Insurance potential customer identification method and device, storage medium and electronic equipment

Abstract

The invention provides an insurance potential customer identification method and device, a storage medium and electronic equipment, and belongs to the field of data processing. The method comprises the following steps: acquiring insurance policy internal data of a to-be-identified user; querying external data related to the to-be-identified user based on the policy internal data, wherein the external data comprises one or more of social identity attributes and financial attributes of the user; generating an index value of an evaluation index of the user based on the external data and the insurance policy internal data; and inputting the index numerical value into a pre-trained potential customer identification model to obtain an identification result of whether the user to be identified belongs to a potential customer or not. According to the invention, the accuracy of insurance potential customer identification can be improved.

Description

Technical Field

[0001] The present application relates to the field of data processing, and in particular to a method, device, storage medium and electronic device for identifying potential insurance customers. Background Art

[0002] In traditional insurance business, the evaluation of customer potential value mainly relies on the experience and intuition of sales staff. This method is not only inefficient, but also unreliable. With the rapid development of big data and artificial intelligence technology, more and more insurance companies are beginning to explore the use of data analysis to optimize the evaluation process of customer potential value.

[0003] However, although there are some customer potential value assessment methods based on data analysis in the market, these methods often only focus on the internal data of the insurance company, such as the customer's policy information, purchase history, etc. This single data source limits the accuracy and comprehensiveness of the assessment model, especially some internal data related to the customer's potential value (such as the user's income, etc.) is not necessarily accurate, resulting in the inability of the relevant assessment model to effectively identify the customer's real needs and potential value.

[0004] Based on this, it is necessary to improve the existing potential customer identification technology in the insurance industry. Summary of the invention

[0005] The purpose of this application is to provide a method, device, storage medium and electronic device for identifying potential insurance customers to improve the accuracy of identifying potential insurance customers.

[0006] In a first aspect, the present application provides a method for identifying potential insurance customers, the method comprising:

[0007] Obtain the internal data of the insurance policy of the user to be identified;

[0008] Based on the internal data of the insurance policy, query the external data related to the user to be identified, wherein the external data includes one or more of the social identity attributes and financial attributes of the user;

[0009] Generate an index value of the user's evaluation index based on the external data and the policy internal data;

[0010] The indicator value is input into a pre-trained potential customer identification model to obtain an identification result of whether the user to be identified is a potential customer.

[0011] Optionally, the evaluation index includes an income index, and the policy internal data includes the user's corporate information and the user's location. The index value of the user's evaluation index generated based on the external data and the policy internal data includes:

[0012] Identify the industry type of the user based on the enterprise information;

[0013] Get the income range that matches the industry type and user location;

[0014] An indicator value related to the income indicator is generated based on the income range and the external data.

[0015] Optionally, identifying the industry type of the user based on the enterprise information includes: converting the enterprise information into a digital vector through a trained BERT model; inputting the digital vector into a trained convolutional neural network model for operation, and outputting the industry type.

[0016] Optionally, the method also includes: obtaining a first sample data set containing enterprise information and industry type labels; using the enterprise information in the first sample data set to iteratively train the BERT model to be trained, outputting a first digital vector with industry characteristic information in the current iteration, calculating the loss function in the current iteration based on the difference between the first digital vector and the industry characteristics corresponding to the industry type label, adjusting the parameters in the BERT model to be trained based on the loss function until the iteration termination condition is met, and using the BERT model after the iteration as the trained BERT model.

[0017] Optionally, the potential customer identification model is a random forest prediction model, and the indicator value is input into a pre-trained potential customer identification model to obtain an identification result of whether the user to be identified is a potential customer, including:

[0018] Input the indicator value into a pre-trained random forest prediction model, and output the evaluation score of the user to be identified;

[0019] Based on the evaluation score, an identification result of whether the user to be identified is a potential customer is obtained.

[0020] Optionally, the method also includes: obtaining a second sample data set containing indicator values; dividing the second sample data set into a positive sample set and a negative sample set according to insurance purchase conditions within a preset time period; and iteratively training the random forest prediction model to be trained based on the positive sample set and the negative sample set to form a trained random forest prediction model.

[0021] Optionally, before obtaining the second sample data set containing indicator values, it also includes: constructing factor values ​​of initial evaluation factors based on internal policy data and external data corresponding to the second sample data set; calculating the correlation between each initial evaluation factor based on the factor values; performing sparse processing on the initial evaluation factors based on the correlation, and using the evaluation factors retained after the sparse processing as the evaluation indicators.

[0022] In a second aspect of the present application, a device for identifying potential insurance customers is provided, the device comprising:

[0023] A data acquisition module, used to acquire the internal data of the insurance policy of the user to be identified, and query the external data related to the user to be identified based on the internal data of the insurance policy, wherein the external data includes one or more of the social identity attributes and financial attributes of the user;

[0024] An indicator value generation module, used to generate an indicator value of a user's evaluation indicator based on the external data and the policy internal data;

[0025] The customer identification module is used to input the indicator value into a pre-trained potential customer identification model to obtain an identification result of whether the user to be identified is a potential customer.

