Insurance product recommendation method and system based on customer life model

Abstract

The invention discloses an insurance product recommendation method based on a customer life model, and the method comprises the following steps: collecting customer related data, carrying out the data cleaning, generating customer attributes, and obtaining a customer attribute data set; summarizing data characteristics based on attribute exploratory data analysis and binning analysis, and calculating client attribute mapping values according to the data characteristics to obtain a client attribute mapping data set; according to the client data mapping data set, client features are generated, and a client feature data set is obtained; constructing an initial customer life model; training the initial customer life model by using the customer feature data set to obtain a customer life model; and based on the customer life model, carrying out actual insurance product recommendation. The invention also discloses a system for realizing the insurance product recommendation method based on the customer life model. According to the method, the customer life model can be established by combining the business condition of an enterprise and applying big data and an intelligent algorithm, and the customer insurance demand can be quickly positioned.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence, and specifically relates to an insurance product recommendation method and system based on a customer life model. Background Art

[0002] Personal insurance is the core business of life insurance. In-depth understanding of existing customers and discovering their insurance protection needs are important tasks in expanding insurance business.

[0003] The insurance needs and plans of customers vary from person to person, which places high demands on the quality of insurance agents. At the same time, the agents in the insurance industry change frequently, with a large number of new employees and resignations each year. The current insurance service agents of a large number of customers are not the original sales agents of the insurance contracts. It is difficult for new agents to break the ice with customers based only on the customers' past insurance contracts. The information system of insurance companies actually records various insurance businesses of customers, but the amount of data is large and is not collected according to customer dimensions, making data application difficult. At the same time, in consideration of customer privacy protection, some types of customer data cannot be displayed directly, which also increases the difficulty for insurance agents to understand customers, resulting in insurance agents being unable to accurately give corresponding insurance product recommendations. Summary of the invention

[0004] One of the purposes of the present invention is to provide an insurance product recommendation method based on a customer life model. By organizing customer data from various systems, establishing a customer life model, and predicting customer portraits, it is possible to reduce the difficulty for insurance agents to understand customers and complete insurance product recommendations.

[0005] A second objective of the present invention is to provide a system for implementing the insurance product recommendation method based on the customer life model.

[0006] The present invention provides an insurance product recommendation method based on a customer life model, comprising the following steps:

[0007] S1. Collect customer-related data, clean the data, generate customer attributes, and obtain a customer attribute data set;

[0008] S2. Summarize data features based on attribute exploratory data analysis and binning analysis, and then calculate customer attribute mapping values ​​based on data features to obtain a customer attribute mapping data set;

[0009] S3. Generate customer features based on the customer data mapping data set obtained in step S2 to obtain a customer feature data set;

[0010] S4. Build an initial customer life model;

[0011] S5. Using the customer feature data set, the initial customer life model obtained in step S4 is trained to obtain a customer life model;

[0012] S6. Based on the customer life model obtained in step S5, make actual insurance product recommendations.

[0013] In step S1, the customer attributes include personal attributes C1, family attributes C2, insurance history C3, social attributes C4, VIP classification C5, sales organization C6, long-term insurance C7, payment and renewal C8, short-term insurance C9, claim records C10, insurance benefits C11, and interaction records C12;

[0014] Sub-attributes of personal attributes include age, gender, birthday, family relationship, employee / agent ID, death ID, and single customer ID;

[0015] The sub-attributes of the family attribute include the purchase identification for the family members and the family member relationship;

[0016] The sub-attributes of historical insurance include youth student insurance identification, group accident insurance identification, employee welfare insurance identification, enterprise annuity insurance identification, and employee welfare insurance rating;

[0017] The sub-attributes of social attributes include the most recently filled annual income, occupation category, work unit classification, and work unit grade;

[0018] The sub-attributes of VIP rating include current / historical highest star rating, current / historical highest star rating points, and the time of the last star rating change;

[0019] The sub-attributes of sales organization include agent team, sales channel, urban area / county / two townships;

[0020] The sub-attributes of long-term insurance include the earliest insurance time, the number of historical policies, the cumulative premium paid, and the number of current policies;

