased insurance and those who have not), the embodiments of this specification perform a binary process on the discretized data, thereby obtaining data sets with different data combinations (i.e., binary orders).

[0139] Among them, the binary tree generated by the binary tree feature nodes can be a division sequence generated by each binary tree feature node according to the node order. In this way, different binary trees can be obtained through different orders of feature nodes. Furthermore, by calculating the Gini coefficient corresponding to each binary tree, the Gini coefficient can be the sum of the Gini coefficients corresponding to different node positions in the binary tree. In this way, the combination of binary tree feature nodes with the smallest Gini coefficient is used as the target node combination, so that the binary tree generated according to the target node combination is used as the training set data of the preset initial classification model, reducing the computational amount during training and improving the efficiency and accuracy of model training.

[0140] Exemplarily, when there are N types of discretized user insurance data and regional insurance data, then N! node combinations can be obtained. Furthermore, N! binary trees can be determined. Calculate the Gini coefficient corresponding to each node in each binary tree in turn, and use the node combination in the binary tree with the smallest sum of Gini coefficients as the target node combination. Therefore, the combination and order of data are integrated in advance from the training set, thereby further improving the training efficiency of the insurance user classification model.

[0141] In some other embodiments, the target node combination can also be determined by performing binary operations layer by layer. For example, first determine the first node in the binary tree. Taking each node as the first node, correspondingly, all nodes other than this node are subsequent nodes. In this way, multiple node sets (i.e., the first node and subsequent nodes) are obtained. Calculate the Gini coefficient corresponding to each node set in turn, and use the first node in the node set with the smallest Gini coefficient as the target first node. Then determine the target second node from the subsequent nodes other than the target first node, and loop the above steps to determine the node order in the subsequent nodes in turn, thereby obtaining the target node combination. Through the above steps, the target node combination can be quickly determined, thereby improving the optimization efficiency of the training set data. Exemplarily, there are n types of insurance data [A1, A2... A n, preferably, A1, A2... A are selected in sequence n as the first node, so that n node sets can be obtained. Taking A1 as an example, A1 is the first node, and the subsequent nodes are [A2... A n , and the Gini coefficient corresponding to each node set is calculated in sequence. When the Gini coefficient corresponding to the node set with A1 as the first node is the smallest, then A1 is used as the target first node, and the subsequent nodes [A2... A n are used as another node. However, in this node, the features of each node are not separated, and the subsequent nodes [A2... A n need to be further divided. The division rule is the same as the above method, and the second node, the third node are determined in sequence until the last node is determined.

[0142] In the embodiments of this specification, as Figure 6 shown, the method of training the preset initial classification model using the training set data to obtain the trained insurance user classification model includes:

[0143] S501: Determine the region to which the preset initial classification model to be trained belongs and the training set data corresponding to the region;

[0144] S502: Input the training set data into the preset initial classification model and train it using the preset loss function to obtain the pre-trained first classification model;

[0145] S503: According to the pre-trained first classification model, determine the gain scores of the user insurance data of different data types during the training process;

[0146] S504: Sort the gain scores of the user insurance data of different data types;

[0147] S505: Use the user insurance data of the preset type quantity as the target user insurance data combination in descending order of the sorting;

[0148] S506: Extract the target training set data from the training set data according to the target user insurance data combination;

[0149] S507: Fine-tune the first classification model through the target training set data to obtain the trained insurance user classification model.

[0150] It can be understood that the training set data can be the training set data determined by the above-mentioned step S103, or the training set data determined by the above-mentioned steps S401 to S404. The preset initial classification model can be an XGboost model. After one iteration by the XGboost model, the weights of the leaf nodes will be multiplied by the learning rate, mainly to weaken the influence of each tree and leave more room for learning later. In the actual training process, the historical data of all types (i.e., the training set data) can be used for training first, and the adjusted model parameters will be obtained, and then the general first classification model will be obtained. Since the models corresponding to different regions are also targeted (i.e., the contribution degrees of each data type are different), during the training process of the XGboost model, the gain scores of each data type are calculated, and the feature with the maximum gain will be selected as the splitting basis, thus memorizing the importance of each feature during model training. Therefore, by sorting the importance of data types, the user insurance data of the preset type quantity can be determined as the target insurance data combination, and then the fine-tuning of the first classification model can be realized, ensuring that different region data corresponds to different classification models and making the trained classification model have regional characteristics.

[0151] In some other embodiments, the user insurance data may include user consumption insurance data and user service insurance data. Therefore, when performing the scoring and sorting, the scoring and sorting of the user consumption insurance data and the scoring and sorting of the user service insurance data can be calculated respectively. Then, for different combinations of input feature factors, that is, the user consumption insurance data combination and the user service insurance data combination. Exemplarily, for example, 15 type quantities can be selected as the user consumption insurance data combination (refer to Table 3), and 8 type quantities can be selected as the user service insurance data combination (refer to Table 4).

