WO2018223719A1 - 用户投保行为预测的方法、装置、计算设备及介质 - Google Patents

用户投保行为预测的方法、装置、计算设备及介质 Download PDF

Info

Publication number
WO2018223719A1
WO2018223719A1 PCT/CN2018/074884 CN2018074884W WO2018223719A1 WO 2018223719 A1 WO2018223719 A1 WO 2018223719A1 CN 2018074884 W CN2018074884 W CN 2018074884W WO 2018223719 A1 WO2018223719 A1 WO 2018223719A1
Authority
WO
WIPO (PCT)
Prior art keywords
user
data
behavior
type
insurance
Prior art date
Application number
PCT/CN2018/074884
Other languages
English (en)
French (fr)
Inventor
刘永凡
Original Assignee
平安科技（深圳）有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 平安科技（深圳）有限公司 filed Critical 平安科技（深圳）有限公司
Publication of WO2018223719A1 publication Critical patent/WO2018223719A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
    • G06Q40/08—Insurance
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9535—Search customisation based on user profiles and personalisation

Definitions

  • the present application belongs to the field of Internet technologies, and in particular, to a method, an apparatus, a computing device, and a medium for predicting a user's insurance behavior.
  • the embodiment of the present application provides a method, a device, a computing device, and a medium for predicting a user's insurance behavior, so as to solve the problem in the prior art that the insurance service personnel subjectively analyze the insurance products that may be of interest to the user, resulting in the user.
  • the problem of predicting efficiency and accuracy of insurance behavior is low.
  • a first aspect of the embodiments of the present application provides a method for predicting a user's insurance behavior, including:
  • Pre-processing the feature data to remove noise data to obtain pre-processed data Pre-processing the feature data to remove noise data to obtain pre-processed data
  • the learning model is configured to match behavior of the first type of user with a preset insurance product in the database, where the first type of user is an insurance customer;
  • a second aspect of the embodiments of the present application provides an apparatus for predicting a user's insurance behavior, including:
  • An obtaining module configured to obtain feature data of a sample user from a user attribute library, where the feature data includes a behavior type of the sample user and a behavior number corresponding to the behavior type, and the behavior type includes a subscription product type and a visiting website type;
  • a preprocessing module configured to preprocess the feature data acquired by the acquiring module, remove the noise data, and obtain the preprocessed data
  • a modeling module configured to establish a learning model according to the preprocessed data obtained by the preprocessing module, wherein the learning model is configured to match the behavior of the first type of user with a preset insurance product in the database, where the first type of user is Insurance customer
  • a prediction module configured to predict a behavior of the first type of user to be predicted based on a learning model established by the modeling module, and obtain a target insurance product of the first type of user
  • the pushing module is configured to push the first type of user to the second type of user, so that the second type of user determines the first type of user as the target user of the target insurance product, wherein the second type of user is an insurance business person.
  • a third aspect of the embodiments of the present application provides a computing device for predicting user insurance behavior, comprising a memory and a processor, wherein the memory stores computer readable instructions executable on the processor, the processor The steps of the method of predicting user insurance behavior as described in the first aspect when the computer readable instructions are executed.
  • a fourth aspect of the embodiments of the present application provides a computer readable storage medium storing computer readable instructions, the computer readable instructions being executed by a processor to implement the first aspect as described in the first aspect The steps of the user's method of insuring behavior prediction.
  • the feature data of the sample user is obtained from the user attribute database
  • the pre-processing data is obtained by denoising the feature data
  • the learning model is established according to the pre-processed data
  • the prediction model is implemented based on the learning model.
  • the first type of user performs behavior prediction, obtains the target insurance product of the first type of user, and can automatically and accurately find the relationship between the user behavior and the insurance product by establishing a learning model, thereby realizing the first type of user feeling.
  • Accurate prediction of the insurance products of interest improve the prediction efficiency and accuracy of the user's insurance behavior, and the level of intelligent prediction, and push the first type of users to the second type of users, so that the second type of users can timely understand the purchase of insurance products.
  • Potential users with high probability so as to carry out targeted continuous tracking, improve the sales success rate and efficiency of insurance products.
  • Embodiment 1 is a schematic flowchart of an implementation method of a method for predicting a user's insurance behavior provided by Embodiment 1 of the present application;
  • FIG. 2 is a schematic flowchart of an implementation method of a method for predicting a user's insurance behavior provided by Embodiment 2 of the present application;
  • FIG. 3 is a schematic flowchart of synchronizing basic data to a user attribute database in a method for predicting user insurance behavior provided by Embodiment 2 of the present application;
  • Embodiment 4 is a schematic diagram of an apparatus for predicting a user's insurance behavior provided by Embodiment 3 of the present application;
  • FIG. 5 is a schematic diagram of an apparatus for predicting user insurance behavior provided by Embodiment 4 of the present application.
  • FIG. 6 is a schematic diagram of a computing device for predicting user insurance behavior provided by Embodiment 5 of the present application.
  • Embodiment 1 is a diagrammatic representation of Embodiment 1:
  • FIG. 1 is a flowchart of a method for predicting a user's insurance behavior according to the first embodiment of the present application.
  • the execution subject of the embodiment of the present application is a computing device, which may be a server, etc., and a user insurance behavior prediction method illustrated in FIG.
  • the method may specifically include steps S101 to S104, which are detailed as follows:
  • S101 Obtain feature data of the sample user from the user attribute database, where the feature data includes a behavior type of the sample user and a behavior number corresponding to the behavior type, and the behavior type includes a subscription product type and a visiting website type.
  • the user attribute library may be a big data platform, and the user attribute library includes feature data of the sample user, the feature data includes a behavior type of the sample user and a behavior number corresponding to each behavior type, and the behavior type may include ordering the product type and accessing Website type.
  • the sample user may be a user who performs behavior actions through various behavior interfaces, wherein the behavior interface includes but is not limited to a website, an application of an intelligent terminal (Application, APP), and an entity channel such as an entity counter or a cooperative company, and the above behavior interface
  • the corresponding behavioral actions include the user accessing the website, operating through the APP, and handling the business through the physical channel.
  • the feature data of each sample user may exist in the form of a feature information table, and the feature information table may be represented by a two-dimensional array, the first dimension of the two-dimensional array is used to identify the behavior type of the sample user, and the second dimension of the two-dimensional array Dimensions are used to identify the number of behaviors for each behavior type.
  • the sample user's behavior type may specifically include the product type of the sample user to order a specific product, and the product type may be a product type purchased by the sample user through the network, such as a travel product, and a transportation product such as an air ticket, and the product type may also be a sample. Insurance products purchased by users through physical channels, etc.
  • the sample user's behavior type may also include the type of website that the sample user visits the website, and the website type may be a shopping website, a news current affairs website, or the like.
  • S102 Perform pre-processing on the feature data of the sample user, remove the noise data, and obtain the pre-processed data.
  • the feature data obtained in step S101 is preprocessed, and the service corresponding to different behavior types of the sample users in the feature data is differentiated, and the noise data such as cheating, brushing, and misoperation is cleared, and the denoised pre-processing is obtained. Data processing.
  • the feature data of the sample user is generated during the various network behaviors or entity channel behaviors of the sample user, there may be a large amount of noise data, and the specific data mining algorithm, such as a clustering algorithm, may be used to filter out the feature data. Noise, so that subsequent analysis and modeling of feature data can be more accurate.
  • S103 Establish a learning model according to the pre-processed data, where the learning model is used to match the behavior of the first type of user with a preset insurance product in the database, and the first type of user is an insurance customer.
  • modeling is performed according to the pre-processed data obtained in step S102, and a learning model is established by using a model training algorithm, for example, a learning model between a user's behavior and an insurance product with purchase intention is established by a big data hadoop learning algorithm. Used to match the behavior of an insurance customer with the insurance products preset in the database.
  • the database may be an insurance product database of the insurance management system in which product information of various insurance products is pre-stored, including insurance type, insurance name, insured data information, insured object, insurance coverage, and income.
  • S104 Perform behavior prediction based on the learning model for the first type of user to be predicted, and obtain a target insurance product of the first type of user.
  • step S103 behavior prediction is performed on a specific insurance customer to be predicted, and the behavior type of the insurance customer to be predicted is input into the learning model, and the calculation and analysis of the learning model are used to obtain the behavior model.
  • Target insurance products can be recommended to insurance customers to be predicted through web push or other push methods. For example, when an insurance customer has an act of purchasing a travel product, or a behavior of purchasing a transportation product such as a train ticket or a ticket, a target insurance product that matches the behavior of the insurance customer can be obtained according to the learning model, such as travel safety insurance, Aviation delay insurance, etc., and push these target insurance products to the insurance customers to be predicted.
  • the learning model such as travel safety insurance, Aviation delay insurance, etc.
  • S105 Push the first type of user to be predicted to the second type of user, so that the second type of user determines the first type of user as the target user of the target insurance product, and the second type of user is the insurance business personnel.
  • the user information of the insurance customer to be predicted is pushed to the insurance business personnel, and the user information may include the name, contact information, insurance products that may be of interest, etc., and the insurance business personnel may determine the insurance customer as the purchase.
  • the target user of the target insurance product conducts targeted and continuous tracking of the insurance user, so that the insurance business personnel no longer blindly select the target user, thereby improving the sales success rate and efficiency of the insurance product.
  • the feature data of the sample user is obtained from the user attribute library, and the feature data is denoised to obtain preprocessed data, and a learning model is established according to the preprocessed data, and the prediction model is implemented based on the learning model.
  • the first type of user performs behavior prediction, obtains the target insurance product of the first type of user, and can automatically and accurately find the relationship between the user behavior and the insurance product by establishing a learning model, thereby achieving the interest of the first type of user.
  • Accurate prediction of insurance products improve the prediction efficiency and accuracy of user insurance behavior, and intelligent prediction level, and push the first type of users to the second type of users, so that the second type of users can know the probability of purchasing insurance products in time. High potential users to carry out targeted continuous tracking to improve the sales success rate and efficiency of insurance products.
  • Embodiment 2 is a diagrammatic representation of Embodiment 1:
  • FIG. 2 is a flowchart of a method for predicting a user's insurance behavior according to the second embodiment of the present application.
  • the execution subject of the embodiment of the present application is a computing device, which may be a server, etc.
  • the method may specifically include steps S201 to S210, which are detailed as follows:
  • the basic data of the sample user is periodically synchronized to the user attribute database, wherein the basic data of the sample user includes network access data and policy data in the insurance management system, and the network access data includes product ordering information and website access information.
  • step S2011 to step S2019 the process of periodically acquiring the basic data of the sample user and synchronizing the basic data to the user attribute database is as shown in FIG. 3, and the specific process may be completed by using step S2011 to step S2019, and the details are as follows:
  • S2011 Get the basic data of the sample user.
  • a sample user can be a user who behaves through various behavioral interfaces. Sample users of different behavioral actions correspond to different basic data.
  • the sample user who accesses the website and operates through the APP the basic data includes network access data
  • the network access data can be analyzed by analyzing the website user's website browsing record, network operation record, etc., to obtain website access information, the website
  • the access information may specifically include the type of content accessed, the product information of interest or purchase, the number of visits, and the like.
  • the basic data includes policy data in the insurance management system.
  • the policy data may include user information for purchasing insurance products, specific insurance product information purchased, and purchase time.