[0021] Sub-attributes of payment and renewal include the average annual payment due amount, the current year's renewal premiums, and the number of cases;

[0022] The sub-attributes of short-term insurance include medical compensation insurance, discontinuation of medical compensation insurance, and earliest / latest insurance record;

[0023] The sub-attributes of the claim record include the number of claims, claim amount, packaged claim mark, major illness claim mark, and non-agreement claim mark;

[0024] The sub-attributes of interaction records include the number of face-to-face interviews, event invitation attendance records, and interaction records with agents;

[0025] Customer data is collected through several business systems and data service platforms within the company; data of the same customer from different systems or multiple policies are merged;

[0026] Obtain data from the data mart and source layer and complete data cleaning.

[0027] Step S2 includes the following steps:

[0028] Based on attribute exploratory data analysis, the basic characteristics of customer attribute data sets are analyzed and summarized through numerical distribution statistics, visualization, missing value processing, and correlation analysis;

[0029] Preprocess customer attribute data based on attribute binning, convert continuous attribute variables into discrete categories, divide the value range into several intervals, and then assign each attribute value to the corresponding interval;

[0030] According to the distribution of the values ​​in the customer attribute data set, the customer attribute data set is regularly mapped to the same value interval of the preset interval to obtain a customer attribute mapping data set.

[0031] Step S3 specifically includes: performing feature transformation on the customer attribute mapping data set obtained in step S2, using normalization, logarithmic transformation, and power transformation, so that the values ​​in the customer attribute mapping data set satisfy the normal distribution, and obtaining the customer feature data set.

[0032] Step S4 specifically includes: constructing an initial customer life model;

[0033] The initial customer life model includes the customer portrait module, RFM module and insurance recommendation module;

[0034] The input of the initial customer life model is the input of the customer portrait module, RFM module and insurance recommendation module; the customer portrait module generates a customer portrait based on the input data; the RFM module obtains the difficulty of customer development based on the input data; the insurance recommendation module obtains insurance recommendations based on the input data; the output of the customer portrait module, RFM module and insurance recommendation module is used as the output of the initial customer life model.

[0035] The input of the customer profiling module is customer attribute data; the output of the customer profiling module is the customer profiling;

[0036] Customer profile generation is first calculated as a calculated value based on customer attribute mapping data values, including initial screening score, insurance acceptance, social attribute score, and insurance benefit value;

[0037] First, the customer samples in the customer attribute mapping data set are preliminarily screened. If the customer attributes contain a death mark and no family member relationship, the customer sample is deleted. Otherwise, the preliminarily screening score of the sample is calculated. The preliminarily screening score is a weighted sum of the mapping data values ​​of the employee / agent mark, customer death mark, family member relationship, single customer mark, major disease compensation mark, and non-agreement compensation mark in the customer attributes.

[0038] Insurance acceptance is weighted and summed using the mapped data values ​​of each sub-attribute in the customer attributes, including family attributes, historical insurance, VIP classification, long-term insurance, payment and renewal, and activity exposure. The results are binned to obtain a score within a preset range.

[0039] The social attribute score is a weighted sum of the historical insurance coverage, social attributes, sub-attributes in the VIP classification, age, and the mapped data values ​​of the urban area / county / two townships in the customer attributes.

[0040] The insurance benefit value is weighted summed by the mapping data values ​​of claim records in customer attributes, various sub-attributes of insurance benefits, age, and accumulated premiums paid;

[0041] When generating a customer profile, it is necessary to assign weights to the data values ​​of each customer attribute mapping. The weights are determined based on the business experience of experts and through historical data verification.

[0042] Based on the big model, the initial screening score, insurance acceptance, social attribute score and insurance benefit value are input to obtain the user portrait, which is as follows:

[0043] According to the social attribute score and insurance benefit value, the life stage of the user portrait is obtained; the life stages include no income, entering the workplace, forming a family, career peak, career end, and asset inheritance;

[0044] Based on the insurance acceptance and social attribute scores, we can obtain the feature judgments in the user portrait and give the key features for the insurance product recommendations for the customer.

[0045] Based on the initial screening score and insurance benefit value, the insurance needs in the user portrait are obtained, and the recommended insurance types are given for the user's insurance products.