[0152] Table 3 User Consumption Insurance Data Combination

[0153] Data type Gain score Total number of purchases 0.9333936 Number of purchases for oneself 0.6448525 Number of product views 0.6245873 Number of medical insurance purchases 0.1735075 Number of wealth management insurance policies 0.1587547 Marital status 0.1568718 Number of platform logins 0.120805 Active in the past 60 days 0.0952249 Number of views in the past year 0.0937207 Active in the past 30 days 0.0824651 WeChat binding status 0.0790407 Active in the past 7 days 0.0468789 Number of purchases for children 0.0356029 Number of purchases for parents 0.0338616 Number of purchases for spouse 0.0256321

[0154] Table 4 User Service Insurance Data Combination

[0155]

[0156]

[0157] In the embodiments of this specification, the preset loss function includes:

[0158]

[0159] Where: L is the total calculated value of the loss function, L i is the calculated value of the loss function of sample i, y iDenote the label of sample \(i\), where the positive class is \(1\) and the negative class is \(0\); \(p\) i Denote the probability that sample \(i\) is predicted as the positive class.

[0160] It should be noted that the classification model can also be other models, such as Gradient Boosting Decision Tree (GBDT). The specific model is not limited in the embodiments of this specification.

[0161] A method for establishing an insurance user classification model provided by an embodiment of this specification. By obtaining the user insurance data and regional insurance data of historical insurance users, and performing binning transcoding on the continuous data in the user insurance data and the regional insurance data to obtain training set data for training, where the user attributes are used as the target output for training, thereby obtaining an insurance user classification model for classifying users, improving the classification accuracy of existing insurance users, and further providing a reliable guidance for insurance services.

[0162] Based on the same inventive concept, an embodiment of this specification also provides an apparatus for establishing an insurance user classification model, as Figure 7 shown. The apparatus includes:

[0163] A data acquisition module 100, configured to acquire the user insurance data and user attributes of historical insurance users, and the regional insurance data of the regions where the historical insurance users are located. The user attributes include existing users who have repurchased insurance and those who have not repurchased insurance.

[0164] A type feature determination module 200, configured to determine the type features of different data in the user insurance data and the regional insurance data. The type features include categorical data and continuous data.

[0165] A processing module 300, configured to perform binning encoding on the continuous data in the user insurance data and the regional insurance data to obtain the converted continuous data, and integrate the categorical data, the encoded continuous data, and the user attributes in the user insurance data and the regional insurance data into training set data.

[0166] A training module 400, configured to train a preset initial classification model using the training set data to obtain a trained insurance user classification model.

[0167] The beneficial effects obtained by the apparatus provided by the embodiments of this specification are the same as those obtained by the above method, and will not be elaborated in the embodiments of this specification.

[0168] Based on the above-provided method for establishing an insurance user classification model, an embodiment of this specification also provides an insurance user classification method, as Figure 8As shown, the method includes:

[0169] S601: Obtain the user insurance data of the user to be classified and the regional insurance data of the region where the user to be classified belongs;

[0170] S602: Substitute the user insurance data and the regional insurance data into the insurance user classification model established by the above-mentioned method to obtain the user attributes of the user to be classified.

[0171] It can be understood that the insurance user classification model established by the above method can be used to classify the existing users in the insurance institution to determine the direction of the repurchase decision of the existing users, so as to facilitate the corresponding service operations of the insurance institution service personnel. For example, when the user attributes obtained by classifying the user belong to the type of repurchase decision (yes), targeted marketing can be carried out on the user, thereby improving the accuracy and efficiency of insurance services.

[0172] Based on the same inventive concept, an embodiment of this specification also provides an insurance user classification device, as Figure 9 shown, the device may include:

[0173] A user data acquisition module 500 for the user to be classified, which acquires the user insurance data of the user to be classified and the regional insurance data of the region where the user to be classified belongs;

[0174] A classification module 600 that substitutes the user insurance data and the regional insurance data into the insurance user classification model established by the above-mentioned method to obtain the user attributes of the user to be classified.

[0175] As Figure 10As shown, a computer device provided by an embodiment of this article. The device in this article can be the computer device in this embodiment, which executes the method in this article. The computer device 1002 may include one or more processors 1004, such as one or more central processing units (CPUs), and each processing unit can implement one or more hardware threads. The computer device 1002 may also include any memory 1006, which is used to store any kind of information such as code, settings, data, etc. Non-limiting examples include any combination of the following: any type of RAM, any type of ROM, flash memory devices, hard disks, optical discs, etc. More generally, any memory can use any technology to store information. Further, any memory can provide volatile or non-volatile retention of information. Further, any memory can represent a fixed or removable component of the computer device 1002. In one case, when the processor 1004 executes the associated instructions stored in any memory or combination of memories, the computer device 1002 can perform any operation of the associated instructions. The computer device 1002 also includes one or more drive mechanisms 1008 for interacting with any memory, such as a hard disk drive mechanism, an optical disc drive mechanism, etc.