  • step S2012 It is judged whether the data cursor of the basic data in the import summary time period is executed, and if the execution is not completed, the process proceeds to step S2013, otherwise, step S2014 is performed.
  • the import summary time period is a periodic time period in which the sample user's basic data is synchronized to the user attribute database in a specified period, for example, the time period may be every 10 minutes, or every 30 minutes, and the like.
  • the time period can be set to a relatively short time length.
  • the time period can be set to a relatively long time length.
  • the specific duration of the import summary period can be set according to the needs of the actual application. There is no restriction here.
  • a cursor acts as a pointer and is a mechanism that extracts one record at a time from a result set that includes multiple data records.
  • the data cursor of the basic data refers to one piece of basic data extracted each time in the set of basic data of the sample user obtained in step S2011. If the execution of the data cursor is completed, the basic data of the sample user obtained in step S2011 has been inserted into the temporary table, and the process jumps to step S2014; if the data cursor is not executed, the process proceeds to step S2013 to extract the next synchronization to be synchronized. The underlying data is inserted into the temporary table.
  • the basic data pointed to by the data cursor is inserted into the temporary table, and the process returns to step S2012 to continue processing the next basic data to be synchronized in the set of basic data through the data cursor.
  • step S2014 Determine whether the analysis cursor for analyzing the basic data is completed. If the execution is not completed, proceed to step S2015. Otherwise, the synchronization process is completed, and the flow jumps to step S2019.
  • the basic data that needs to be synchronized to the user attribute database has been completely inserted into the temporary table, and the temporary table is extracted each time by analyzing the cursor in the temporary table.
  • a basic data in the process is processed by the synchronization process. If the analysis cursor execution is completed, the synchronization process is completed, and the process jumps to step S2019; if the analysis cursor is not executed, the process proceeds to step S2015, and the next basic data to be synchronized is extracted for synchronization processing.
  • the analysis cursor for analyzing the basic data in the temporary table is not completed, the basic data pointed to by the analysis cursor is analyzed and summarized, and the synchronization data and the corresponding data record table are obtained.
  • the data to be synchronized may include access record data, service feature data, common link data, and the like of the sample user, and the data of each data type may be saved by means of the data record table. , a data type of data is saved to a data record table.
  • step S2016 In the current user attribute database, it is determined whether the record of the data to be synchronized already exists in the data record table corresponding to the data to be synchronized, if yes, step S2018 is performed, otherwise step S2017 is performed.
  • the user attribute database may be a big data platform.
  • the data record table of the current user attribute database only the latest value of the type data is recorded for each data to be synchronized, so it is necessary to determine the data record table corresponding to the data to be synchronized. Whether there is already a record of the data to be synchronized in the middle, if it already exists, step S2018 is performed, otherwise step S2017 is performed.
  • the data to be synchronized is inserted into the corresponding data record table. After the insertion operation is completed, the flow jumps to step S2019.
  • the data to be synchronized is updated to the corresponding record that already exists.
  • S202 Perform cluster analysis on the basic data of the sample user to obtain feature data of the sample user.
  • the step S201 is synchronized to the basic data of the sample user in the user attribute library for cluster analysis, and the feature data of the sample user is obtained, wherein the feature data includes the behavior type of the sample user and the number of behaviors corresponding to the behavior type, and the behavior Types include ordering product types and accessing website types.
  • the clustering analysis can be analyzed by calculating the degree of closeness between different basic data units, and the degree of proximity can be expressed by the distance index, that is, the smaller the distance index is, the more similar the data of the corresponding two basic data units are.
  • the distance index can be calculated according to the nature of the data, such as Euclidean distance, Chebychev distance, Chi-Square measure distance. Wait.
  • the feature data of the sample user obtained by the cluster analysis may exist in the form of a feature information table, and the feature information table may be represented by a two-dimensional array, and the first dimension of the two-dimensional array is used to identify the behavior type of the sample user, the second The second dimension of the dimension array is used to identify the number of behaviors for each behavior type.
  • the type of behavior of the sample user may specifically include the type of product in which the sample user subscribes to the specific product, and the type of website on which the sample user visits the website.
  • S203 Acquire feature data of the sample user from the user attribute library.
  • the basic data of the sample user is periodically synchronized to the user attribute library according to steps S201 to S202, and the basic data of the sample user is clustered to obtain the feature data of the sample user, and then the samples are obtained from the user attribute library.
  • User's feature data is obtained from the user attribute library.
  • S204 Preprocess the feature data of the sample user, remove the noise data, and obtain the preprocessed data.
  • step S203 the feature data obtained in step S203 is preprocessed, and the noise data is removed to obtain preprocessed data.
  • This step is the same as the step S102 in the first embodiment, and the same processing result is obtained, and details are not described herein again.
  • S205 Select training sample data and test sample data from the preprocessed data.
  • the pre-processed data is selected as the training sample data according to the first preset ratio, and the pre-processed data is selected as the test sample data according to the second preset ratio.
  • the first preset ratio may be set to 80%
  • the second preset ratio may be set to 20%, that is, 80% of the preprocessed data is used as the training sample data, and the remaining 20% of the preprocessed data is used as the test sample data. It can be understood that the first preset ratio and the second preset ratio can be reasonably set according to the needs of the actual application, and no limitation is made here.
  • S206 Perform an association model between the behavior of the sample user and the preset insurance product in the database according to the training sample data in a preset training period.
  • the preset training period can be set according to the behavior characteristics of the sample user, and the training period can be adjusted according to the training result of the associated model.
  • the corresponding sample user behavior characteristics are different, according to the behavior characteristics of the sample users to determine the corresponding insurance business, the insurance business determines the length of the training cycle.
  • the training period is adjusted according to the training result of the association model, which can be determined according to multiple experiments.
  • the appropriate training period can effectively improve the accuracy of the association model. Since the recent behavior of the sample user can better reflect the next possible behavior, the training period needs to consider the influence of the time attenuation factor.
  • CF Collaborative Filtering
  • the collaborative filtering algorithm includes a User Collaboration Filter (UserCF) algorithm and a Product Collaboration Filter (ItemCF) algorithm.
  • the UserCF algorithm can be used in training the association model, or the ItemCF algorithm can be used, and the UserCF and ItemCF algorithms can also be used at the same time.
  • the user in the collaborative filtering algorithm is the sample user
  • the product is the insurance product preset in the database.
  • the database is the insurance product database of the insurance management system, in which the product information of various insurance products is pre-stored, including the insurance type, the insurance name, the insured data information, the insured object, the insurance coverage, As well as income and so on.
  • the UserCF algorithm can be directly used to establish the association model.
  • the core idea of the UserCF algorithm is that when user A needs personalized recommendation, other users who have similar interests with user A can be found first, and then these other users have already The purchased insurance product or the insurance product of interest is associated with user A.
  • the search can be performed by calculating the similarity degree between the two users.
  • the similarity algorithm can be used, for example, The Log Likelihood Estimate algorithm calculates the similarity of interest between the sample user u and the sample user v.
  • the degree of interest of the sample user u in the insurance product can be calculated by the formula (1).
  • p(u,i) represents the degree of interest of the sample user u for the insurance product i
  • S(u,K) represents the K sample users closest to the interest of the sample user u
  • N(i) represents the insurance product i has a set of sample users who have purchased behavior
  • w uv represents the degree of similarity between the sample user u and the sample user v
  • r vi represents the degree of interest of the sample user v for the insurance product i.
  • the AssociationCF algorithm When the AssociationCF algorithm is used to establish the association model, since the core idea of the ItemCF algorithm is to recommend other products similar to the products that the user is interested in, it is possible to calculate the similarity between different insurance products by analyzing the behavior of the sample users. And storing the similarity between the preset insurance products in the form of the similarity matrix, and analyzing the insurance products associated with the behavior of the sample user according to the similarity matrix and the historical behavior of the sample user.
  • the degree of interest of the sample user u in the insurance product j can be calculated by the formula (2).
  • p(u,j) represents the degree of interest of the sample user u for the insurance product j
  • M(u) represents the set of insurance products of interest to the sample user u
  • T(i,K) represents the most similar to the insurance product i K insurance product sets
  • w ji represent the similarity between the insurance product j and the insurance product i
  • r ui represents the degree of interest of the sample user u for the insurance product i.
  • the sample user u is interested in different insurance products, and selects the insurance products whose interest degree exceeds the interest threshold, and establishes the behavior of the sample user u and the selected ones.
  • the association model between the behavior of the training sample user and the preset insurance product in the database may also be implemented by the following steps S2061 to S2063, which are described in detail as follows:
  • S2061 Analyze the network behavior of the sample user according to the training sample data in a preset training period, and determine an associated product corresponding to the network behavior.
  • the UserCF algorithm based on the above formula (1) or the ItemCF algorithm of the formula (2) is used to determine the insurance product of interest to the sample user, and the sample is combined with the sample.
  • the user's behavior type and the number of behaviors corresponding to the behavior type are summarized and analyzed to determine the associated product corresponding to the network behavior of the sample user.
  • the insurance product that the sample user searches in the training period but does not purchase is determined, and the sample user is analyzed for the insurance product in combination with the number of searches of the insurance product by the sample user.
  • the degree of interest, and the insurance product whose degree of interest reaches the interest threshold is determined as the associated product corresponding to the web search behavior.
  • S2062 Perform weighted training on the network behavior of the sample user and the associated product corresponding to the network behavior, and obtain an association model between the behavior of the sample user and the preset insurance product in the database.
  • test sample data determined in step S205 the test sample data is brought into the correlation model determined in step S206 for testing to verify whether the accuracy of the associated model reaches a preset test requirement.
  • the association model is used as a learning model, and the learning model is used to match the behavior of the first type of user with the preset insurance product in the database, and the first type of user is an insurance customer. .
  • step S207 if the test result of step S207 meets the preset test requirement, the association model determined in step S206 is used as a learning model, and the learning model is used to match the behavior of the first type of user with the preset insurance product in the database. .
  • the first type of user is an insurance customer, that is, a potential customer who has purchased insurance intentions and an existing customer who has purchased insurance.
  • the database is specifically the insurance product database of the insurance management system, in which the product information of various insurance products is pre-stored, including insurance type, insurance name, insurance amount data information, insurance object, insurance coverage, and income.
  • test requirements can be set according to the needs of the actual application, and there is no restriction here.
  • test requirements can be set to use test sample data to test the associated model with a correct rate of 90%.
  • step S207 If the test result of step S207 does not meet the preset test requirements, the associated model is continuously trained and tested until the preset test requirements are met.
  • training sample data and the test sample data are data combined with the insurance business, and the data is associated by the user's telephone number or identification number and the like, thereby ensuring user consistency.
  • S209 Based on the learning model, predicting behavior of the first type of user to be predicted, and obtaining a target insurance product of the first type of user.
  • step S208 behavior prediction is performed on the first type of user to be predicted, and the target insurance product of the first type of user is obtained.
  • This step is the same as the step S104 in the first embodiment, and the same processing result is obtained, and details are not described herein again.