[0026] In a third aspect of the present application, a computer-readable storage medium is provided, on which executable instructions are stored. When the executable instructions are executed by a processor, the processor executes the method described in any embodiment of the present application.

[0027] In a fourth aspect of the present application, an electronic device is provided, comprising: one or more processors; and a memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors execute the method described in any embodiment of the present application.

[0028] Compared with the prior art that only uses the internal data of the insurance company, the insurance potential customer identification method, device, storage medium and electronic device in this application integrates internal and external data resources to construct a scientific and reasonable evaluation index system, so that the constructed evaluation index is more comprehensive and accurate. On this basis, with the help of relevant machine learning models for identification and judgment, the target customer group can be more accurately locked in, thereby improving the accuracy of identifying potential customers. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope of the present application.

[0030] Figure 1 A schematic diagram of a process of identifying potential insurance customers in one embodiment;

[0031] Figure 2 A schematic diagram of a training process of a random forest model in one embodiment;

[0032] Figure 3 A schematic diagram of a process of creating an evaluation index in one embodiment;

[0033] Figure 4 A schematic diagram of the structure of a device for identifying potential insurance customers in one embodiment;

[0034] Figure 5 It is a structural schematic diagram of a device for identifying potential insurance customers in another embodiment;

[0035] Figure 6 It is a structural schematic diagram of a device for identifying potential insurance customers in another embodiment;

[0036] Figure 7 FIG. 4 is a diagram showing the internal structure of an electronic device in one embodiment. DETAILED DESCRIPTION

[0037] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0038] All terms (including technical and scientific terms) used in this application have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used here should be interpreted as having a meaning consistent with the context of this specification, and should not be interpreted in an idealized or overly rigid manner.

[0039] For example, the terms "first", "second", etc. used in this application may be used herein to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish a first element from another element. For example, without departing from the scope of this application, a first sample data set may be referred to as a second sample data set, and similarly, a second sample data set may be referred to as a first sample data set. Both the first sample data set and the second sample data set are sample data sets, but they are not the same sample data set.

[0040] For another example, the terms “include”, “comprising”, etc. used in this application indicate the existence of features, steps, operations and / or components, but do not exclude the existence or addition of one or more other features, steps, operations or components.

[0041] This application provides a method for identifying potential insurance customers. Figure 1 The method comprises:

[0042] Step 110, obtaining the internal data of the insurance policy of the user to be identified.

[0043] In this embodiment, the user to be identified is a user who needs to be identified as a potential insurance customer. Potential insurance customers refer to pre-defined customers who meet preset protection needs and have marketing value for insurance companies. The internal policy data is the data left by the user at the insurance agency. The data can be the data recorded on the user's insurance policy, which may include personal information such as the user's name, age, the user's company information, position, user location, and other relevant policy information. The company information can be the company name filled in by the user.

[0044] Optionally, the enterprise information in the internal data of the insurance policy is not necessarily a standardized enterprise name, and it may be the enterprise abbreviation, the full name of the enterprise, or part of the full name of the enterprise, and may not even be the real enterprise where the user is located.

[0045] The database corresponding to the electronic device stores the internal policy data of a large number of users, including the above-mentioned users to be identified and users whose identification results have been identified. The internal policy data of the corresponding users to be identified can be extracted from the database.

[0046] Step 120, query external data related to the user to be identified based on the internal data of the insurance policy.

[0047] External data refers to data that does not belong to the internal data of the insurance policy, that is, external data is the user's data collected from the outside. For the user to be identified, in addition to the internal data of the insurance policy, information related to the user to be identified is queried from an external server or platform by calling relevant interfaces. Based on the preset data format of external data, the collected information is converted into corresponding external data. External data can not only help insurance companies to have a more comprehensive understanding of the background information of the user to be identified, but also provide richer input features for the potential customer identification model, thereby improving the accuracy and efficiency of identification.

[0048] The user's name and other personal information in the insurance's internal data can be used as search conditions, and through data acquisition methods such as relevant API interface calls, data crawlers or data cooperation, inquiries can be made from relevant government public data, financial institutions such as banks and securities, third-party data providers, social media, public network platforms and other channels. Relevant data that matches the user's personal information can be queried from each channel to form corresponding external data.

[0049] Specifically, the authenticity of the acquired data is verified to ensure that the acquired data is the authentic data of the user. The data that has passed the authenticity verification is cleaned and integrated, such as removing duplicate data, processing missing values, converting data formats, merging data from different data sources, etc., to form external data. The external data finally obtained meets the requirements of accuracy and consistency, providing a reliable foundation for subsequent data analysis and potential customer identification.