[0176] The computer device 1002 may also include an input / output module 1010 (I / O), which is used to receive various inputs (via the input device 1012) and to provide various outputs (via the output device 1014). A specific output mechanism may include a presentation device 1016 and an associated graphical user interface (GUI) 1018. In other embodiments, the input / output module 1010 (I / O), the input device 1012, and the output device 1014 may not be included, and it only serves as a computer device in the network. The computer device 1002 may also include one or more network interfaces 1020, which are used to exchange data with other devices via one or more communication links 1022. One or more communication buses 1024 couple the components described above together.

[0177] The communication link 1022 can be implemented in any way, for example, through a local area network, a wide area network (e.g., the Internet), a point-to-point connection, etc., or any combination thereof. The communication link 1022 may include any combination of hardwired links, wireless links, routers, gateway functions, name servers, etc. governed by any protocol or combination of protocols.

[0178] Corresponding to Figures 2 - 6 or Figure 8 For the method in, an embodiment of this article also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is run by a processor, it executes the steps of the above method.

[0179] Embodiments of the present disclosure also provide a computer-readable instruction. When a processor executes the instruction, the program therein causes the processor to execute the method as Figures 2 - 6 or Figure 8 shown.

[0180] It should be understood that in various embodiments of the present disclosure, the magnitudes of the sequence numbers of the above processes do not imply the order of execution. The order of execution of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present disclosure.

[0181] It should also be understood that in the embodiments of the present disclosure, the term "and / or" is merely a description of the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B may represent: A exists alone, A and B exist simultaneously, and B exists alone. In addition, the character " / " in the present disclosure generally represents an "or" relationship between the associated objects before and after.

[0182] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of the two. To clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described according to functions in the above description. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present disclosure.

[0183] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the systems, devices, and units described above can refer to the corresponding processes in the foregoing method embodiments, and will not be elaborated herein.

[0184] In several embodiments provided in the present disclosure, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division, and there may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed coupling or direct coupling or communication connection to each other can be an indirect coupling or communication connection through some interfaces, devices, or units, and can also be an electrical, mechanical, or other form of connection.

[0185] The unit described as a separation component may or may not be physically separated. The component displayed as a unit may or may not be a physical unit, that is, it may be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiments in this article.

[0186] In addition, each functional unit in the embodiments of this article can be integrated into a processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of a software functional unit.

[0187] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution in this article, in essence, or the part that contributes to the prior art, or all or part of this technical solution can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the embodiments of this article. The aforementioned storage medium includes: various media that can store program codes such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs.

[0188] Specific embodiments are applied in this article to elaborate on the principles and implementation manners of this article. The description of the above embodiments is only used to help understand the method and its core idea in this article; at the same time, for those of ordinary skill in the art, according to the idea of this article, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to this article.

Claims

1. A method for establishing an insurance user classification model, characterized in that, The method includes: Obtaining the user insurance data and user attributes of historical insurance users, as well as the regional insurance data of the regions to which the historical insurance users belong. The user attributes include existing users who have repurchased insurance and those who have not. The repurchased insurance means purchasing insurance that meets the correlation threshold multiple times; Determining the type characteristics of different data in the user insurance data and the regional insurance data. The type characteristics include categorical data and continuous data; Processing the continuous data in the user insurance data and the regional insurance data through binning encoding to obtain the converted discrete data, and integrating the categorical data, the encoded discrete data, and the user attributes in the user insurance data and the regional insurance data into training set data. The categorical data and the encoded discrete data represent training features, and the user attributes represent training labels; Training a preset initial classification model using the training set data to obtain a trained insurance user classification model; Wherein, after obtaining the regional insurance data of the region to which the historical insurance user belongs, it further includes: Calculating a regional insurance consumption coefficient based on the insurance density and insurance penetration of the region to which the historical insurance user belongs. The insurance consumption coefficient represents the proportion of users in the region for insurance consumption; Correcting the insurance consumption coefficient according to the range interval where the insurance consumption coefficient is located and the preset correction rules corresponding to the range interval; Taking the corrected insurance consumption coefficient and the insurance coverage of the historical insurance user as the regional insurance data of the region to which the historical insurance user belongs.