  • S210 Push the first type of user to be predicted to the second type of user, so that the second type of user determines the first type of user as the target user of the target insurance product, and the second type of user is the insurance business personnel.
  • this step is the same as the step S105 in the first embodiment, and the same processing result is obtained, and details are not described herein again.
  • the feature data of the sample user is obtained, and then the sample user is obtained from the user attribute library.
  • the feature data is subjected to denoising processing to obtain preprocessed data.
  • the targeted pre-processed data is obtained, so as to ensure that the subsequent analysis and modeling of the feature data can be more accurate and improve the prediction accuracy. rate.
  • the training sample data and the test sample data are selected from the pre-processed data, and the behavior of the sample user is trained according to the user-based collaborative filtering algorithm or the product-based collaborative filtering algorithm according to the training sample data in a preset training period.
  • the association model between the preset insurance products in the database, and the test model data is used to test the association model. If the test result satisfies the preset test requirements, the association model is used as the learning model, and the prediction model is implemented based on the learning model.
  • the first type of user performs behavior prediction, obtains the target insurance product of the first type of user, and automatically and accurately finds the relationship between the user behavior and the insurance product by using the CF algorithm to establish the learning model, thereby achieving the first Accurate prediction of insurance products of type interest, improve the prediction efficiency and accuracy of user insurance behavior, and intelligent prediction level, and push the first type of users to the second type of users, so that the second type of users can understand in time
  • the potential for high probability of purchasing insurance products Households thereby targeted to keep track of and improve sales success rate and efficiency of insurance products.
  • Embodiment 3 is a diagrammatic representation of Embodiment 3
  • the apparatus for predicting the user's insurance behavior as illustrated in FIG. 4 may be the execution subject of the method for predicting the user's insurance behavior provided by the foregoing first embodiment.
  • the apparatus for predicting the user's insurance behavior as illustrated in FIG. 4 includes: an obtaining module 31, a pre-processing module 32, a modeling module 33, a prediction module 34, and a pushing module 35, and each functional module is described in detail as follows:
  • the obtaining module 31 is configured to obtain feature data of the sample user from the user attribute database, where the feature data includes a behavior type of the sample user and a behavior number corresponding to the behavior type, and the behavior type includes a subscription product type and a visiting website type;
  • the pre-processing module 32 is configured to perform pre-processing on the feature data acquired by the obtaining module 31, remove the noise data, and obtain pre-processed data;
  • the modeling module 33 is configured to establish a learning model according to the pre-processed data obtained by the pre-processing module 32, wherein the learning model is configured to match the behavior of the first type of user with a preset insurance product in the database, the first type The user is an insurance customer;
  • the prediction module 34 is configured to perform behavior prediction on the first type of user to be predicted based on the learning model established by the modeling module 33, to obtain a target insurance product of the first type of user;
  • the pushing module 35 is configured to push the first type of user to the second type of user, so that the second type of user determines the first type of user as the target user of the target insurance product, wherein the second type of user is the insurance business personnel .
  • the apparatus for predicting the behavior of the user's insurance behavior as exemplified in the above FIG. 4 is known.
  • the feature data of the sample user is obtained from the user attribute library, and the pre-processing data is obtained by denoising the feature data, and according to The pre-processed data establishes a learning model, and based on the learning model, the behavior of the first type of user to be predicted is predicted, and the target insurance product of the first type of user is obtained, and the user behavior can be automatically and accurately found by establishing a learning model.
  • the relationship between insurance products thereby achieving accurate prediction of insurance products of interest to the first type of users, improving the predictive efficiency and accuracy of the user's insurance behavior, and the level of intelligent prediction, while pushing the first type of users to the first
  • the second type of user enables the second type of user to know the potential users who have high probability of purchasing the insurance product in time, so as to carry out targeted continuous tracking and improve the sales success rate and efficiency of the insurance product.
  • Embodiment 4 is a diagrammatic representation of Embodiment 4:
  • FIG. 5 is a schematic structural diagram of an apparatus for predicting user insurance behavior provided by Embodiment 4 of the present application. For convenience of description, only parts related to the embodiment of the present application are shown.
  • the apparatus for predicting user insurance behavior behavior illustrated in FIG. 5 may be the execution body of the method for predicting user insurance behavior provided by the foregoing second embodiment.
  • the apparatus for predicting user insurance behavior as illustrated in FIG. 5 includes: an obtaining module 41, a preprocessing module 42, a modeling module 43, a prediction module 44, and a pushing module 45, and each functional module is described in detail as follows:
  • the obtaining module 41 is configured to obtain feature data of the sample user from the user attribute database, where the feature data includes a behavior type of the sample user and a behavior number corresponding to the behavior type, and the behavior type includes a subscription product type and a visiting website type;
  • the pre-processing module 42 is configured to perform pre-processing on the feature data acquired by the obtaining module 41, remove the noise data, and obtain pre-processed data;
  • the modeling module 43 is configured to establish a learning model according to the pre-processed data obtained by the pre-processing module 42, wherein the learning model is configured to match the behavior of the first type of user with a preset insurance product in the database, the first type The user is an insurance customer;
  • the prediction module 44 is configured to perform behavior prediction on the first type of user to be predicted based on the learning model established by the modeling module 43 to obtain a target insurance product of the first type of user;
  • the pushing module 45 is configured to push the first type of user to the second type of user, so that the second type of user determines the first type of user as the target user of the target insurance product, wherein the second type of user is the insurance business personnel. .
  • the device further includes:
  • the synchronization module 46 is configured to periodically synchronize the basic data of the sample user to the user attribute database, where the basic data includes network access data and policy data in the insurance management system, and the network access data includes product ordering information and website access information;
  • the clustering module 47 is configured to perform cluster analysis on the basic data in the user attribute database to obtain feature data of the sample user.
  • modeling module 43 includes:
  • a selection submodule 431, configured to select training sample data and test sample data from the preprocessed data obtained by the preprocessing module 42;
  • the training sub-module 432 is configured to train an association model between the behavior of the sample user and the preset insurance product in the database according to the training sample data determined by the selection sub-module 431 within a preset training period;
  • test sub-module 433 configured to test the association model determined by the training sub-module 432 using the test sample data determined by the selection sub-module 431;
  • the determining sub-module 434 is configured to use the correlation model determined by the training sub-module 432 as a learning model if the test result obtained by the test sub-module 433 meets the preset test requirement.
  • training sub-module 432 is further configured to:
  • the collaborative filtering algorithm is used to train the association model between the behavior of the sample user and the preset insurance product in the database.
  • training sub-module 434 is further configured to:
  • the network behavior of the sample user is analyzed according to the training sample data determined by the selection sub-module 431 in a preset training period, and the associated product corresponding to the network behavior is determined;
  • the network behavior and related products are weighted and trained to obtain the association model between the behavior of the sample user and the preset insurance products in the database.
  • the device for predicting the behavior of the user's insurance behavior as illustrated in FIG. 5 above that, in this embodiment, first, by periodically synchronizing the basic data of the sample user to the user attribute database, and performing cluster analysis on the basic data of the sample user, The feature data of the sample user is then obtained from the user attribute library to obtain the feature data of the sample user, and the feature data is denoised to obtain the preprocessed data.
  • the targeted pre-processed data is obtained, so as to ensure that the subsequent analysis and modeling of the feature data can be more accurate and improve the prediction accuracy. rate.
  • the training sample data and the test sample data are selected from the pre-processed data, and the behavior of the sample user is trained according to the user-based collaborative filtering algorithm or the product-based collaborative filtering algorithm according to the training sample data in a preset training period.
  • the association model between the preset insurance products in the database, and the test model data is used to test the association model. If the test result satisfies the preset test requirements, the association model is used as the learning model, and the prediction model is implemented based on the learning model.
  • the first type of user performs behavior prediction, obtains the target insurance product of the first type of user, and automatically and accurately finds the relationship between the user behavior and the insurance product by using the CF algorithm to establish the learning model, thereby achieving the first Accurate prediction of insurance products of type interest, improve the prediction efficiency and accuracy of user insurance behavior, and intelligent prediction level, and push the first type of users to the second type of users, so that the second type of users can understand in time
  • the potential for high probability of purchasing insurance products Households thereby targeted to keep track of and improve sales success rate and efficiency of insurance products.
  • Embodiment 5 is a diagrammatic representation of Embodiment 5:
  • FIG. 6 is a schematic diagram of a computing device for predicting user insurance behavior provided by Embodiment 5 of the present application.
  • the computing device 6 of this embodiment includes a processor 60 and a memory 61 in which computer readable instructions 62 executable on the processor 60, such as user insurance behavior predictions, are stored. program of.
  • the steps in the method embodiment of implementing the above-described prediction of each user's insurance behavior when the processor 60 executes the computer readable instructions 62 such as steps S101 to S105 shown in FIG.
  • the functions of the modules/units in the apparatus embodiment for implementing the above-described prediction of each user's insurance behavior when the processor 60 executes the computer readable instructions 62 such as the functions of the modules 31 to 35 shown in FIG.
  • the computer readable instructions 62 may be partitioned into one or more modules/units that are stored in the memory 61 and executed by the processor 60, To complete this application.
  • the one or more modules/units may be a series of computer readable instruction segments capable of performing a particular function for describing the execution of the computer readable instructions 62 in the computing device 6.
  • the computing device 6 can be a computing device such as a desktop computer, a notebook, a palmtop computer, and a cloud server.
  • the computing device can include, but is not limited to, processor 60 and memory 61. It will be understood by those skilled in the art that FIG. 6 is merely an example of the computing device 6 for user insurance behavior prediction, and does not constitute a limitation of the computing device 6 for predicting the user's insurance behavior, and may include more or fewer components than illustrated. Alternatively, some components may be combined, or different components, such as the computing device, which may also include input and output devices, network access devices, buses, and the like.
  • the so-called processor 60 may be a central processing unit (CPU), or may be other general-purpose processors, a digital signal processor (DSP), an application specific integrated circuit (ASIC), Field-Programmable Gate Array (FPGA) or other programmable logic device, discrete gate or transistor logic device, discrete hardware components, etc.
  • the general purpose processor may be a microprocessor or the processor or any conventional processor or the like.
  • the memory 61 may be an internal storage unit of the computing device 6, such as a hard disk or memory of the computing device 6.
  • the memory 61 may also be an external storage device of the computing device 6, such as a plug-in hard disk equipped on the computing device 6, a smart memory card (SMC), and a secure digital (SD). Card, flash card, etc. Further, the memory 61 may also include both an internal storage unit of the computing device 6 and an external storage device.
  • the memory 61 is for storing the computer readable instructions and other programs and data required by the computing device.
  • the memory 61 can also be used to temporarily store data that has been output or is about to be output.
  • each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.
  • the above integrated unit can be implemented in the form of hardware or in the form of a software functional unit.
  • the integrated unit if implemented in the form of a software functional unit and sold or used as a standalone product, may be stored in a computer readable storage medium.
  • a computer readable storage medium A number of instructions are included to cause a computer device (which may be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present application.
  • the foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, and the like, which can store program codes. .