[0050] External data includes one or more of the user's social identity attributes and financial attributes. Social identity attributes and financial attributes are important dimensions for evaluating the potential of the user to be identified. Social identity attributes mainly reflect information about the user's social status, professional background, educational background, and social relationships. These attributes help insurance companies understand the user's professional stability, social status, and potential risk tolerance, so as to more accurately evaluate their value as potential customers.

[0051] The social identity attributes may include one or more types of data such as occupational identity, education background, social status, social labels, etc. Occupational identity may include civil servants, corporate executives, teachers, doctors, etc. Educational background may include academic qualifications, graduation schools, majors studied, etc. Social status may include the user's social position, honorary awards, etc. Social relationships may include family members, social circles, etc.

[0052] Financial attributes mainly reflect information about the user's financial status, investment preferences, risk tolerance, etc. These attributes help insurance companies understand the user's asset status, investment habits, and potential insurance needs, so as to provide them with more accurate insurance products and services.

[0053] Financial attributes may include one or more of income level, asset status, investment preference, credit record, etc. Income level includes income data such as annual salary, monthly salary, bonus, allowance, etc. Asset status includes data such as real estate, vehicles, deposits, stocks, etc. Investment preference may include data such as stock investment, fund investment, bond investment, etc. Credit record may include bank loan records, credit card usage, etc.

[0054] Step 130, generating the indicator value of the user's evaluation indicator based on the external data and the policy internal data.

[0055] In this embodiment, the evaluation indicators are quantitative standards for measuring the potential of the user to be identified. They are based on the internal and external data of the user and are derived through specific calculation methods and logic. These indicators can reflect the performance of the user in multiple dimensions, such as personal identity, asset allocation preferences, regional economic conditions, etc.

[0056] Electronic devices are pre-set with a variety of evaluation indicators, which can be divided into primary indicators, secondary indicators under primary indicators, etc. according to the level. Primary indicators are the top-level framework of the evaluation indicator system. They usually cover the main aspects of user potential. Primary indicators may include indicators of dimensions such as personal identity characteristics, asset allocation and preferences, regional economic conditions, cross-checking and risk assessment. Cross-checking is used to measure the comparison between the user's standard premium data and the user's consumption data. Secondary indicators are further refinements of primary indicators, and they specifically describe the key elements under each primary indicator. For example, under the primary indicator of personal identity characteristics, secondary indicators such as professional identity, educational background, and social status can be refined; under the primary indicator of asset allocation and preference, secondary indicators such as income level, asset status, and investment preferences can be refined.

[0057] Furthermore, an indicator value calculation engine is pre-configured, in which calculation rules for various evaluation indicators are configured. The calculation rules may include relevant mathematical calculation formulas, statistical methods or business logic, etc., which are used to convert relevant information extracted from internal and external data of insurance policies into indicator values ​​with clear meanings and quantitative standards.

[0058] For the acquired external data and policy internal data, the corresponding indicator value calculation engine can be used to calculate the indicator value of each evaluation indicator. For example, the asset status evaluation indicator can calculate a corresponding value through the asset information such as real estate, vehicles, deposits, etc. in the external data; the income indicator can predict the user's income level based on relevant data such as financial attributes and social identity attributes, and calculate the corresponding value of the income indicator based on the income level.

[0059] Taking occupational identity as one of the evaluation indicators, the electronic device pre-classifies different occupational types and determines the numerical quantification of each occupational type. For example, the occupational type may include civil servants, corporate executives, teachers, doctors, etc. Based on the acquired social identity attribute information, the occupational type matching the user is determined, and the indicator value corresponding to the occupational identity evaluation indicator is calculated based on the quantified value.

[0060] Step 140, input the indicator value into a pre-trained potential customer identification model to obtain an identification result of whether the user to be identified is a potential customer.

[0061] In this embodiment, the electronic device is pre-set with a corresponding potential customer identification model, which can perform calculations based on the input indicator values ​​of each evaluation indicator and output an identification result for the customer corresponding to the indicator value. The identification result can be the probability of the potential customer to which the user to be identified belongs, or can be the output of the corresponding potential customer level or related evaluation score.

[0062] The potential customer identification model can be a model pre-trained according to a relevant machine learning algorithm, for example, the machine learning algorithm can be a random forest prediction model. After obtaining the indicator values ​​of each evaluation indicator, the indicator values ​​can be input into the corresponding identification model to obtain the corresponding identification results.

[0063] Compared with the existing technology that only uses internal data of insurance companies, this application integrates internal and external data resources to construct a scientific and reasonable evaluation index system, making the constructed evaluation indicators more comprehensive and accurate. On this basis, with the help of relevant machine learning models for identification and judgment, the target customer group can be more accurately locked in, thereby improving the accuracy of identifying potential customers.

[0064] In one embodiment, the evaluation index includes an income index, and the internal data of the insurance policy includes the user's business information and the user's location; the indicator value of the user's evaluation index is generated based on the external data and the internal data of the insurance policy, including: identifying the user's industry type based on the business information; obtaining the income range that matches the industry type and the user's location; generating the indicator value of the income index based on the income range and the external data.