[0046] The input of the RFM module is customer feature data; the output of the RFM module is the difficulty of customer development;

[0047] The RFM module extracts the consumption time interval R, the number of consumption times F and the consumption amount M of the customer in the preset time period through the customer feature data; the R, F and M data are standardized by the custom activation function derank of the sigmoid class to obtain the final scores of the three eigenvalues ​​R, F and M; the three eigenvalues ​​R, F and M extracted from the customer sample data in the customer feature data set are used to construct the RFM eigenvalue space; according to the numerical distribution of the three eigenvalues ​​R, F and M in the RFM eigenvalue space, the distribution interval is preset, and the two dividing points A and B are calculated using the numpy.percentile function, and A>B; when the eigenvalue is greater than A, it is defined as high, between A and B is defined as medium, and less than B is defined as low. According to the definition combination of the three eigenvalues ​​of R, F and M "high, medium and low", 8 customer types are obtained; according to the 8 obtained customer types, the customer development ease is obtained, which is as follows:

[0048] If the customer's R classification feature value is high or medium, the F classification feature value is high or medium, and the M classification feature value is high or medium, then the customer's development ease is 8;

[0049] If the customer's R classification feature value is low, the F classification feature value is high or medium, and the M classification feature value is high or medium, then the customer's development ease is 7;

[0050] If the customer's R classification feature value is high or medium, the F classification feature value is low, and the M classification feature value is high or medium, then the customer's development ease is 6;

[0051] If the customer's R classification feature value is low, the F classification feature value is low, and the M classification feature value is high or medium, then the customer's development ease is 5;

[0052] If the customer's R classification feature value is high or medium, the F classification feature value is high or medium, and the M classification feature value is low, then the customer's development ease is 4;

[0053] If the customer's R classification feature value is low, the F classification feature value is high or medium, and the M classification feature value is low, then the customer's development ease is 3;

[0054] If the customer's R classification feature value is high or medium, the F classification feature value is low, and the M classification feature value is low, then the customer's development ease is 2;

[0055] If the customer's R classification characteristic value is low, the F classification characteristic value is low, and the M classification characteristic value is low, then the customer's development ease is 1.

[0056] An insurance recommendation module is constructed based on the random forest algorithm; the input of the insurance recommendation module is customer feature data; the output of the insurance recommendation module is the recommended insurance type, recommended annual premium, and recommended number of payment years.

[0057] In the training described in step S5, the model parameter weights are firstly trained iteratively using Mean Decrease Impurity (MDI) and Mean Decrease Accuracy (MDA) to determine the influencing factors of the model parameters. MDI is expressed by the following formula:

[0058]

[0059] Where I(j) is the weight of model parameter j; T is the number of customer attributes; S(t) is the total sample of customer attribute t; Δi j (s, t) is the change value of model parameter j related to sample s and dimension t; s is the sample;

[0060] MDA is expressed using the following formula:

[0061]

[0062] Where accuracy(t) is the accuracy of the model based on the original data in the tth cycle; accuracy(t,permutedj) is the influence of the replacement of model parameter j in the tth cycle.

[0063] The present invention also provides a system for implementing the insurance product recommendation method based on the customer life model, comprising a data acquisition module, a data mapping module, a feature extraction module, a model construction module, a model training module, and a product recommendation module;

[0064] The data acquisition module collects customer-related data, performs data cleaning, generates customer attributes, obtains customer attribute data sets, and uploads the data to the data mapping module;

[0065] The data mapping module summarizes the data features based on the received data, attribute exploratory data analysis and binning analysis, and then calculates the customer attribute mapping value based on the data features to obtain the customer attribute mapping data set, and uploads the data to the feature extraction module;

[0066] The feature extraction module generates customer features based on the received data and maps the data set according to the customer data, obtains the customer feature data set, and uploads the data to the model building module;

[0067] The model building module builds an initial customer life model based on the received data and uploads the data to the model training module;

[0068] The model training module uses the customer feature data set to train the initial customer life model based on the received data, obtains the customer life model, and uploads the data to the product recommendation module;

[0069] The product recommendation module makes actual insurance product recommendations based on the received data and the customer’s life model.