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Databases & Information Systems (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Accounting & Taxation (AREA)
  • Finance (AREA)
  • General Physics & Mathematics (AREA)
  • Development Economics (AREA)
  • Technology Law (AREA)
  • General Business, Economics & Management (AREA)
  • Strategic Management (AREA)
  • Marketing (AREA)
  • Economics (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • Financial Or Insurance-Related Operations Such As Payment And Settlement (AREA)

Abstract

本方案提供了一种用户投保行为预测的方法、装置、计算设备及介质，适用于互联网技术领域，该方法包括：从用户属性库中获取样本用户的特征数据；对特征数据进行预处理，去除噪声数据，得到预处理数据；根据预处理数据，建立学习模型；基于该学习模型，对待预测的第一类型用户进行行为预测，得到该第一类型用户的目标保险产品；将该第一类型用户推送给第二类型用户，以使第二类型用户将该第一类型用户确定为目标保险产品的目标用户。本方案实现了对用户感兴趣的保险产品的准确预测，提高对用户投保行为的预测效率和准确率，以及智能预测水平，同时使保险业务人员能够及时了解购买保险产品概率高的潜在用户，提高保险产品的销售成功率和效率。

Description

用户投保行为预测的方法、装置、计算设备及介质
本申请要求于2017年06月09日提交中国专利局、申请号为201710434568.8、发明名称为“一种用户投保行为预测的方法及装置”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请属于互联网技术领域，尤其涉及一种用户投保行为预测的方法、装置、计算设备及介质。
背景技术
目前，保险业务人员在开发客户时，往往需要经过与客户进行深入沟通，或者通过查阅客户购买保险的历史记录，才能了解到客户的投保需求或者投保意愿，并在此基础上进一步通过人工分析，确定客户可能感兴趣的保险产品，这种方式导致工作效率低且准确率不高，并且对于陌生客户则无法进行有针对性的保险产品的推荐，导致保险产品的成交量无法有效提高。
技术问题
有鉴于此，本申请实施例提供了一种用户投保行为预测的方法、装置、计算设备及介质，以解决现有技术中通过保险业务人员主观地分析用户可能感兴趣的保险产品，导致对用户投保行为预测效率和准确率低的问题。
技术解决方案
本申请实施例的第一方面提供了一种用户投保行为预测的方法，包括：
从用户属性库中获取样本用户的特征数据，其中，所述特征数据包括所述样本用户的行为类型和所述行为类型对应的行为次数，所述行为类型包括订购产品类型和访问网站类型；
对所述特征数据进行预处理，去除噪声数据，得到预处理数据；
根据所述预处理数据，建立学习模型，所述学习模型用于将第一类型用户的行为与数据库中预设的保险产品进行匹配，所述第一类型用户为保险客户；
基于所述学习模型，对待预测的所述第一类型用户进行行为预测，得到该第一类型用户的目标保险产品；
将该第一类型用户推送给第二类型用户，以使所述第二类型用户将该第一类型用户确 定为所述目标保险产品的目标用户，所述第二类型用户为保险业务人员。
本申请实施例的第二方面提供了一种用户投保行为预测的装置，包括：
获取模块，用于从用户属性库中获取样本用户的特征数据，其中，特征数据包括样本用户的行为类型和该行为类型对应的行为次数，行为类型包括订购产品类型和访问网站类型；
预处理模块，用于对获取模块获取的特征数据进行预处理，去除噪声数据，得到预处理数据；
建模模块，用于根据预处理模块得到的预处理数据，建立学习模型，其中，该学习模型用于将第一类型用户的行为与数据库中预设的保险产品进行匹配，第一类型用户为保险客户；
预测模块，用于基于建模模块建立的学习模型，对待预测的第一类型用户进行行为预测，得到该第一类型用户的目标保险产品；
推送模块，用于将该第一类型用户推送给第二类型用户，以使第二类型用户将该第一类型用户确定为目标保险产品的目标用户，其中，第二类型用户为保险业务人员。
本申请实施例的第三方面提供了一种用户投保行为预测的计算设备，包括存储器以及处理器，所述存储器中存储有可在所述处理器上运行的计算机可读指令，所述处理器执行所述计算机可读指令时实现如第一方面所述的用户投保行为预测的方法的步骤。
本申请实施例的第四方面提供了一种计算机可读存储介质，所述计算机可读存储介质存储有计算机可读指令，所述计算机可读指令被处理器执行时实现如第一方面所述的用户投保行为预测的方法的步骤。
有益效果
本申请实施例中，通过从用户属性库中获取样本用户的特征数据，对该特征数据进行去噪处理后得到预处理数据，并根据该预处理数据建立学习模型，基于该学习模型实现对待预测的第一类型用户进行行为预测，得到该第一类型用户的目标保险产品，通过建立学习模型的方式能够自动且准确地找到用户行为与保险产品之间的关系，从而实现对第一类型用户感兴趣的保险产品的准确预测，提高对用户投保行为的预测效率和准确率，以及智能预测水平，同时将该第一类型用户推送给第二类型用户，使得第二类型用户能够及时了解购买保险产品概率高的潜在用户，从而进行有针对性的持续跟踪，提高保险产品的销售成功率和效率。
附图说明
图1是本申请实施例一提供的一种用户投保行为预测的方法的实现流程示意图；
图2是本申请实施例二提供的一种用户投保行为预测的方法的实现流程示意图；
图3是本申请实施例二提供的一种用户投保行为预测的方法中将基础数据同步到用户属性库的流程示意图；
图4是本申请实施例三提供的一种用户投保行为预测的装置的示意图；
图5是本申请实施例四提供的一种用户投保行为预测的装置的示意图；
图6是本申请实施例五提供的一种用户投保行为预测的计算设备的示意图。
本发明的实施方式
以下描述中，为了说明而不是为了限定，提出了诸如特定系统结构、技术之类的具体细节，以便透彻理解本申请实施例。然而，本领域的技术人员应当清楚，在没有这些具体细节的其它实施例中也可以实现本申请。在其它情况中，省略对众所周知的系统、装置、电路以及方法的详细说明，以免不必要的细节妨碍本申请的描述。
为了说明本申请所述的技术方案，下面通过具体实施例来进行说明。
实施例一：
图1是本申请实施例一提供的一种用户投保行为预测的方法的流程图，本申请实施例的执行主体为计算设备，其具体可以是服务器等，图1示例的一种用户投保行为预测的方法具体可以包括步骤S101至步骤S104，详述如下：
S101：从用户属性库中获取样本用户的特征数据，其中，特征数据包括样本用户的行为类型和该行为类型对应的行为次数，行为类型包括订购产品类型和访问网站类型。
具体地，用户属性库可以是大数据平台，用户属性库包含样本用户的特征数据，该特征数据包括样本用户的行为类型和每种行为类型对应的行为次数，行为类型可以包括订购产品类型和访问网站类型。
样本用户可以是通过各种行为接口进行行为动作的用户，其中，行为接口包括但不限于网站、智能终端的应用(Application，APP)，以及实体柜台或者合作公司这样的实体渠道，以上行为接口所分别对应的行为动作包括用户对网站进行访问、通过APP进行操作，以及通过实体渠道办理业务。
每个样本用户的特征数据可以以特征信息表的形式存在，特征信息表可以采用二维数组表示，该二维数组的第一维用于标识样本用户的行为类型，该二维数组的第二维用于标识每个行为类型对应的行为次数。
样本用户的行为类型具体可以包括样本用户订购具体产品的产品类型，该产品类型可以是样本用户通过网络购买的产品类型，例如旅游产品，以及机票等交通运输产品等，该产品类型还可以是样本用户通过实体渠道购买的保险产品等。
样本用户的行为类型还可以包括样本用户访问网站的网站类型，该网站类型可以是购物类网站、新闻时事类网站等。
S102：对样本用户的特征数据进行预处理，去除噪声数据，得到预处理数据。
具体地，对步骤S101得到的特征数据进行预处理，根据特征数据中样本用户的不同行为类型对应的业务进行差异化处理，清除作弊、刷单、误操作等噪声数据，得到去噪后的预处理数据。
由于样本用户的特征数据是在样本用户的各种网络行为或实体渠道行为过程中产生的，可能存在大量的噪声数据，具体可以通过常用的数据挖掘算法，例如聚类算法过滤掉特征数据中的噪声，从而使得后续对特征数据的分析和建模能够更加精确。
S103：根据预处理数据，建立学习模型，该学习模型用于将第一类型用户的行为与数据库中预设的保险产品进行匹配，第一类型用户为保险客户。
具体地，根据步骤S102得到的预处理数据进行建模，采用模型训练算法建立学习模型，例如通过大数据hadoop学习算法建立用户的行为与具有购买意向的保险产品之间的学习模型，该学习模型用于将保险客户的行为与数据库中预设的保险产品进行匹配。
保险客户为具有购买保险意向的潜在客户和已经购买过保险的现有客户。数据库可以是保险管理系统的保险产品数据库，在该数据库中预存了各种保险产品的产品信息，包括保险类型、保险名称、保额数据信息、参保对象、保险范围、以及收益等。
S104：基于学习模型，对待预测的第一类型用户进行行为预测，得到该第一类型用户的目标保险产品。
具体地，基于步骤S103建立的学习模型，对某一个具体的待预测的保险客户进行行为预测，即将待预测的保险客户的行为类型输入该学习模型，通过该学习模型的计算和分析，得到该待预测的保险客户感兴趣的目标保险产品。
目标保险产品可以通过网络推送或者其他推送方式被推荐给待预测的保险客户。例如，当保险客户发生购买旅游产品的行为，或者发生购买火车票、机票等交通运输产品的行为时，根据学习模型可以得到与该保险客户的行为相匹配的目标保险产品，例如旅游安全险、航空延误险等，并且将这些目标保险产品推送给待预测的保险客户。
S105：将待预测的第一类型用户推送给第二类型用户，以使第二类型用户将该第一类型用户确定为目标保险产品的目标用户，第二类型用户为保险业务人员。
具体地，将待预测的保险客户的用户信息推送给保险业务人员，该用户信息可以包括保险客户的姓名、联系方式、可能感兴趣的保险产品等，保险业务人员可以将该保险客户确定为购买目标保险产品的目标用户，对该保险用户进行有针对性的持续跟踪，使得保险业务 人员不再盲目的选择目标用户，从而提高保险产品的销售成功率和效率。
本实施例中，通过从用户属性库中获取样本用户的特征数据，对该特征数据进行去噪处理后得到预处理数据，并根据该预处理数据建立学习模型，基于该学习模型实现对待预测的第一类型用户进行行为预测，得到该第一类型用户的目标保险产品，通过建立学习模型的方式能够自动且准确地找到用户行为与保险产品之间的关系，从而实现对第一类型用户感兴趣的保险产品的准确预测，提高对用户投保行为的预测效率和准确率，以及智能预测水平，同时将该第一类型用户推送给第二类型用户，使得第二类型用户能够及时了解购买保险产品概率高的潜在用户，从而进行有针对性的持续跟踪，提高保险产品的销售成功率和效率。
实施例二：
图2是本申请实施例二提供的一种用户投保行为预测的方法的流程图，本申请实施例的执行主体为计算设备，其具体可以是服务器等，图2示例的一种用户投保行为预测的方法具体可以包括步骤S201至步骤S210，详述如下：
S201：定期将样本用户的基础数据同步到用户属性库，其中，样本用户的基础数据包括网络访问数据和保险管理系统中的保单数据，网络访问数据包括产品订购信息和网站访问信息。
具体地，定期获取样本用户的基础数据，并将该基础数据同步到用户属性库的过程如图3所示，其具体可以通过步骤S2011至步骤S2019完成，详细说明如下：
S2011：获取样本用户的基础数据。
样本用户可以是通过各种行为接口进行行为动作的用户。不同行为动作的样本用户对应不同的基础数据。
其中，对网站进行访问、通过APP进行操作的样本用户，其基础数据包括网络访问数据，网络访问数据可以通过对样本用户的网站浏览记录、网络操作记录等进行分析，得到网站访问信息，该网站访问信息具体可以包括访问内容的类型、所关注或者购买的产品信息，以及访问次数等。