[0065] In this embodiment, the income index is one of the important indicators for evaluating potential customers. Generally speaking, the income data of the user filled in the internal data of the insurance policy is not real data. The indicator value calculated directly using the income data in the internal data of the insurance policy is not only inaccurate, but will reduce the accuracy of identifying potential customers.

[0066] Based on this, this application obtains the enterprise information filled in by the user from the internal data of the insurance policy, interprets the enterprise information, and then accurately identifies the industry type of the user's enterprise through the industry classification database or industry identification algorithm. Optionally, the obtained enterprise information is further cross-validated through external data to identify whether the enterprise information in the internal data of the insurance policy is accurate, and on this basis, the corresponding industry type is further determined, so that the calculated industry type is more accurate. The industry type reflects the overall economic status and income level of the user's industry, and is an important reference for evaluating the user's income potential.

[0067] For the identified industry type, the income range corresponding to the industry can be determined based on the user's location in the internal data of the insurance policy. The income range can be obtained based on a large amount of historical data and statistical analysis of related channels. The electronic device pre-sets multiple industry types and sets the income range of each industry type in different regions.

[0068] After obtaining the income range that matches the industry type and user location, we further combine data such as social identity attributes and / or financial attributes in external data to estimate the user's actual income and determine the user's relative position within the income range, thereby generating a quantitative income indicator value.

[0069] The income indicator value can be a specific value or a relative value (such as a percentage of the income range). The income indicator value reflects the user's income level and relative position in the industry and region, and is an important basis for evaluating the user as a potential customer.

[0070] In one embodiment, the industry type of the user is identified based on the enterprise information, including: converting the enterprise information into a digital vector through a trained BERT model; inputting the digital vector into a trained convolutional neural network model for operation, and outputting the industry type.

[0071] In this embodiment, the electronic device is pre-set with a trained BERT (Bidirectional Encoder Representations from Transformers) model and a related convolutional neural network (CNN) model. The BERT model is used to convert the verified enterprise information into a corresponding digital vector as the input of the neural network model, and the neural network model runs based on the input information and outputs the corresponding industry type. The convolutional neural network model can be any suitable convolutional neural network model such as LeNet-5, AlexNet, VGGNet, GoogLeNet, ResNet, DenseNet, MobileNet, etc.

[0072] Enterprise information includes the enterprise name carried in the internal data of the insurance policy, and may also include the full name of the enterprise identified in the external data and related business scope and other data. The enterprise name and the enterprise data in the queried external data are checked for consistency, and the checked data is used as the input of the BERT model. Among them, when the two are completely consistent, the enterprise data in the internal and external data are integrated as the input of the BERT model. When the two are inconsistent, the accuracy of the enterprise name is determined based on the inconsistent result. If it is inaccurate, the identification of the enterprise information is cancelled. When the enterprise data related to the user to be identified cannot be obtained from the external data, only the enterprise name is used as all the enterprise information.

[0073] By inputting enterprise information into the BERT model, the BERT model is able to generate a fixed-length digital vector that accurately captures the semantic characteristics of the enterprise information.

[0074] After obtaining the digital vector of enterprise information, it is input into the trained convolutional neural network model for further processing. The CNN model extracts local features from the input data through convolution and pooling operations, learns the mapping relationship between these features and output labels, and identifies the industry type based on this output.

[0075] Optionally, the CNN model may be able to output the probability distribution of the industry type to which the enterprise information belongs, and based on the probability distribution, select the industry type with the highest probability as the prediction result of the model.

[0076] By combining the BERT model and the CNN model, the BERT model can quickly and accurately output digital vectors suitable for the CNN model, so that the accuracy of the industry type finally output is higher.

[0077] In one embodiment, the above method also includes a training process for the BERT model. The process includes: obtaining a first sample data set containing enterprise information and industry type labels; using the enterprise information in the first sample data set to iteratively train the BERT model to be trained, outputting a first digital vector with industry characteristic information in the current iteration, calculating the loss function in the current iteration based on the difference between the first digital vector and the industry characteristics corresponding to the industry type label, adjusting the parameters in the BERT model to be trained based on the loss function until the iteration termination condition is met; and using the BERT model after the iteration as the trained BERT model.

[0078] In this embodiment, the first sample data set includes multiple first sample data, each of which includes corresponding enterprise information and an industry type label matching the enterprise information. The enterprise information may be enterprise information extracted from internal data of each historical insurance policy, and the corresponding industry type label may be a verified industry type label.

[0079] The industry type label is used to identify the industry type corresponding to the enterprise information. When the enterprise information is false information that does not meet the authenticity verification or does not meet the filling normative requirements, the corresponding industry type label can indicate that the enterprise information does not belong to any industry type or belongs to a false enterprise. For the first sample data of this type, it can be used as the first negative sample data. For the first sample data of enterprise information that meets the authenticity verification or meets the filling normative requirements, it can be used as the first positive sample data.