[0070] The present invention discloses an insurance product recommendation method and system based on a customer life model. The method of the present invention can combine the business situation of an enterprise, use big data and intelligent algorithms, establish a customer life model, and quickly locate the customer's insurance needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] Figure 1 It is a schematic diagram of the process of the present invention;

[0072] Figure 2 It is a schematic diagram of the structure of the system of the present invention. DETAILED DESCRIPTION

[0073] The present invention provides an insurance product recommendation method based on a customer life model, and its flow chart is as follows: Figure 1 As shown, the following steps are included:

[0074] S1. Collect customer-related data, clean the data, generate customer attributes, and obtain a customer attribute data set;

[0075] In step S1, the customer attributes include personal attributes C1, family attributes C2, insurance history C3, social attributes C4, VIP classification C5, sales organization C6, long-term insurance C7, payment and renewal C8, short-term insurance C9, claim records C10, insurance benefits C11, and interaction records C12;

[0076] Sub-attributes of personal attributes include age, gender, birthday, family relationship, employee / agent ID, death ID, and single customer ID;

[0077] The sub-attributes of the family attribute include the purchase identification for the family members and the family member relationship;

[0078] The sub-attributes of historical insurance include youth student insurance identification, group accident insurance identification, employee welfare insurance identification, enterprise annuity insurance identification, and employee welfare insurance rating;

[0079] The sub-attributes of social attributes include the most recently filled annual income, occupation category, work unit classification, and work unit grade;

[0080] The sub-attributes of VIP rating include current / historical highest star rating, current / historical highest star rating points, and the time of the last star rating change;

[0081] The sub-attributes of sales organization include agent team, sales channel, urban area / county / two townships;

[0082] The sub-attributes of long-term insurance include the earliest insurance time, the number of historical policies, the cumulative premium paid, and the number of current policies;

[0083] Sub-attributes of payment and renewal include the average annual payment due amount, the current year's renewal premiums, and the number of cases;

[0084] The sub-attributes of short-term insurance include medical compensation insurance, discontinuation of medical compensation insurance, and earliest / latest insurance record;

[0085] The sub-attributes of the claim record include the number of claims, claim amount, packaged claim mark, major illness claim mark, and non-agreement claim mark;

[0086] The sub-attributes of interaction records include the number of face-to-face interviews, event invitation attendance records, and interaction records with agents;

[0087] Customer data is collected through several business systems and data service platforms within the company; data of the same customer from different systems or multiple policies are merged;

[0088] Obtain data from the data mart and source layer and complete data cleaning.

[0089] S2. Summarize the data features based on attribute exploratory data analysis and binning analysis, and then calculate the customer attribute mapping value according to the data features to obtain the customer attribute mapping data set, including the following steps:

[0090] Based on attribute exploratory data analysis, the basic characteristics of customer attribute data sets are analyzed and summarized through numerical distribution statistics, visualization, missing value processing, and correlation analysis;

[0091] Preprocess customer attribute data based on attribute binning, convert continuous attribute variables into discrete categories, divide the value range into several intervals, and then assign each attribute value to the corresponding interval;

[0092] According to the distribution of the values ​​in the customer attribute data set, the customer attribute data set is regularly mapped to the same value interval of the preset interval to obtain a customer attribute mapping data set.

[0093] S3. Generate customer features based on the customer data mapping data set obtained in step S2 to obtain a customer feature data set, specifically:

[0094] The customer attribute mapping data set obtained in step S2 is subjected to feature transformation, using normalization, logarithmic transformation, and power transformation, so that the values ​​in the customer attribute mapping data set satisfy a normal distribution, thereby obtaining a customer feature data set.

[0095] S4. Build an initial customer life model;

[0096] The initial customer life model includes the customer portrait module, RFM module and insurance recommendation module;

[0097] The input of the initial customer life model is the input of the customer portrait module, RFM module and insurance recommendation module; the customer portrait module generates a customer portrait based on the input data; the RFM module obtains the difficulty of customer development based on the input data; the insurance recommendation module obtains insurance recommendations based on the input data; the output of the customer portrait module, RFM module and insurance recommendation module is used as the output of the initial customer life model.