通过实体渠道办理业务的样本用户，其基础数据包括保险管理系统中的保单数据，保单数据可以包括购买保险产品的用户信息，具体购买的保险产品信息，购买时间等。
S2012：判断导入汇总时间段内基础数据的数据游标是否执行完成，若未执行完成，则继续执行步骤S2013，否则执行步骤S2014。
具体地，导入汇总时间段是指定期将样本用户的基础数据同步到用户属性库的定期时间段，例如该时间段可以是每10分钟，或者每30分钟等。当用户通过各种行为接口进行频繁的行为动作时，该时间段可以设置相对较短的时间长度，当用户的行为动作不是很频繁时， 该时间段可以是设置相对较长的时间长度。导入汇总时间段的具体时长可以根据实际应用的需要进行设置，此处不做限制。
游标充当指针的作用，是一种能从包括多条数据记录的结果集中每次提取一条记录的机制。这里，基础数据的数据游标是指在步骤S2011得到的样本用户的基础数据的集合中，每次提取一条基础数据。若数据游标执行完成，则说明步骤S2011得到的样本用户的基础数据已经插入到临时表中，流程跳转到步骤S2014；若数据游标未执行完成，则继续执行步骤S2013，提取下一条待同步的基础数据进行插入临时表的操作。
S2013：将获取到的基础数据插入到临时表中，并返回步骤S2012。
具体地，将数据游标当前指向的基础数据插入到临时表中，并返回步骤S2012，继续通过数据游标对基础数据的集合中的下一条待同步的基础数据进行处理。
S2014：判断对基础数据进行分析用的分析游标是否执行完成，若未执行完成，则继续执行步骤S2015，否则该同步流程处理完成，则流程跳转到步骤S2019。
具体地，当导入汇总时间段内基础数据的数据游标执行完成时，需要同步到用户属性库中的基础数据已经全部插入到临时表中，在该临时表中通过分析游标每次提取该临时表中的一条基础数据进行同步流程处理。若分析游标执行完成，则同步流程处理完成，流程跳转到步骤S2019；若分析游标未执行完成，则继续执行步骤S2015，提取下一条待同步的基础数据进行同步处理。
S2015：对基础数据进行分析和汇总操作，得到待同步数据和对应的数据记录表。
具体地，若临时表中对基础数据进行分析用的分析游标未执行完成，则对分析游标当前指向的基础数据进行分析和汇总操作，得到同步数据和对应的数据记录表。
可以理解的是，根据基础数据的特征，该待同步数据可以包括样本用户的访问记录数据、业务特征数据，常用链接数据等，并且每种数据类型的数据均可以通过数据记录表的方式进行保存，一种数据类型的数据保存至一份数据记录表中。
S2016：在当前的用户属性库中，判断待同步数据对应的数据记录表中是否已经存在待同步数据的记录，若已经存在则执行步骤S2018，否则执行步骤S2017。
具体地，用户属性库可以是大数据平台，在当前的用户属性库的数据记录表中，对每个待同步数据只记录该类型数据的最新值，因此需判断待同步数据对应的数据记录表中是否已经存在该待同步数据的记录，若已经存在则执行步骤S2018，否则执行步骤S2017。
S2017：将待同步的数据插入到指定的数据记录表中，并跳转到步骤S2019。
具体地，若待同步数据对应的数据记录表中尚不存在该待同步数据的记录，则将该待同步的数据插入到对应的数据记录表中。完成插入操作后流程跳转到步骤S2019。
S2018：将待同步的数据更新到对应的数据记录表中的对应记录中。
具体地，若待同步数据对应的数据记录表中已经存在该待同步数据的记录，则将该待同步数据更新到已经存在的对应记录中。
S2019：流程结束。
S202：对样本用户的基础数据进行聚类分析，得到样本用户的特征数据。
具体地，将步骤S201同步到用户属性库中的样本用户的基础数据进行聚类分析，得到样本用户的特征数据，其中，特征数据包括样本用户的行为类型和该行为类型对应的行为次数，行为类型包括订购产品类型和访问网站类型。
聚类分析可以通过计算不同基础数据单元之间的接近程度进行分析，该接近程度可以用距离指标表示，即距离指标越小表示对应的两个基础数据单元的数据越具有相似性。计算距离指标的方法有很多，按照数据性质的不同，可以选用不同的距离指标算法，如欧氏距离(Euclidean distance)、切比雪夫距离(Chebychev distance)、卡方距离(Chi-Square measure distance)等。
通过聚类分析后得到的样本用户的特征数据可以以特征信息表的形式存在，特征信息表可以采用二维数组表示，该二维数组的第一维用于标识样本用户的行为类型，该二维数组的第二维用于标识每个行为类型对应的行为次数。样本用户的行为类型具体可以包括样本用户订购具体产品的产品类型，以及样本用户访问网站的网站类型。
S203：从用户属性库中获取样本用户的特征数据。
具体地，根据步骤S201至步骤S202定期将样本用户的基础数据同步到用户属性库，并对样本用户的基础数据进行聚类分析，得到样本用户的特征数据之后，从用户属性库中获取这些样本用户的特征数据。
S204：对样本用户的特征数据进行预处理，去除噪声数据，得到预处理数据。
具体地，对步骤S203得到的特征数据进行预处理，去除噪声数据，得到预处理数据。
本步骤与实施例一中的步骤S102采用完全相同的处理过程，并得到相同的处理结果，此处不再赘述。
S205：从预处理数据中选择训练样本数据和测试样本数据。
具体地，从步骤S204得到的预处理数据中，按照第一预设比例选择预处理数据作为训练样本数据，按照第二预设比例选择预处理数据作为测试样本数据。
例如，第一预设比例可以设置为80％，第二预设比例可以设置为20％，即将80％的预处理数据作为训练样本数据，将剩下20％的预处理数据作为测试样本数据。可以理解的是，第一预设比例和第二预设比例可以根据实际应用的需要进行合理设置，此处不做限制。
S206：在预设的训练周期内根据训练样本数据，训练样本用户的行为与数据库中预设的保险产品之间的关联模型。
预设的训练周期可以依据样本用户的行为特征进行设置，并且该训练周期可以根据关联模型的训练结果进行调整。
不同投保业务，其对应的样本用户行为特征不同，根据样本用户的行为特征确定对应的投保业务，由投保业务决定训练周期的长短。根据关联模型的训练结果对训练周期进行调整，具体可以根据多次实验来确定。
合适的训练周期能够有效提升关联模型的准确度，由于样本用户在近期的行为更加能反映出接下来可能的行为动作，因此训练周期的设置还需要考虑时间衰减因素的影响。
具体地，在预设的训练周期内根据步骤S205确定的训练样本数据，采用协同过滤(Collaborative Filtering，CF)算法，训练样本用户的行为与数据库中预设的保险产品之间的关联模型。
协同过滤算法包括基于用户的协同过滤(User CollaborationFilter，UserCF)算法和基于产品的协同过滤(ItemCollaborationFilter，ItemCF)算法。在训练关联模型时可以采用UserCF算法，或者采用ItemCF算法，还可以同时采用UserCF和ItemCF算法。
可以理解的是，协同过滤算法中的用户即样本用户，产品即数据库中预设的保险产品。在本申请实施例中，数据库即保险管理系统的保险产品数据库，在该数据库中预存了各种保险产品的产品信息，包括保险类型、保险名称、保额数据信息、参保对象、保险范围、以及收益等。
当样本用户的数量较小时，可以直接采用UserCF算法建立关联模型，UserCF算法核心思想是当用户A需要个性化推荐时，可以先找到和用户A有相似兴趣的其他用户，然后把这些其他用户已经购买的保险产品或者感兴趣的保险产品与用户A进行关联匹配。
在查找和用户A有相似兴趣的其他用户时，可以通过计算两个用户之间的兴趣相似度的方法进行查找，例如，给定样本用户u和样本用户v，可以通过相似度算法，例如对数似然估计(Log Likelihood Estimate)算法，计算样本用户u和样本用户v之间的兴趣相似度。
具体可以通过公式(1)计算样本用户u对保险产品的感兴趣程度。
Figure PCTCN2018074884-appb-000001
其中，p(u,i)表示样本用户u对保险产品i的感兴趣程度，S(u,K)表示和样本用户u的兴趣最接近的K个样本用户，N(i)表示对保险产品i有过购买行为的样本用户的集合，w uv表 示样本用户u和样本用户v之间的兴趣相似度，r vi表示样本用户v对保险产品i的感兴趣程度。
当采用ItemCF算法建立关联模型时，由于ItemCF算法的核心思想是给用户推荐和该用户感兴趣的产品类似的其他产品，因此可以通过分析样本用户的行为，计算不同保险产品之间的相似度，并且通过相似度矩阵的形式保存数据中预设的保险产品之间的相似度，根据该相似度矩阵和样本用户的历史行为分析与该样本用户的行为相关联的保险产品。
具体可以通过公式(2)计算样本用户u对保险产品j的感兴趣程度。
Figure PCTCN2018074884-appb-000002
其中，p(u,j)表示样本用户u对保险产品j的感兴趣程度，M(u)表示样本用户u感兴趣的保险产品集合，T(i,K)表示和保险产品i最相似的K个保险产品集合，j∈T(i,K)，w ji表示保险产品j和保险产品i之间的相似度，r ui表示样本用户u对保险产品i的感兴趣程度。
根据上述公式(1)或公式(2)计算出的样本用户u对不同保险产品的感兴趣程度，从其中选取感兴趣程度超过兴趣阈值的保险产品，建立样本用户u的行为与选取出的这些保险产品之间的关联关系，进而得到样本用户的行为与数据库中预设的保险产品之间的关联模型。
进一步地，在预设的训练周期内根据训练样本数据，训练样本用户的行为与数据库中预设的保险产品之间的关联模型还可以通过如下步骤S2061至步骤S2063实现，详细说明如下：
S2061：在预设的训练周期内根据训练样本数据，对样本用户的网络行为进行分析，确定网络行为对应的关联产品。
具体地，在预设的训练周期内根据步骤S205确定的训练样本数据，采用基于上述公式(1)的UserCF算法或者公式(2)的ItemCF算法，确定样本用户感兴趣的保险产品，并结合样本用户的行为类型和该行为类型对应的行为次数进行归纳和分析，确定样本用户的网络行为对应的关联产品。
例如，通过对样本用户所访问的网站类型信息进行分析，确定样本用户在训练周期内搜索却没有购买的保险产品，同时，结合样本用户对该保险产品的搜索次数，分析样本用户对该保险产品的感兴趣的程度，并将感兴趣程度达到兴趣阈值的保险产品确定为该网络搜索行为对应的关联产品。
S2062：对样本用户的网络行为和网络行为对应的关联产品进行加权训练，得到样本用户的行为与数据库中预设的保险产品之间的关联模型。
具体地，对网络行为和其对应的关联产品进行加权计算，根据训练样本数据中的网络 行为以及每个网络行为对应的关联产品的出现频次，确定网络行为的权重参数以及该网络行为对应的关联产品的权重值，出现频次高的网络行为或者关联产品，其对应的权重值就大，反之出现频次低的网络行为或者关联产品，其对应的权重值就小。通过对训练样本数据中每种网络行为及其对应的关联产品进行加权计算，根据计算得到的综合权重值，将综合权重值超过预设权重阈值的网络行为及其对应的关联产品提取出来，建立网络行为与关联产品之间的关联模型，即样本用户的行为与数据库中预设的保险产品之间的关联模型。
S207：使用测试样本数据对关联模型进行测试。
具体地，根据步骤S205确定的测试样本数据，将该测试样本数据带入步骤S206确定的关联模型中进行测试，验证关联模型的准确率是否达到预设的测试要求。
S208：若测试结果满足预设的测试要求，则将关联模型作为学习模型，该学习模型用于将第一类型用户的行为与数据库中预设的保险产品进行匹配，第一类型用户为保险客户。
具体地，若步骤S207的测试结果满足预设的测试要求，则将步骤S206确定的关联模型作为学习模型，该学习模型用于将第一类型用户的行为与数据库中预设的保险产品进行匹配。
第一类型用户为保险客户，即具有购买保险意向的潜在客户和已经购买过保险的现有客户。数据库具体为保险管理系统的保险产品数据库，在该数据库中预存了各种保险产品的产品信息，包括保险类型、保险名称、保额数据信息、参保对象、保险范围、以及收益等。
预设的测试要求可以根据实际应用的需要进行设置，此处不做限制。例如，测试要求可以设置为使用测试样本数据对关联模型进行测试的测试结果的正确率达到90％。
若步骤S207的测试结果不满足预设的测试要求，则继续对关联模型进行训练和测试，直到满足预设的测试要求为止。