[0080] Optionally, the first sample data set can be divided into a first training set and a first validation set, wherein the first training set and the first validation set both contain an appropriate number of first positive sample data and first negative sample data. The first sample data in the first training set is used to iteratively train the BERT model. The first sample data in the first validation set is used to verify the trained BERT model, and compare whether the accuracy of the trained BERT model meets the preset accuracy requirements. If not, the BERT model continues to be trained until the output BERT model meets the accuracy requirements, and the final output BERT model is used as the trained BERT model.

[0081] Among them, in each iteration process of the BERT model, the BERT model outputs a corresponding digital vector based on the input enterprise information. The corresponding loss function is calculated based on the output digital vector and the industry type label corresponding to the enterprise information, and the relevant parameters in the BERT model are adjusted based on the loss function. The BERT model with adjusted parameters is retrained by performing multiple iterations of training until the output BERT model meets the iteration termination condition. The iteration termination condition may be that the number of iterations meets a preset iteration number threshold, and / or the value of the loss function is less than a preset difference threshold. For example, when the number of iterations reaches the number threshold, the iteration is terminated, and the BERT model after the iteration is completed is used as the trained BERT model.

[0082] The loss function is an indicator used to measure the difference between the prediction result of the BERT model and the actual result. It is usually a non-negative real number. The smaller the value of the loss function, the closer the prediction result of the model is to the actual result. The loss function can be one or more of the cross-entropy loss and negative log-likelihood loss.

[0083] Based on the calculated loss function, the back propagation algorithm is used to adjust the parameters in the BERT model to be trained. The value of the loss function is gradually reduced by calculating the gradient of the loss function to the model parameters and updating the parameter values ​​in the opposite direction of the gradient.

[0084] In one embodiment, the potential customer identification model is a random forest prediction model, and the indicator value is input into a pre-trained potential customer identification model to obtain an identification result of whether the user to be identified is a potential customer, including: inputting the indicator value into a pre-trained random forest prediction model, outputting an evaluation score of the user to be identified; and obtaining an identification result of whether the user to be identified is a potential customer based on the evaluation score.

[0085] In this embodiment, a random forest prediction model is used to identify potential customers. The random forest prediction model adds two layers of randomness on the basis of the decision tree algorithm, namely, random sampling of data and random extraction of features, so that the model is not prone to overfitting.

[0086] The random forest prediction model can perform calculations based on the indicator values ​​of each input evaluation indicator and output a corresponding evaluation score, which reflects the possibility of a customer becoming a potential customer. Specifically, a mapping relationship between different evaluation scores and potential customers is set, and multiple potential customer levels are set, each level corresponding to a different evaluation score range. After obtaining the corresponding evaluation score, the corresponding potential customer level can be obtained based on the score range in which it is located, thereby outputting the corresponding recognition result. Optionally, the recognition result may include the determined potential customer level and the corresponding evaluation score, which helps business personnel to have a deeper understanding and analysis of the recognition results. The higher the evaluation score, the higher the marketing value of the corresponding user to be identified.

[0087] In one embodiment, the electronic device can also set a corresponding threshold value according to the evaluation score or the predicted probability to classify customers into potential customers and non-potential customers. If the evaluation score or the predicted probability is higher than the threshold value, the customer is determined to be a potential customer; otherwise, it is determined to be a non-potential customer.

[0088] In one embodiment, Figure 2 As shown, the above method also includes a training process for a random forest model. This process:

[0089] Step 210: Obtain a second sample data set containing indicator values.

[0090] In this embodiment, the second sample data set includes multiple second sample data, each of which is formed based on the indicator values ​​of various evaluation indicators of multiple users whose potential customer identification results have been clarified, that is, the indicator value has a corresponding identification result.

[0091] Step 220: divide the second sample data set into a positive sample set and a negative sample set according to the insurance purchase situation within a preset time period.

[0092] The insurance purchase situation is used to determine the recognition result of the corresponding second sample data. Specifically, the recognition result corresponding to the second sample data is the result formed by the insurance purchase situation of the corresponding user within a preset time period. The preset time period can be any suitable time period such as the past year or the past two years. The insurance purchase situation is used to reflect the amount of insurance purchased or the situation of adding or renewing insurance within the observation time period.

[0093] Specifically, a standard premium growth threshold is pre-set, and the growth value of the standard premium of the user within the preset time period is identified. The growth value is compared to see whether it exceeds the growth threshold. If it exceeds, the second sample data is added with a positive sample label as the second positive sample data. Otherwise, a negative sample label is added to the second sample data and classified as the second negative sample data. The set of the second positive sample data containing the positive sample label constitutes the positive sample set; the set of the second negative sample data containing the negative sample label constitutes the negative sample set.

[0094] Step 230, iteratively train the random forest prediction model to be trained based on the positive sample set and the negative sample set to form a trained random forest prediction model.