[0098] The input of the customer profiling module is customer attribute data; the output of the customer profiling module is the customer profiling;

[0099] Customer profile generation is first calculated as a calculated value based on customer attribute mapping data values, including initial screening score, insurance acceptance, social attribute score, and insurance benefit value;

[0100] First, the customer samples in the customer attribute mapping data set are preliminarily screened. If the customer attributes contain a death mark and no family member relationship, the customer sample is deleted. Otherwise, the preliminarily screening score of the sample is calculated. The preliminarily screening score is a weighted sum of the mapping data values ​​of the employee / agent mark, customer death mark, family member relationship, single customer mark, major disease compensation mark, and non-agreement compensation mark in the customer attributes.

[0101] Insurance acceptance is weighted and summed using the mapped data values ​​of each sub-attribute in the customer attributes, including family attributes, historical insurance, VIP classification, long-term insurance, payment and renewal, and activity exposure. The results are binned to obtain a score within a preset range.

[0102] The social attribute score is a weighted sum of the historical insurance coverage, social attributes, sub-attributes in the VIP classification, age, and the mapped data values ​​of the urban area / county / two townships in the customer attributes.

[0103] The insurance benefit value is weighted summed by the mapping data values ​​of claim records in customer attributes, various sub-attributes of insurance benefits, age, and accumulated premiums paid;

[0104] When generating a customer profile, it is necessary to assign weights to the data values ​​of each customer attribute mapping. The weights are determined based on the business experience of experts and through historical data verification.

[0105] Based on the big model, the initial screening score, insurance acceptance, social attribute score and insurance benefit value are input to obtain the user portrait, which is as follows:

[0106] According to the social attribute score and insurance benefit value, the life stage of the user portrait is obtained; the life stages include no income, entering the workplace, forming a family, career peak, career end, and asset inheritance;

[0107] Based on the insurance acceptance and social attribute scores, we can obtain the feature judgments in the user portrait and give the key features for the insurance product recommendations for the customer.

[0108] Based on the initial screening score and insurance benefit value, the insurance needs in the user portrait are obtained, and the recommended insurance types are given for the user's insurance products.

[0109] The input of the RFM module is customer feature data; the output of the RFM module is the difficulty of customer development;

[0110] The RFM module extracts the consumption time interval R, the number of consumption times F and the consumption amount M of the customer in the preset time period through the customer feature data; the R, F and M data are standardized by the custom activation function derank of the sigmoid class to obtain the final scores of the three eigenvalues ​​R, F and M; the three eigenvalues ​​R, F and M extracted from the customer sample data in the customer feature data set are used to construct the RFM eigenvalue space; according to the numerical distribution of the three eigenvalues ​​R, F and M in the RFM eigenvalue space, the distribution interval is preset, and the two dividing points A and B are calculated using the numpy.percentile function, and A>B; when the eigenvalue is greater than A, it is defined as high, between A and B is defined as medium, and less than B is defined as low. According to the definition combination of the three eigenvalues ​​of R, F and M "high, medium and low", 8 customer types are obtained; according to the 8 obtained customer types, the customer development ease is obtained, which is as follows:

[0111] If the customer's R classification feature value is high or medium, the F classification feature value is high or medium, and the M classification feature value is high or medium, then the customer's development ease is 8;

[0112] If the customer's R classification feature value is low, the F classification feature value is high or medium, and the M classification feature value is high or medium, then the customer's development ease is 7;

[0113] If the customer's R classification feature value is high or medium, the F classification feature value is low, and the M classification feature value is high or medium, then the customer's development ease is 6;

[0114] If the customer's R classification feature value is low, the F classification feature value is low, and the M classification feature value is high or medium, then the customer's development ease is 5;

[0115] If the customer's R classification feature value is high or medium, the F classification feature value is high or medium, and the M classification feature value is low, then the customer's development ease is 4;

[0116] If the customer's R classification feature value is low, the F classification feature value is high or medium, and the M classification feature value is low, then the customer's development ease is 3;

[0117] If the customer's R classification feature value is high or medium, the F classification feature value is low, and the M classification feature value is low, then the customer's development ease is 2;

[0118] If the customer's R classification characteristic value is low, the F classification characteristic value is low, and the M classification characteristic value is low, then the customer's development ease is 1.