需要说明的是，训练样本数据和测试样本数据均是与保险业务结合的数据，其通过用户的电话号码或者身份证号码等标识信息进行数据关联，从而保证用户的一致性。
S209：基于学习模型，对待预测的第一类型用户进行行为预测，得到该第一类型用户的目标保险产品。
具体地，基于步骤S208确定的学习模型，对待预测的第一类型用户进行行为预测，得到该第一类型用户的目标保险产品。
本步骤与实施例一中的步骤S104采用完全相同的处理过程，并得到相同的处理结果，此处不再赘述。
S210：将待预测的第一类型用户推送给第二类型用户，以使第二类型用户将该第一类型用户确定为目标保险产品的目标用户，第二类型用户为保险业务人员。
具体地，本步骤与实施例一中的步骤S105采用完全相同的处理过程，并得到相同的处理结果，此处不再赘述。
本实施例中，首先，通过定期将样本用户的基础数据同步到用户属性库，并对样本用户的基础数据进行聚类分析，得到样本用户的特征数据，然后从用户属性库中获取样本用户的特征数据，对该特征数据进行去噪处理后得到预处理数据。通过在用户属性库这种大数据平台上对特征数据进行大数据分析，去噪等，得到具有针对性的预处理数据，从而保证后续对特征数据的分析和建模能够更加精确，提高预测准率。然后，从预处理数据中选择训练样本数据和测试样本数据，并在预设的训练周期内根据训练样本数据，采用基于用户的协同过滤算法或者基于产品的协同过滤算法，训练样本用户的行为与数据库中预设的保险产品之间的关联模型，并使用测试样本数据对关联模型进行测试，若测试结果满足预设的测试要求，则将关联模型作为学习模型，并基于该学习模型实现对待预测的第一类型用户进行行为预测，得到该第一类型用户的目标保险产品，通过利用CF算法建立学习模型的方式能够自动且准确地找到用户行为与保险产品之间的关系，从而实现对第一类型用户感兴趣的保险产品的准确预测，提高对用户投保行为的预测效率和准确率，以及智能预测水平，同时将该第一类型用户推送给第二类型用户，使得第二类型用户能够及时了解购买保险产品概率高的潜在用户，从而进行有针对性的持续跟踪，提高保险产品的销售成功率和效率。
实施例三：
图4是本申请实施例三提供的一种用户投保行为预测的装置的结构示意图，为了便于说明，仅示出了与本申请实施例相关的部分。图4示例的一种用户投保行为预测的装置可以是前述实施例一提供的用户投保行为预测的方法的执行主体。图4示例的一种用户投保行为预测的装置包括：获取模块31、预处理模块32、建模模块33、预测模块34和推送模块35，各功能模块详细说明如下：
获取模块31，用于从用户属性库中获取样本用户的特征数据，其中，特征数据包括样本用户的行为类型和该行为类型对应的行为次数，行为类型包括订购产品类型和访问网站类型；
预处理模块32，用于对获取模块31获取的特征数据进行预处理，去除噪声数据，得到预处理数据；
建模模块33，用于根据预处理模块32得到的预处理数据，建立学习模型，其中，该学习模型用于将第一类型用户的行为与数据库中预设的保险产品进行匹配，第一类型用户为保险客户；
预测模块34，用于基于建模模块33建立的学习模型，对待预测的第一类型用户进行行 为预测，得到该第一类型用户的目标保险产品；
推送模块35，用于将该第一类型用户推送给第二类型用户，以使第二类型用户将该第一类型用户确定为目标保险产品的目标用户，其中，第二类型用户为保险业务人员。
本实施例提供的一种用户投保行为预测的装置中各模块实现各自功能的过程，具体可参考前述图1所示实施例的描述，此处不再赘述。
从上述图4示例的一种用户投保行为预测的装置可知，本实施例中，通过从用户属性库中获取样本用户的特征数据，对该特征数据进行去噪处理后得到预处理数据，并根据该预处理数据建立学习模型，基于该学习模型实现对待预测的第一类型用户进行行为预测，得到该第一类型用户的目标保险产品，通过建立学习模型的方式能够自动且准确地找到用户行为与保险产品之间的关系，从而实现对第一类型用户感兴趣的保险产品的准确预测，提高对用户投保行为的预测效率和准确率，以及智能预测水平，同时将该第一类型用户推送给第二类型用户，使得第二类型用户能够及时了解购买保险产品概率高的潜在用户，从而进行有针对性的持续跟踪，提高保险产品的销售成功率和效率。
实施例四：
图5是本申请实施例四提供的一种用户投保行为预测的装置的结构示意图，为了便于说明，仅示出了与本申请实施例相关的部分。图5示例的一种用户投保行为预测的装置可以是前述实施例二提供的用户投保行为预测的方法的执行主体。图5示例的一种用户投保行为预测的装置包括：获取模块41、预处理模块42、建模模块43、预测模块44和推送模块45，各功能模块详细说明如下：
获取模块41，用于从用户属性库中获取样本用户的特征数据，其中，特征数据包括样本用户的行为类型和该行为类型对应的行为次数，行为类型包括订购产品类型和访问网站类型；
预处理模块42，用于对获取模块41获取的特征数据进行预处理，去除噪声数据，得到预处理数据；
建模模块43，用于根据预处理模块42得到的预处理数据，建立学习模型，其中，该学习模型用于将第一类型用户的行为与数据库中预设的保险产品进行匹配，第一类型用户为保险客户；
预测模块44，用于基于建模模块43建立的学习模型，对待预测的第一类型用户进行行为预测，得到该第一类型用户的目标保险产品；
推送模块45，用于将该第一类型用户推送给第二类型用户，以使第二类型用户将该第一类型用户确定为目标保险产品的目标用户，其中，第二类型用户为保险业务人员。
进一步地，该装置还包括：
同步模块46，用于定期将样本用户的基础数据同步到用户属性库，其中，基础数据包括网络访问数据和保险管理系统中的保单数据，网络访问数据包括产品订购信息和网站访问信息；
聚类模块47，用于对用户属性库中的基础数据进行聚类分析，得到样本用户的特征数据。
进一步地，建模模块43包括：
选择子模块431，用于从预处理模块42得到的预处理数据中选择训练样本数据和测试样本数据；
训练子模块432，用于在预设的训练周期内根据选择子模块431确定的训练样本数据，训练样本用户的行为与数据库中预设的保险产品之间的关联模型；
测试子模块433，用于使用选择子模块431确定的测试样本数据对训练子模块432确定的关联模型进行测试；
判断子模块434，用于若测试子模块433得到的测试结果满足预设的测试要求，则将训练子模块432确定的关联模型作为学习模型。
进一步地，训练子模块432还用于：
在预设的训练周期内根据选择子模块431确定的训练样本数据，采用协同过滤算法训练样本用户的行为与数据库中预设的保险产品之间的关联模型。
进一步地，训练子模块434还用于：
在预设的训练周期内根据选择子模块431确定的训练样本数据，对样本用户的网络行为进行分析，确定网络行为对应的关联产品；
对网络行为和关联产品进行加权训练，得到样本用户的行为与数据库中预设的保险产品之间的关联模型。
本实施例提供的一种用户投保行为预测的装置中各模块实现各自功能的过程，具体可参考前述图2所示实施例的描述，此处不再赘述。
从上述图5示例的一种用户投保行为预测的装置可知，本实施例中，首先，通过定期将样本用户的基础数据同步到用户属性库，并对样本用户的基础数据进行聚类分析，得到样本用户的特征数据，然后从用户属性库中获取样本用户的特征数据，对该特征数据进行去噪处理后得到预处理数据。通过在用户属性库这种大数据平台上对特征数据进行大数据分析，去噪等，得到具有针对性的预处理数据，从而保证后续对特征数据的分析和建模能够更加精确，提高预测准率。然后，从预处理数据中选择训练样本数据和测试样本数据，并在预设的 训练周期内根据训练样本数据，采用基于用户的协同过滤算法或者基于产品的协同过滤算法，训练样本用户的行为与数据库中预设的保险产品之间的关联模型，并使用测试样本数据对关联模型进行测试，若测试结果满足预设的测试要求，则将关联模型作为学习模型，并基于该学习模型实现对待预测的第一类型用户进行行为预测，得到该第一类型用户的目标保险产品，通过利用CF算法建立学习模型的方式能够自动且准确地找到用户行为与保险产品之间的关系，从而实现对第一类型用户感兴趣的保险产品的准确预测，提高对用户投保行为的预测效率和准确率，以及智能预测水平，同时将该第一类型用户推送给第二类型用户，使得第二类型用户能够及时了解购买保险产品概率高的潜在用户，从而进行有针对性的持续跟踪，提高保险产品的销售成功率和效率。
应理解，上述实施例中各步骤的序号的大小并不意味着执行顺序的先后，各过程的执行顺序应以其功能和内在逻辑确定，而不应对本申请实施例的实施过程构成任何限定。
实施例五：
图6是本申请实施例五提供的用户投保行为预测的计算设备的示意图。如图6所示，该实施例的计算设备6包括：处理器60以及存储器61，所述存储器61中存储有可在所述处理器60上运行的计算机可读指令62，例如用户投保行为预测的程序。所述处理器60执行所述计算机可读指令62时实现上述各个用户投保行为预测的方法实施例中的步骤，例如图1所示的步骤S101至S105。或者，所述处理器60执行所述计算机可读指令62时实现上述各用户投保行为预测的装置实施例中各模块/单元的功能，例如图4所示模块31至35的功能。
示例性的，所述计算机可读指令62可以被分割成一个或多个模块/单元，所述一个或者多个模块/单元被存储在所述存储器61中，并由所述处理器60执行，以完成本申请。所述一个或多个模块/单元可以是能够完成特定功能的一系列计算机可读指令段，该指令段用于描述所述计算机可读指令62在所述计算设备6中的执行过程。
所述计算设备6可以是桌上型计算机、笔记本、掌上电脑及云端服务器等计算设备。所述计算设备可包括，但不仅限于，处理器60以及存储器61。本领域技术人员可以理解，图6仅仅是用户投保行为预测的计算设备6的示例，并不构成对用户投保行为预测的计算设备6的限定，可以包括比图示更多或更少的部件，或者组合某些部件，或者不同的部件，例如所述计算设备还可以包括输入输出设备、网络接入设备、总线等。
所称处理器60可以是中央处理单元(Central Processing Unit，CPU)，还可以是其他通用处理器、数字信号处理器(Digital Signal Processor，DSP)、专用集成电路(Application Specific Integrated Circuit，ASIC)、现成可编程门阵列(Field-Programmable Gate Array，FPGA)或者其他可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件等。通用处理器可以是微 处理器或者该处理器也可以是任何常规的处理器等。
所述存储器61可以是所述计算设备6的内部存储单元，例如所述计算设备6的硬盘或内存。所述存储器61也可以是所述计算设备6的外部存储设备，例如所述计算设备6上配备的插接式硬盘，智能存储卡(Smart Media Card，SMC)，安全数字(Secure Digital,SD)卡，闪存卡(Flash Card)等。进一步地，所述存储器61还可以既包括所述计算设备6的内部存储单元也包括外部存储设备。所述存储器61用于存储所述计算机可读指令以及所述计算设备所需的其他程序和数据。所述存储器61还可以用于暂时地存储已经输出或者将要输出的数据。
另外，在本申请各个实施例中的各功能单元可以集成在一个处理单元中，也可以是各个单元单独物理存在，也可以两个或两个以上单元集成在一个单元中。上述集成的单元既可以采用硬件的形式实现，也可以采用软件功能单元的形式实现。
所述集成的单元如果以软件功能单元的形式实现并作为独立的产品销售或使用时，可以存储在一个计算机可读取存储介质中。基于这样的理解，本申请的技术方案本质上或者说对现有技术做出贡献的部分或者该技术方案的全部或部分可以以软件产品的形式体现出来，该计算机软件产品存储在一个存储介质中，包括若干指令用以使得一台计算机设备(可以是个人计算机，服务器，或者网络设备等)执行本申请各个实施例所述方法的全部或部分步骤。而前述的存储介质包括：U盘、移动硬盘、只读存储器(Read-Only Memory，ROM)、随机存取存储器(Random Access Memory，RAM)、磁碟或者光盘等各种可以存储程序代码的介质。
以上所述，以上实施例仅用以说明本申请的技术方案，而非对其限制；尽管参照前述实施例对本申请进行了详细的说明，本领域的普通技术人员应当理解：其依然可以对前述各实施例所记载的技术方案进行修改，或者对其中部分技术特征进行等同替换；而这些修改或者替换，并不使相应技术方案的本质脱离本申请各实施例技术方案的精神和范围。