[0095] Specifically, step 230 includes: 1) initializing the model; 2) iterative training and model evaluation and adjustment; 3) forming a trained model and other processes.

[0096] Among them, initializing the model: the electronic device pre-sets a suitable random forest algorithm implementation and initializes the model parameters, which include the number of decision trees, the maximum depth, the minimum number of sample splits, etc.

[0097] Iterative training and model evaluation and adjustment: The positive sample set and the negative sample set are input into the random forest prediction model as training data. The model learns how to distinguish positive samples from negative samples based on the input feature values. During the training process, the model will continuously adjust the splitting conditions of the decision tree and the prediction values ​​of the leaf nodes to improve the classification accuracy of the model.

[0098] After each iteration of training, the model is evaluated using the validation set data. Specifically, the electronic device sets corresponding evaluation indicators, which may include indicators such as the accuracy, recall rate, and F1 score of the model to evaluate the performance of the model. After each iteration, the corresponding evaluation indicator value is calculated, and the model parameters are adjusted based on the evaluation indicator value, such as increasing or decreasing the number of decision trees, adjusting the maximum depth, etc., to optimize the model performance.

[0099] Through multiple iterations of training, the model will gradually converge to a stable state, at which point the model's classification performance reaches the optimal state. When the number of iterations reaches the preset threshold, or when the evaluation index value meets the termination condition, the iteration is terminated to form a trained model, otherwise the iteration continues.

[0100] Forming a trained model: After multiple iterations of training and parameter adjustment, the model reaches the predetermined performance requirements. At this point, the model is saved as a trained random forest prediction model for subsequent potential customer identification tasks.

[0101] Through the iterative training process of the random forest prediction model described in this embodiment, a more accurate and stable potential customer identification model can be constructed. The model can be applied to the customer management and marketing strategies of insurance companies, helping insurance companies to more accurately identify potential customer groups, thereby improving customer satisfaction and business growth rate.

[0102] In one embodiment, the present application also includes a process of creating an evaluation index. This process may be performed before the above step 210, such as Figure 3 As shown, the process includes:

[0103] Step 310: construct factor values ​​of initial evaluation factors based on the policy internal data and external data corresponding to the second sample data set.

[0104] In this embodiment, the electronic device pre-constructs a first number of initial evaluation factors, which are similar to evaluation indicators and can also be divided into multiple levels, such as primary initial evaluation factors, secondary initial evaluation factors under the primary initial evaluation factors, etc. The initial evaluation factors include, but are not limited to, factors of multiple dimensions such as the insured's age, gender, occupation, health status, historical claims record, credit score, regional economic conditions, and industry development trends.

[0105] Each initial evaluation factor of the second sample data has a corresponding factor value, and the factor value may be a factor value calculated based on the above-mentioned indicator value calculation engine.

[0106] Step 320: Calculate the correlation between the initial evaluation factors based on the factor values.

[0107] For the factor values ​​of each second sample data in the formed second sample data set, the correlation between each evaluation factor can be calculated using relevant statistical methods or machine learning algorithms. For example, the correlation between the initial evaluation factors can be achieved by calculating one or more indicators such as the Pearson correlation coefficient, the Spearman rank correlation coefficient or the mutual information. The calculation of the correlation helps us understand which factors have redundant or complementary relationships, thereby providing a basis for subsequent sparse processing. The higher the correlation, the higher the correlation or repetition between the corresponding initial evaluation factors. The calculated correlation can be between [-1,1], where 1 represents a complete positive correlation, -1 represents a complete negative correlation, and 0 represents no correlation.

[0108] In the process of correlation calculation, the factor values ​​of each initial evaluation factor are first cleaned and standardized to ensure that the data of each factor value is complete, consistent in format, and at the same order of magnitude.

[0109] For the factor values ​​after data cleaning and standardization, the factor type of the initial evaluation factor can be determined and the corresponding correlation calculation model can be used for correlation calculation. The factor type is used to identify whether the corresponding initial evaluation factor belongs to a continuous variable, discrete variable, linear correlation or nonlinear correlation and other characteristic types.

[0110] For example, for the initial evaluation factors of two continuous variables, the Pearson Correlation Coefficient can be used to calculate the linear correlation between the two. For the initial evaluation factors of two discrete variables, the Spearman's Rank Correlation Coefficient can be used to calculate the correlation. For nonlinear situations, the mutual information can be used to calculate the correlation.

[0111] After completing the correlation calculation between the initial evaluation factors, a corresponding correlation matrix is ​​formed. The row and column of each element in the matrix correspond to an evaluation factor, and the elements in the matrix represent the correlation values ​​between the corresponding factor pairs.

[0112] Step 330 , performing a sparse processing on the initial evaluation factors based on the correlation, and using the evaluation factors retained after the sparse processing as evaluation indicators.

[0113] Specifically, one or more sparse methods such as threshold method, stepwise regression method, principal component analysis method, etc. may be combined to screen the evaluation factors.