[0119] An insurance recommendation module is constructed based on the random forest algorithm; the input of the insurance recommendation module is customer feature data; the output of the insurance recommendation module is the recommended insurance type, recommended annual premium, and recommended number of payment years.

[0120] S5. Using the customer feature data set, the initial customer life model obtained in step S4 is trained to obtain a customer life model;

[0121] In the training described in step S5, the model parameter weights are firstly trained iteratively using Mean Decrease Impurity (MDI) and Mean Decrease Accuracy (MDA) to determine the influencing factors of the model parameters. MDI is expressed by the following formula:

[0122]

[0123] Where I(j) is the weight of model parameter j; T is the number of customer attributes; S(t) is the total sample of customer attribute t; Δi j (s, t) is the change value of model parameter j related to sample s and dimension t; s is the sample;

[0124] MDA is expressed using the following formula:

[0125]

[0126] Where accuracy(t) is the accuracy of the model based on the original data in the tth cycle; accuracy(t,permutedj) is the influence of the replacement of model parameter j in the tth cycle.

[0127] S6. Based on the customer life model obtained in step S5, make actual insurance product recommendations.

[0128] The present invention also provides a system for implementing the insurance product recommendation method based on the customer life model, and its structural diagram is shown in FIG. Figure 2 As shown, it includes a data acquisition module, a data mapping module, a feature extraction module, a model building module, a model training module, and a product recommendation module;

[0129] The data acquisition module collects customer-related data, performs data cleaning, generates customer attributes, obtains customer attribute data sets, and uploads the data to the data mapping module;

[0130] The data mapping module summarizes the data features based on the received data, attribute exploratory data analysis and binning analysis, and then calculates the customer attribute mapping value based on the data features to obtain the customer attribute mapping data set, and uploads the data to the feature extraction module;

[0131] The feature extraction module generates customer features based on the received data and maps the data set according to the customer data, obtains the customer feature data set, and uploads the data to the model building module;

[0132] The model building module builds an initial customer life model based on the received data and uploads the data to the model training module;

[0133] The model training module uses the customer feature data set to train the initial customer life model based on the received data, obtains the customer life model, and uploads the data to the product recommendation module;

[0134] The product recommendation module makes actual insurance product recommendations based on the received data and the customer’s life model.

Claims

1. A method for recommending insurance products based on a customer life model, characterized in that: The following steps are involved: S1. Collect customer-related data, clean the data, generate customer attributes, and obtain a customer attribute data set; S2. Summarize data features based on attribute exploratory data analysis and binning analysis, and then calculate customer attribute mapping values ​​based on data features to obtain a customer attribute mapping data set; S3. Generate customer features based on the customer data mapping data set obtained in step S2 to obtain a customer feature data set; S4. Build an initial customer life model; S5. Using the customer feature data set, the initial customer life model obtained in step S4 is trained to obtain a customer life model; S6. Based on the customer life model obtained in step S5, make actual insurance product recommendations.

2. The insurance product recommendation method based on customer life model according to claim 1, characterized in that: In step S1, the customer attributes include personal attributes C1, family attributes C2, insurance history C3, social attributes C4, VIP classification C5, sales organization C6, long-term insurance C7, payment and renewal C8, short-term insurance C9, claim records C10, insurance benefits C11, and interaction records C12; Sub-attributes of personal attributes include age, gender, birthday, family relationship, employee / agent ID, death ID, and single customer ID; The sub-attributes of the family attribute include the purchase identification for the family members and the family member relationship; The sub-attributes of historical insurance include youth student insurance identification, group accident insurance identification, employee welfare insurance identification, enterprise annuity insurance identification, and employee welfare insurance rating; The sub-attributes of social attributes include the most recently filled annual income, occupation category, work unit classification, and work unit grade; The sub-attributes of VIP rating include current / historical highest star rating, current / historical highest star rating points, and the time of the last star rating change; The sub-attributes of sales organization include agent team, sales channel, urban area / county / two townships; The sub-attributes of long-term insurance include the earliest insurance time, the number of historical policies, the cumulative premium paid, and the number of current policies; Sub-attributes of payment and renewal include the average annual payment due amount, the current year's renewal premiums, and the number of cases; The sub-attributes of short-term insurance include medical compensation insurance, discontinuation of medical compensation insurance, and earliest / latest insurance record; The sub-attributes of the claim record include the number of claims, claim amount, packaged claim mark, major illness claim mark, and non-agreement claim mark; The sub-attributes of interaction records include the number of face-to-face interviews, event invitation attendance records, and interaction records with agents; Customer data is collected through several business systems and data service platforms within the company; data of the same customer from different systems or multiple policies are merged; Obtain data from the data mart and source layer and complete data cleaning.