Claims (20)

  1. 一种用户投保行为预测的方法，其特征在于，包括：
    从用户属性库中获取样本用户的特征数据，其中，所述特征数据包括所述样本用户的行为类型和所述行为类型对应的行为次数，所述行为类型包括订购产品类型和访问网站类型；
    对所述特征数据进行预处理，去除噪声数据，得到预处理数据；
    根据所述预处理数据，建立学习模型，其中，所述学习模型用于将第一类型用户的行为与数据库中预设的保险产品进行匹配，所述第一类型用户为保险客户；
    基于所述学习模型，对待预测的所述第一类型用户进行行为预测，得到该第一类型用户的目标保险产品；
    将该第一类型用户推送给第二类型用户，以使所述第二类型用户将该第一类型用户确定为所述目标保险产品的目标用户，其中，所述第二类型用户为保险业务人员。
  2. 如权利要求1所述的用户投保行为预测的方法，其特征在于，所述从用户属性库中获取样本用户的特征数据之前，所述方法还包括：
    定期将所述样本用户的基础数据同步到所述用户属性库，其中，所述基础数据包括网络访问数据和保险管理系统中的保单数据，所述网络访问数据包括产品订购信息和网站访问信息；
    对所述基础数据进行聚类分析，得到所述样本用户的特征数据。
  3. 如权利要求1或2所述的用户投保行为预测的方法，其特征在于，所述根据所述预处理数据，建立学习模型包括：
    从所述预处理数据中选择训练样本数据和测试样本数据；
    在预设的训练周期内根据所述训练样本数据，训练所述样本用户的行为与数据库中预设的保险产品之间的关联模型；
    使用所述测试样本数据对所述关联模型进行测试；
    若测试结果满足预设的测试要求，则将所述关联模型作为所述学习模型。
  4. 如权利要求3所述的用户投保行为预测的方法，其特征在于，所述在预设的训练周期内根据所述训练样本数据，训练所述样本用户的行为与数据库中预设的保险产品之间的关联模型包括：
    在预设的训练周期内根据所述训练样本数据，采用协同过滤算法训练所述关联模型。
  5. 如权利要求3所述的用户投保行为预测的方法，其特征在于，所述在预设的训练周期内根据所述训练样本数据，训练所述样本用户的行为与数据库中预设的保险产品之间的关 联模型还包括：
    在所述训练周期内根据所述训练样本数据，对所述样本用户的网络行为进行分析，确定所述网络行为对应的关联产品；
    对所述网络行为和所述关联产品进行加权训练，得到所述关联模型。
  6. 一种用户投保行为预测的装置，其特征在于，包括：
    获取模块，用于从用户属性库中获取样本用户的特征数据，其中，特征数据包括样本用户的行为类型和该行为类型对应的行为次数，行为类型包括订购产品类型和访问网站类型；
    预处理模块，用于对获取模块获取的特征数据进行预处理，去除噪声数据，得到预处理数据；
    建模模块，用于根据预处理模块得到的预处理数据，建立学习模型，其中，该学习模型用于将第一类型用户的行为与数据库中预设的保险产品进行匹配，第一类型用户为保险客户；
    预测模块，用于基于建模模块建立的学习模型，对待预测的第一类型用户进行行为预测，得到该第一类型用户的目标保险产品；
    推送模块，用于将该第一类型用户推送给第二类型用户，以使第二类型用户将该第一类型用户确定为目标保险产品的目标用户，其中，第二类型用户为保险业务人员。
  7. 根据权利要求6所述的用户投保行为预测的装置，其特征在于，还包括：
    同步模块，用于定期将样本用户的基础数据同步到用户属性库，其中，基础数据包括网络访问数据和保险管理系统中的保单数据，网络访问数据包括产品订购信息和网站访问信息；
    聚类模块，用于对用户属性库中的基础数据进行聚类分析，得到样本用户的特征数据。
  8. 根据权利要求6或7所述的用户投保行为预测的装置，其特征在于，所述建模模块包括：
    选择子模块，用于从预处理模块得到的预处理数据中选择训练样本数据和测试样本数据；
    训练子模块，用于在预设的训练周期内根据选择子模块确定的训练样本数据，训练样本用户的行为与数据库中预设的保险产品之间的关联模型；
    测试子模块，用于使用选择子模块确定的测试样本数据对训练子模块确定的关联模型进行测试；
    判断子模块，用于若测试子模块得到的测试结果满足预设的测试要求，则将训练子模块确定的关联模型作为学习模型。
  9. 根据权利要求8所述的用户投保行为预测的装置，其特征在于，所述训练子模块还用于：
    在预设的训练周期内根据选择子模块确定的训练样本数据，采用协同过滤算法训练样本用户的行为与数据库中预设的保险产品之间的关联模型。
  10. 根据权利要求8所述的用户投保行为预测的装置，其特征在于，训练子模块还用于：
    在预设的训练周期内根据选择子模块确定的训练样本数据，对样本用户的网络行为进行分析，确定网络行为对应的关联产品；
    对网络行为和关联产品进行加权训练，得到样本用户的行为与数据库中预设的保险产品之间的关联模型。
  11. 一种用户投保行为预测的计算设备，其特征在于，包括存储器以及处理器，所述存储器中存储有可在所述处理器上运行的计算机可读指令，所述处理器执行所述计算机可读指令时实现如下步骤：
    从用户属性库中获取样本用户的特征数据，其中，所述特征数据包括所述样本用户的行为类型和所述行为类型对应的行为次数，所述行为类型包括订购产品类型和访问网站类型；
    对所述特征数据进行预处理，去除噪声数据，得到预处理数据；
    根据所述预处理数据，建立学习模型，其中，所述学习模型用于将第一类型用户的行为与数据库中预设的保险产品进行匹配，所述第一类型用户为保险客户；
    基于所述学习模型，对待预测的所述第一类型用户进行行为预测，得到该第一类型用户的目标保险产品；
    将该第一类型用户推送给第二类型用户，以使所述第二类型用户将该第一类型用户确定为所述目标保险产品的目标用户，其中，所述第二类型用户为保险业务人员。
  12. 根据权利要求11所述的计算设备，其特征在于，所述处理器执行所述计算机可读指令时还实现如下步骤：
    定期将所述样本用户的基础数据同步到所述用户属性库，其中，所述基础数据包括网络访问数据和保险管理系统中的保单数据，所述网络访问数据包括产品订购信息和网站访问信息；
    对所述基础数据进行聚类分析，得到所述样本用户的特征数据。
  13. 根据权利要求11或12所述的计算设备，其特征在于，所述根据所述预处理数据，建立学习模型包括：
    从所述预处理数据中选择训练样本数据和测试样本数据；
    在预设的训练周期内根据所述训练样本数据，训练所述样本用户的行为与数据库中预 设的保险产品之间的关联模型；
    使用所述测试样本数据对所述关联模型进行测试；
    若测试结果满足预设的测试要求，则将所述关联模型作为所述学习模型。
  14. 根据权利要求13所述的计算设备，其特征在于，所述在预设的训练周期内根据所述训练样本数据，训练所述样本用户的行为与数据库中预设的保险产品之间的关联模型包括：
    在预设的训练周期内根据所述训练样本数据，采用协同过滤算法训练所述关联模型。
  15. 根据权利要求13所述的计算设备，其特征在于，所述在预设的训练周期内根据所述训练样本数据，训练所述样本用户的行为与数据库中预设的保险产品之间的关联模型还包括：
    在所述训练周期内根据所述训练样本数据，对所述样本用户的网络行为进行分析，确定所述网络行为对应的关联产品；
    对所述网络行为和所述关联产品进行加权训练，得到所述关联模型。
  16. 一种计算机可读存储介质，所述计算机可读存储介质存储有计算机可读指令，其特征在于，所述计算机可读指令被至少一个处理器执行时实现如下步骤：
    从用户属性库中获取样本用户的特征数据，其中，所述特征数据包括所述样本用户的行为类型和所述行为类型对应的行为次数，所述行为类型包括订购产品类型和访问网站类型；
    对所述特征数据进行预处理，去除噪声数据，得到预处理数据；
    根据所述预处理数据，建立学习模型，其中，所述学习模型用于将第一类型用户的行为与数据库中预设的保险产品进行匹配，所述第一类型用户为保险客户；
    基于所述学习模型，对待预测的所述第一类型用户进行行为预测，得到该第一类型用户的目标保险产品；
    将该第一类型用户推送给第二类型用户，以使所述第二类型用户将该第一类型用户确定为所述目标保险产品的目标用户，其中，所述第二类型用户为保险业务人员。
  17. 根据权利要求16所述的计算机可读存储介质，其特征在于，所述计算机可读指令被至少一个处理器执行时还实现如下步骤：
    定期将所述样本用户的基础数据同步到所述用户属性库，其中，所述基础数据包括网络访问数据和保险管理系统中的保单数据，所述网络访问数据包括产品订购信息和网站访问信息；
    对所述基础数据进行聚类分析，得到所述样本用户的特征数据。
  18. 根据权利要求16或17所述的计算机可读存储介质，其特征在于，所述根据所述预处理数据，建立学习模型包括：
    从所述预处理数据中选择训练样本数据和测试样本数据；
    在预设的训练周期内根据所述训练样本数据，训练所述样本用户的行为与数据库中预设的保险产品之间的关联模型；
    使用所述测试样本数据对所述关联模型进行测试；
    若测试结果满足预设的测试要求，则将所述关联模型作为所述学习模型。
  19. 根据权利要求18所述的计算机可读存储介质，其特征在于，所述在预设的训练周期内根据所述训练样本数据，训练所述样本用户的行为与数据库中预设的保险产品之间的关联模型包括：
    在预设的训练周期内根据所述训练样本数据，采用协同过滤算法训练所述关联模型。
  20. 根据权利要求18所述的计算机可读存储介质，其特征在于，所述在预设的训练周期内根据所述训练样本数据，训练所述样本用户的行为与数据库中预设的保险产品之间的关联模型还包括：
    在所述训练周期内根据所述训练样本数据，对所述样本用户的网络行为进行分析，确定所述网络行为对应的关联产品；
    对所述网络行为和所述关联产品进行加权训练，得到所述关联模型。
PCT/CN2018/074884 2017-06-09 2018-01-31 用户投保行为预测的方法、装置、计算设备及介质 WO2018223719A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201710434568.8A CN107689008A (zh) 2017-06-09 2017-06-09 一种用户投保行为预测的方法及装置
CN201710434568.8 2017-06-09