[0114] For example, for the threshold method, a suitable correlation threshold can be set, and at least one factor in the factor pair below the threshold can be retained. Specifically, the retained factors can be selected to be retained, or multiple factors in the factor pair can be combined to form a new evaluation factor, and the new evaluation factor can be retained. For example, the correlation threshold can be set to 0.9. For the stepwise regression method, the initial evaluation factors that contribute less to the model can be gradually eliminated through stepwise regression analysis. For the principal component analysis (PCA) method, high-dimensional data is projected into a low-dimensional space through PCA, the main information components are retained, and the corresponding initial evaluation factors are screened out based on this.

[0115] By combining one or more of the above sparsification methods, the initial evaluation factors that have a small contribution to the model or a large correlation can be removed, so that the evaluation factors retained in the end have an appropriate number and have a greater contribution value to the final potential customer identification model. For example, the number of initial evaluation factors is 300. After screening through the above process, the number of evaluation factors retained is 137. The 137 retained evaluation factors are used as corresponding evaluation indicators to form 137 evaluation indicators.

[0116] After the sparse processing, the evaluation factors that have the greatest impact on risk assessment are retained as the final evaluation indicators, and the redundancy between evaluation factors is reduced. These indicators are not only fewer in number, but also better reflect the core risk characteristics of insurance business.

[0117] This embodiment optimizes the selection process of evaluation indicators by constructing factor values ​​of initial evaluation factors, calculating correlations, and performing sparse processing, etc. This not only improves the accuracy of risk assessment, but also reduces the complexity of the potential customer identification model and improves the efficiency of insurance potential customer identification.

[0118] In one embodiment, Figure 4 As shown, a device for identifying potential insurance customers is provided, the device comprising:

[0119] The data acquisition module 410 is used to acquire the internal data of the insurance policy of the user to be identified, and query the external data related to the user to be identified based on the internal data of the insurance policy, wherein the external data includes one or more of the user's social identity attributes and financial attributes.

[0120] The indicator value generation module 420 is used to generate the indicator value of the user's evaluation indicator based on external data and policy internal data.

[0121] The customer identification module 430 is used to input the indicator value into a pre-trained potential customer identification model to obtain an identification result of whether the user to be identified is a potential customer.

[0122] In one embodiment, the evaluation index includes an income index, and the internal data of the insurance policy includes the enterprise information and the location of the user. The index value generation module 420 is also used to identify the industry type of the user based on the enterprise information; obtain the income range that matches the industry type and the location of the user; and generate an index value corresponding to the income index based on the income range and external data.

[0123] The indicator value generation module 420 is also used to convert enterprise information into digital vectors through the trained BERT model; the digital vectors are input into the trained convolutional neural network model for operation, and the industry type is output.

[0124] In one embodiment, Figure 5 As shown, the device also includes: a model training module 405, which is used to obtain a first sample data set containing enterprise information and industry type labels; use the enterprise information in the first sample data set to iteratively train the BERT model to be trained, output a first digital vector with industry characteristic information in the current iteration, calculate the loss function in the current iteration based on the difference between the first digital vector and the industry characteristics corresponding to the industry type label, adjust the parameters in the BERT model to be trained based on the loss function until the iteration termination condition is met, and use the BERT model after the iteration as the trained BERT model.

[0125] In one embodiment, the potential customer identification model is a random forest prediction model, and the customer identification module 430 is also used to input the indicator value into a pre-trained random forest prediction model, output the evaluation score of the user to be identified, and obtain the identification result of whether the user to be identified is a potential customer based on the evaluation score.

[0126] In one embodiment, the model training module 405 is also used to obtain a second sample data set containing indicator values; divide the second sample data set into a positive sample set and a negative sample set according to the insurance purchase situation within a preset time period; and iteratively train the random forest prediction model to be trained based on the positive sample set and the negative sample set to form a trained random forest prediction model.

[0127] In one embodiment, Figure 6 As shown, the device also includes: an evaluation index creation module 401, which is used to construct factor values ​​of initial evaluation factors based on the internal data and external data of the insurance policy corresponding to the second sample data set; calculate the correlation between each initial evaluation factor based on the factor value; perform sparse processing on the initial evaluation factors based on the correlation, and use the evaluation factors retained after the sparse processing as evaluation indicators.

[0128] In one embodiment, a computer-readable storage medium is provided, on which executable instructions are stored. When the instructions are executed by a processor, the processor executes the steps in the above-mentioned method embodiments.

[0129] In one embodiment, an electronic device is also provided, including one or more processors; a memory, in which one or more programs are stored, wherein when the one or more programs are executed by one or more processors, the one or more processors execute the steps in the above-mentioned method embodiments.

[0130] In one embodiment, an electronic device is provided, which may be a terminal or a server. Figure 7As shown, the electronic device 700 includes a central processing unit (CPU) 701, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 702 or a program loaded from a storage portion 708 into a random access memory (RAM) 703. In the RAM 703, various programs and data required for the operation of the electronic device 700 are also stored. The CPU 701, the ROM 702, and the RAM 703 are connected to each other via a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.