3. The insurance product recommendation method based on customer life model according to claim 1, characterized in that: Step S2 includes the following steps: Based on attribute exploratory data analysis, the basic characteristics of customer attribute data sets are analyzed and summarized through numerical distribution statistics, visualization, missing value processing, and correlation analysis; Preprocess customer attribute data based on attribute binning, convert continuous attribute variables into discrete categories, divide the value range into several intervals, and then assign each attribute value to the corresponding interval; According to the distribution of the values ​​in the customer attribute data set, the customer attribute data set is regularly mapped to the same value interval of the preset interval to obtain a customer attribute mapping data set.

4. The insurance product recommendation method based on customer life model according to claim 1, characterized in that: Step S3 specifically includes: performing feature transformation on the customer attribute mapping data set obtained in step S2, using normalization, logarithmic transformation, and power transformation, so that the values ​​in the customer attribute mapping data set satisfy the normal distribution, and obtaining the customer feature data set.

5. The insurance product recommendation method based on customer life model according to claim 1, characterized in that: The initial customer life model includes the customer portrait module, RFM module and insurance recommendation module; The input of the initial customer life model is the input of the customer portrait module, RFM module and insurance recommendation module; the customer portrait module generates a customer portrait based on the input data; the RFM module obtains the difficulty of customer development based on the input data; the insurance recommendation module obtains insurance recommendations based on the input data; the output of the customer portrait module, RFM module and insurance recommendation module is used as the output of the initial customer life model.

6. The insurance product recommendation method based on customer life model according to claim 5, characterized in that: The input of the customer profiling module is customer attribute data; the output of the customer profiling module is the customer profiling; Customer profile generation is first calculated as a calculated value based on customer attribute mapping data values, including initial screening score, insurance acceptance, social attribute score, and insurance benefit value; First, the customer samples in the customer attribute mapping data set are preliminarily screened. If the customer attributes contain a death mark and no family member relationship, the customer sample is deleted. Otherwise, the preliminarily screening score of the sample is calculated. The preliminarily screening score is a weighted sum of the mapping data values ​​of the employee / agent mark, customer death mark, family member relationship, single customer mark, major disease compensation mark, and non-agreement compensation mark in the customer attributes. Insurance acceptance is weighted and summed using the mapped data values ​​of each sub-attribute in the customer attributes, including family attributes, historical insurance, VIP classification, long-term insurance, payment and renewal, and activity exposure. The results are binned to obtain a score within a preset range. The social attribute score is a weighted sum of the historical insurance coverage, social attributes, sub-attributes in the VIP classification, age, and the mapped data values ​​of the urban area / county / two townships in the customer attributes. The insurance benefit value is weighted summed by the mapping data values ​​of claim records in customer attributes, various sub-attributes of insurance benefits, age, and accumulated premiums paid; When generating a customer profile, it is necessary to assign weights to the data values ​​of each customer attribute mapping. The weights are determined based on the business experience of experts and through historical data verification. Based on the big model, the initial screening score, insurance acceptance, social attribute score and insurance benefit value are input to obtain the user portrait, which is as follows: According to the social attribute score and insurance benefit value, the life stage of the user portrait is obtained; the life stages include no income, entering the workplace, forming a family, career peak, career end, and asset inheritance; Based on the insurance acceptance and social attribute scores, we can obtain the feature judgments in the user portrait and give the key features for the insurance product recommendations for the customer. Based on the initial screening score and insurance benefit value, the insurance needs in the user portrait are obtained, and the recommended insurance types are given for the user's insurance products.