Publications (1)

Publication Number Publication Date
WO2018223719A1 true WO2018223719A1 (zh) 2018-12-13

Family

ID=61152613

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/074884 WO2018223719A1 (zh) 2017-06-09 2018-01-31 用户投保行为预测的方法、装置、计算设备及介质

Country Status (2)

Country Link
CN (1) CN107689008A (zh)
WO (1) WO2018223719A1 (zh)

Families Citing this family (21)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108492194A (zh) * 2018-03-06 2018-09-04 平安科技（深圳）有限公司 产品推荐方法、装置及存储介质
CN108664550B (zh) * 2018-03-29 2021-10-01 北京邮电大学 一种对用户行为数据进行漏斗分析方法及装置
CN108520444B (zh) * 2018-04-12 2023-06-27 中国平安人寿保险股份有限公司 保险产品推荐方法、设备、装置及计算机可读存储介质
CN108734550A (zh) * 2018-04-28 2018-11-02 平安科技（深圳）有限公司 推送数据的方法、装置、设备及存储介质
CN108961071B (zh) * 2018-06-01 2023-07-21 中国平安人寿保险股份有限公司 自动预测组合业务收益的方法及终端设备
CN108985950B (zh) * 2018-07-13 2023-04-18 平安科技（深圳）有限公司 电子装置、用户骗保风险预警方法及存储介质
CN109711856B (zh) * 2018-08-17 2023-06-27 中国平安人寿保险股份有限公司 基于大数据的用户分类方法、装置、服务器及存储介质
CN109189372B (zh) * 2018-08-22 2023-12-05 中国平安人寿保险股份有限公司 保险产品的开发脚本生成方法及终端设备
CN109460816B (zh) * 2018-11-16 2020-09-18 焦点科技股份有限公司 一种基于深度学习的用户行为预测方法
CN109300054A (zh) * 2018-11-27 2019-02-01 泰康保险集团股份有限公司 保险产品推荐方法、装置、服务器及存储介质
CN110008977B (zh) * 2018-12-05 2023-08-11 创新先进技术有限公司 聚类模型构建方法以及装置
CN111292194B (zh) * 2018-12-06 2023-08-22 泰康保险集团股份有限公司 网上投保客户数据处理方法、装置、介质及电子设备
CN109785000A (zh) * 2019-01-16 2019-05-21 深圳壹账通智能科技有限公司 客户资源分配方法、装置、存储介质和终端
CN110297968B (zh) * 2019-05-22 2023-10-31 中国平安人寿保险股份有限公司 产品推送方法、装置、计算机设备及存储介质
CN110349034A (zh) * 2019-05-30 2019-10-18 阿里巴巴集团控股有限公司 基于物联网终端的项目推荐方法以及装置
CN110363244A (zh) * 2019-07-16 2019-10-22 中国工商银行股份有限公司 一种营销数据处理的方法和装置
CN112330016A (zh) * 2020-11-04 2021-02-05 广东工业大学 一种基于集成学习的社交网络用户行为预测方法
CN113379455B (zh) * 2021-06-10 2024-02-09 中国铁道科学研究院集团有限公司电子计算技术研究所 订单量预测方法和设备
CN113344723B (zh) * 2021-06-11 2024-02-02 北京十一贝科技有限公司 用户保险认知演进路径预测方法、装置和计算机设备
CN113656702B (zh) * 2021-08-27 2023-07-14 建信基金管理有限责任公司 用户行为的预测方法及装置
CN113850686B (zh) * 2021-10-08 2023-11-28 同盾网络科技有限公司 投保概率确定方法、装置、存储介质及电子设备

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20150088662A1 (en) * 2012-10-10 2015-03-26 Nugg.Ad Ag Predictive Behavioural Targeting
CN105023175A (zh) * 2015-07-24 2015-11-04 金鹃传媒科技股份有限公司 一种基于消费者行为数据分析和分类技术的在线广告分类推送方法及其系统
CN106022800A (zh) * 2016-05-16 2016-10-12 北京百分点信息科技有限公司 一种用户特征数据的处理方法和装置
CN106649774A (zh) * 2016-12-27 2017-05-10 北京百度网讯科技有限公司 基于人工智能的对象推送方法及装置

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20150088662A1 (en) * 2012-10-10 2015-03-26 Nugg.Ad Ag Predictive Behavioural Targeting
CN105023175A (zh) * 2015-07-24 2015-11-04 金鹃传媒科技股份有限公司 一种基于消费者行为数据分析和分类技术的在线广告分类推送方法及其系统
CN106022800A (zh) * 2016-05-16 2016-10-12 北京百分点信息科技有限公司 一种用户特征数据的处理方法和装置
CN106649774A (zh) * 2016-12-27 2017-05-10 北京百度网讯科技有限公司 基于人工智能的对象推送方法及装置

Also Published As

Publication number Publication date
CN107689008A (zh) 2018-02-13

Similar Documents

Publication Publication Date Title
WO2018223719A1 (zh) 用户投保行为预测的方法、装置、计算设备及介质
CN110097066B (zh) 一种用户分类方法、装置及电子设备
CN108170692B (zh) 一种热点事件信息处理方法和装置
Figueiredo et al. Migration and regional trade agreements: A (new) gravity estimation
CN112889042A (zh) 机器学习中超参数的识别与应用
CN106682686A (zh) 一种基于手机上网行为的用户性别预测方法
CN109492180A (zh) 资源推荐方法、装置、计算机设备及计算机可读存储介质
WO2017133615A1 (zh) 一种业务参数获取方法及装置
CN107451832B (zh) 推送信息的方法和装置
US20200286154A1 (en) Utilizing item-level importance sampling models for digital content selection policies
JP2018516404A (ja) 情報推奨方法および装置、ならびにサーバ
CN108021651B (zh) 一种网络舆情风险评估方法及装置
CN107944032B (zh) 用于生成信息的方法和装置
US20120311140A1 (en) Method of processing web access information and server implementing same
WO2017071474A1 (zh) 一种语料处理方法和装置及语料分析方法和装置
KR20130134046A (ko) 하이브리드 협업적 여과 방법을 이용한 코사인 유사도 기반 전문가 추천 장치 및 방법
CN114943279A (zh) 招投标合作关系的预测方法、设备及系统
US20170140003A1 (en) Method for creating individual user profile, electronic device, and non-transitory computer-readable storage medium
WO2019242453A1 (zh) 信息处理方法及装置、存储介质、电子装置
US20220222752A1 (en) Methods for analyzing insurance data and devices thereof
KR20200145346A (ko) 상품 또는 컨텐츠 추천 서비스 제공 프로그램
CN109242690A (zh) 理财产品推荐方法、装置、计算机设备及可读存储介质
CN115048487A (zh) 基于人工智能的舆情分析方法、装置、计算机设备及介质
US11822609B2 (en) Prediction of future prominence attributes in data set
CN114399352A (zh) 一种信息推荐方法、装置、电子设备及存储介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 18814471

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 11/03/2020)

122 Ep: pct application non-entry in european phase

Ref document number: 18814471

Country of ref document: EP

Kind code of ref document: A1