[0131] The following components are connected to the I / O interface 705: an input section 706 including a keyboard, a mouse, etc.; an output section 707 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 708 including a hard disk, etc.; and a communication section 709 including a network interface card such as a LAN card, a modem, etc. The communication section 709 performs communication processing via a network such as the Internet. A drive 710 is also connected to the I / O interface 705 as needed. A removable medium 711, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 710 as needed, so that a computer program read therefrom is installed into the storage section 708 as needed.

[0132] In particular, according to an embodiment of the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present application includes a computer program product, including a computer-readable medium carrying instructions, in such an embodiment, the instructions can be downloaded and installed from a network through a communication portion 709, and / or installed from a removable medium 711. When the instructions are executed by a central processing unit (CPU) 701, the various method steps described in the present invention are executed.

[0133] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present application.

[0134] In addition, those skilled in the art will appreciate that, although some embodiments herein include certain features included in other embodiments but not other features, the combination of features of different embodiments is meant to be within the scope of the present application and form different embodiments. For example, all of the above embodiments may be used in any combination. The information disclosed in this background technology section is intended only to deepen the understanding of the overall background technology of the present application and should not be regarded as an admission or in any form to imply that the information constitutes prior art known to those skilled in the art.

Claims

1. A method for identifying potential insurance customers, characterized in that: The method comprises: Obtain the internal data of the insurance policy of the user to be identified; Based on the internal data of the insurance policy, query the external data related to the user to be identified, wherein the external data includes one or more of the social identity attributes and financial attributes of the user; Generate an index value of the user's evaluation index based on the external data and the policy internal data; The indicator value is input into a pre-trained potential customer identification model to obtain an identification result of whether the user to be identified is a potential customer.

2. The method according to claim 1, characterized in that: The evaluation index includes an income index, and the internal data of the insurance policy includes the enterprise information and the location of the user; the index value of the evaluation index of the user generated based on the external data and the internal data of the insurance policy includes: Identify the industry type of the user based on the enterprise information; Get the income range that matches the industry type and user location; An indicator value related to the income indicator is generated based on the income range and the external data.

3. The method according to claim 2, characterized in that The identifying the industry type of the user based on the enterprise information includes: The enterprise information is converted into a digital vector through a trained BERT model; The digital vector is input into a trained convolutional neural network model for operation, and the industry type is output.

4. The method according to claim 3, characterized in that The method further comprises: Obtain a first sample data set including enterprise information and industry type labels; Iteratively train the BERT model to be trained using the enterprise information in the first sample data set, output a first digital vector with industry characteristic information in the iteration, calculate the loss function in the iteration based on the difference between the first digital vector and the industry characteristics corresponding to the industry type label, adjust the parameters in the BERT model to be trained based on the loss function until the iteration termination condition is met, and use the BERT model after the iteration as the trained BERT model.

5. The method according to claim 1, characterized in that The potential customer identification model is a random forest prediction model, and the indicator value is input into a pre-trained potential customer identification model to obtain an identification result of whether the user to be identified is a potential customer, including: Input the indicator value into a pre-trained random forest prediction model, and output the evaluation score of the user to be identified; Based on the evaluation score, an identification result of whether the user to be identified is a potential customer is obtained.

6. The method according to claim 5, characterized in that The method further comprises: Obtain a second sample data set containing indicator values; Dividing the second sample data set into a positive sample set and a negative sample set according to insurance purchase conditions within a preset period of time; The random forest prediction model to be trained is iteratively trained based on the positive sample set and the negative sample set to form a trained random forest prediction model.

7. The method according to claim 6, characterized in that Before obtaining the second sample data set containing the indicator values, the method further includes: Constructing factor values ​​of initial evaluation factors based on the policy internal data and external data corresponding to the second sample data set; Calculate the correlation between each initial evaluation factor based on the factor value; The initial evaluation factors are subjected to sparse processing based on the correlation, and the evaluation factors retained after the sparse processing are used as the evaluation indicators.

8. A device for identifying potential insurance customers, characterized in that: The device comprises: A data acquisition module, used to acquire the internal data of the insurance policy of the user to be identified, and query the external data related to the user to be identified based on the internal data of the insurance policy, wherein the external data includes one or more of the social identity attributes and financial attributes of the user; An indicator value generation module, used to generate an indicator value of a user's evaluation indicator based on the external data and the policy internal data; The customer identification module is used to input the indicator value into a pre-trained potential customer identification model to obtain an identification result of whether the user to be identified is a potential customer.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores executable instructions, and when the executable instructions are executed by a processor, the processor is enabled to perform the method according to any one of claims 1 to 7.

10. An electronic device, characterized in that: include: one or more processors; A memory for storing one or more programs, which, when executed by the one or more processors, causes the one or more processors to perform the method according to any one of claims 1 to 7.

Images