7. The insurance product recommendation method based on customer life model according to claim 5, characterized in that: The input of the RFM module is customer feature data; the output of the RFM module is the difficulty of customer development; The RFM module extracts the consumption time interval R, the number of consumption times F and the consumption amount M of the customer in the preset time period through the customer feature data; the R, F and M data are standardized using the sigmoid-like custom activation function derank to obtain the final scores of the three feature values ​​of R, F and M; the RFM feature value space is constructed based on the three feature values ​​of R, F and M extracted from the customer sample data in the customer feature data set; According to the numerical distribution of the three eigenvalues ​​R, F, and M in the RFM eigenvalue space, the distribution interval is preset, and the two dividing points A and B are calculated using the numpy.percentile function, and A>B; when the eigenvalue is greater than A, it is defined as high, between A and B is defined as medium, and less than B is defined as low. According to the definition combination of "high, medium, and low" of the three eigenvalues ​​of R, F, and M, 8 customer types are obtained; according to the obtained 8 customer types, the ease of customer development is obtained, which is specifically: If the customer's R classification feature value is high or medium, the F classification feature value is high or medium, and the M classification feature value is high or medium, then the customer's development ease is 8; If the customer's R classification feature value is low, the F classification feature value is high or medium, and the M classification feature value is high or medium, then the customer's development ease is 7; If the customer's R classification feature value is high or medium, the F classification feature value is low, and the M classification feature value is high or medium, then the customer's development ease is 6; If the customer's R classification feature value is low, the F classification feature value is low, and the M classification feature value is high or medium, then the customer's development ease is 5; If the customer's R classification feature value is high or medium, the F classification feature value is high or medium, and the M classification feature value is low, then the customer's development ease is 4; If the customer's R classification feature value is low, the F classification feature value is high or medium, and the M classification feature value is low, then the customer's development ease is 3; If the customer's R classification feature value is high or medium, the F classification feature value is low, and the M classification feature value is low, then the customer's development ease is 2; If the customer's R classification characteristic value is low, the F classification characteristic value is low, and the M classification characteristic value is low, then the customer's development ease is 1.

8. The insurance product recommendation method based on customer life model according to claim 5, characterized in that: An insurance recommendation module is constructed based on the random forest algorithm; the input of the insurance recommendation module is customer feature data; the output of the insurance recommendation module is the recommended insurance type, recommended annual premium, and recommended number of payment years.

9. The insurance product recommendation method based on customer life model according to claim 1, characterized in that: In the training described in step S5, the model parameter weights are firstly trained iteratively using Mean Decrease Impurity and Mean Decrease Accuracy to determine the influencing factors of the model parameters. MDI is expressed by the following formula: Where I(j) is the weight of model parameter j; T is the number of customer attributes; S(t) is the total sample of customer attribute t; Δi j (s, t) is the change value of model parameter j related to sample s and dimension t; s is the sample; MDA is expressed using the following formula: Where accuracy(t) is the accuracy of the model based on the original data in the tth cycle; accuracy(t,permutedj) is the influence of the replacement of model parameter j in the tth cycle.

10. A system for implementing the insurance product recommendation method based on customer life model as claimed in any one of claims 1 to 9, characterized in that It includes data acquisition module, data mapping module, feature extraction module, model building module, model training module and product recommendation module; The data acquisition module collects customer-related data, performs data cleaning, generates customer attributes, obtains customer attribute data sets, and uploads the data to the data mapping module; The data mapping module summarizes the data features based on the received data, attribute exploratory data analysis and binning analysis, and then calculates the customer attribute mapping value based on the data features to obtain the customer attribute mapping data set, and uploads the data to the feature extraction module; The feature extraction module generates customer features based on the received data and maps the data set according to the customer data, obtains the customer feature data set, and uploads the data to the model building module; The model building module builds an initial customer life model based on the received data and uploads the data to the model training module; The model training module uses the customer feature data set to train the initial customer life model based on the received data, obtains the customer life model, and uploads the data to the product recommendation module; The product recommendation module makes actual insurance product recommendations based on the received data and the customer’s life model.

Images