2. The method according to claim 1, wherein: The regional insurance data at least includes insurance density, insurance penetration, and insurance coverage; The insurance penetration is the ratio of the annual premium income of the region to the annual gross domestic product of the region; The insurance density is the ratio of the annual premium income of the region to the annual total population of the region; The insurance coverage is the ratio of the number of people who purchase insurance annually in the region to the annual total population of the region.

3. The method according to claim 1, characterized in that, The processing of the continuous data in the user insurance data and the regional insurance data through binning encoding to obtain the encoded discrete data includes: Performing the following steps for each type of continuous data: Sorting the sample values corresponding to the continuous data from smallest to largest to obtain a sample value queue; Taking the average value of two adjacent sample values in the sample value queue as the leaf division point to obtain a set of leaf division points; Calculating the Gini coefficient of each leaf division point in the set of leaf division points as a binary classification point in turn, and taking the leaf division point with the smallest Gini coefficient as the optimal division point; Dividing the sample values corresponding to the continuous data into two subsets with the optimal division point, and encoding the two subsets according to a preset encoding method to obtain the encoded data.

4. The method according to claim 1, wherein Processing the continuous data in the user insurance data and the regional insurance data through binning encoding to obtain the converted discrete data, and integrating the categorical data, the encoded discrete data, and the user attributes in the user insurance data and the regional insurance data into training set data, including: Regarding different data types in the user insurance data and the regional insurance data as different binary tree feature nodes; Generating multiple binary trees according to the binary tree feature nodes, where each binary tree includes different combinations of binary tree feature node partitions; Calculating the Gini coefficient of each binary tree in sequence according to the data corresponding to the binary tree feature nodes in each binary tree; Regarding the combination of binary tree feature node partitions corresponding to the binary tree with the smallest Gini coefficient as the target node combination, so that the binary tree generated according to the target node combination and the user attributes are used as the training set data of the preset initial classification model.

5. The method according to claim 1 or 4, characterized in that, Training the preset initial classification model using the training set data to obtain a trained insurance user classification model, including: Determining the region to which the preset initial classification model to be trained belongs and the training set data corresponding to the region; Inputting the training set data into the preset initial classification model and training it using a preset loss function to obtain a pre-trained first classification model; Determining the gain scores of the user insurance data of different data types during the training process according to the pre-trained first classification model; Performing score ranking on the gain scores of the user insurance data of different data types; Regarding the user insurance data of the preset type quantity as the target user insurance data combination in descending order of the ranking; Extracting the target training set data from the training set data according to the target user insurance data combination; Fine-tuning the first classification model through the target training set data to obtain a trained insurance user classification model.

6. A method for classifying insurance users, characterized in that, The method includes: Obtaining the user insurance data of the user to be classified and the regional insurance data of the region to which the user to be classified belongs; Substituting the user insurance data and the regional insurance data into the insurance user classification model established by the method according to any one of claims 1 to 5 to obtain the user attributes of the user to be classified.

7. An apparatus for establishing an insurance user classification model, characterized in that, The device includes: A data acquisition module for acquiring the user insurance data and user attributes of historical insurance users, and the regional insurance data of the regions to which the historical insurance users belong. The user attributes include existing users who have repurchased insurance and those who have not. Repurchasing insurance means purchasing insurance that meets the correlation threshold multiple times; A type feature determination module for determining the type features of different data in the user insurance data and the regional insurance data. The type features include categorical data and continuous data; A processing module, configured to perform binning encoding processing on continuous data in the user insurance data and the regional insurance data to obtain transformed discrete data, and integrate categorical data, the encoded discrete data, and user attributes in the user insurance data and the regional insurance data into training set data; the categorical data and the encoded discrete data represent training features, and the user attributes represent training labels; A training module, configured to train a preset initial classification model using the training set data to obtain a trained insurance user classification model; Wherein, after obtaining the regional insurance data of the historical insurance user's affiliated region, it further includes: Calculating a regional insurance consumption coefficient based on the insurance density and insurance depth of the region where the historical insurance user belongs, and the insurance consumption coefficient represents the proportion of users in the region for insurance consumption; According to the range interval where the insurance consumption coefficient is located and the preset correction rule corresponding to the range interval, correcting the insurance consumption coefficient; Taking the corrected insurance consumption coefficient and the insurance coverage of the historical insurance user as the regional insurance data of the region where the historical insurance user belongs.

8. An insurance user classification device, characterized in that, The apparatus includes: A to-be-classified user data acquisition module, configured to acquire user insurance data of a to-be-classified user and regional insurance data of the region where the to-be-classified user belongs; A classification module, configured to substitute the user insurance data and the regional insurance data into an insurance user classification model established by the method according to any one of claims 1 to 5 to obtain the user attributes of the to-be-classified user.

9. A computer device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method according to any one of claims 1 to 6.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, and when the computer program is executed by the processor, it implements the method according to any one of claims 1 to 6.

Images