Insurance customer resource optimization configuration method and system

Abstract

The invention discloses an insurance customer resource optimal configuration method and system, and mainly relates to the technical field of insurance service resource management. Comprising the following steps: acquiring customer label data of a plurality of customers and salesman label data of a plurality of salesmen; performing preprocessing and feature coding on the customer label data and the salesman label data, and converting non-numerical labels into numerical features; carrying out feature importance analysis on the basis of the preprocessed numerical features, and selecting key features; based on the key features, calculating an association degree matrix between the customer tag and the salesman tag; based on the correlation degree matrix, calculating the matching degree between the to-be-distributed customer and the plurality of salesmen; and selecting a target salesman from the plurality of salesmen according to the matching degree, and allocating the to-be-allocated customer to the target salesman. The method has the beneficial effects that accurate and automatic distribution of customers and salesmen based on portrait label matching is realized, and meanwhile, the service efficiency and customer satisfaction are improved.

Description

Technical Field

[0001] This invention relates to the field of insurance business resource management technology, specifically a method and system for optimizing the allocation of insurance customer resources based on big data profile tag matching. Background Technology

[0002] In the life insurance and other insurance industries, the agent workforce inevitably changes due to career development and staff turnover. When an agent leaves the company, the policyholders they served need to be redistributed to ensure continued customer service and potential secondary development opportunities. The efficiency and accuracy of this customer resource allocation directly affect the insurance company's customer satisfaction, operating costs, and long-term profitability. Currently, the industry mainly relies on the following traditional methods for redistributing such "orphan policy" customers: 1. Allocation based on sales team lineage: This means prioritizing the reassignment of departing sales representatives' clients to their former team leaders or members. This method is primarily based on the convenience of the internal management structure and aims to maintain the stability of team interests. 2. Assignment based on customer's registered geographical location: This involves assigning customers to the nearest sales outlet or agent within that outlet based on their registered residential or workplace address in their policy. This method aims to reduce agents' commuting costs and improve visit efficiency.

[0003] However, the aforementioned existing technologies have revealed numerous problems in practice and are no longer sufficient to meet the needs of modern insurance services. The main problems are as follows: First, a mismatch between the service capabilities of clients and sales agents leads to a decline in service quality. Traditional allocation methods fail to adequately consider whether the knowledge structure, sales expertise, and service experience of the agent match the client's actual needs. For example, an agent specializing in health insurance might be assigned to a client with complex wealth transfer needs, or the significant differences in the client's occupation or educational background might hinder effective communication, ultimately impacting the service experience. Secondly, customer resources are underutilized, resulting in a low success rate for secondary development. Simple kinship or geographical allocation is a passive and inefficient management model. It fails to proactively identify customers' potential and diverse insurance needs and match them with agents best suited to selling corresponding products and managing such customer groups. This leads to the underutilization of customer resources, a core asset, resulting in missed cross-selling and upselling opportunities. Third, allocation strategies lack data support and adaptability, making them difficult to handle complex scenarios. Existing methods mostly rely on static, single rules (such as organizational relationships or geographical locations), lacking in-depth mining and analysis of massive amounts of historical business data, customer profiles, and salesperson profiles. Their allocation logic is rigid and cannot be dynamically adjusted with market trends, changes in customer lifecycles, and the growth of salesperson capabilities, resulting in poor adaptability.

[0004] Therefore, there is an urgent need for a method and system for optimizing the allocation of insurance customer resources based on big data profile tag matching to solve the above problems. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for optimizing the allocation of insurance customer resources based on big data profile tag matching. It achieves accurate and automated allocation of customers and salespersons based on profile tag matching, while improving service efficiency and customer satisfaction.

[0006] To achieve the above objectives, the present invention employs the following technical solution: On the one hand, the present invention provides a method for optimizing the allocation of insurance customer resources, comprising the following steps: Step S1: Obtain customer tag data for a number of customers and salesperson tag data for a number of salespersons. The customer tag data includes, but is not limited to: customer's occupation, marital status, education level, age group, policy type, and customer star rating. The salesperson tag data includes, but is not limited to: the number of salespersons with different occupations, marital statuses, education levels, age groups, policy types, and average customer star rating. Step S2: Preprocess and feature encode the customer tag data and salesperson tag data from step S1, converting non-numerical tags into numerical features; Step S3: Based on the preprocessed numerical features, perform feature importance analysis and select key features; Step S4: Based on key features, calculate the correlation matrix between customer tags and salesperson tags, wherein the correlation matrix is ​​calculated by combining an improved Pearson correlation coefficient with information entropy weighting. Step S5: Based on the correlation matrix, calculate the matching degree between the customer to be assigned and several salespersons; Step S6: Based on the matching degree, select a target salesperson from several salespersons and assign the customers to be assigned to the target salesperson.

[0007] Preferably, in step S2, the customer tag data and salesperson tag data are preprocessed and feature-encoded, specifically as follows: For customer tag data, one-hot encoding is used to encode occupation, marital status, education level, and place of residence, generating binary feature vectors; ordinal encoding is used to encode age group and income level, where age group is mapped to ordinal values ​​according to age intervals, and income level is mapped to ordinal values ​​according to annual income range; numerical mapping is used to encode policy type and customer star rating, where policy type is mapped to integer values ​​according to insurance type classification, and customer star rating is directly mapped to integer values. For salesperson tag data, the mode is used to aggregate and encode the salesperson's customer occupation, marital status, education, age group, and customer policy type, and the mean is used to aggregate and encode the customer star rating.

[0008] Preferably, in step S3, the feature importance analysis uses principal component analysis for dimensionality reduction, including the following steps: Step S31: Construct a feature matrix from the preprocessed label data; Step S32: Perform principal component analysis on the feature matrix and select the principal component with the highest contribution rate as the key feature; Principal component analysis is represented as follows: ; in, Indicates selection Principal components, It is an integer, determined based on the feature contribution rate; The key characteristics include, but are not limited to: occupation, marital status, education level, age group, policy type, and customer rating.

[0009] Preferably, in step S4, calculating the correlation matrix includes: Construct a customer-salesperson joint feature matrix, where each row represents a salesperson and the columns include the aggregated value of the salesperson's tags and the average value of the tags of the customers under their name; For each pair of customer tags and salesperson tags, calculate the Pearson correlation coefficient; Calculate the information entropy of each label, and use the information entropy to weight and adjust the Pearson correlation coefficient to obtain the adjusted correlation degree; The adjusted correlation values ​​are normalized to the [0,1] interval, and a threshold is applied to filter weak correlations.

[0010] Preferably, the Pearson correlation coefficient is expressed as follows: ; in, Indicates the first Sample values ​​for each salesperson's tag. Indicates the first Sample values ​​of customer tags, This represents the sample mean of the salesperson tags. This represents the sample mean of customer tags. Indicates the number of samples.

[0011] Preferably, the information entropy is expressed as: ; in, This indicates the probability of a label value appearing in a sample. Indicates the number of different values ​​that the label can take; The information entropy is used to measure the discriminative power of the labels; the higher the entropy value, the greater the discriminative power of the features.

[0012] The preferred weighted correlation is expressed as follows: ; in, The Pearson correlation coefficient is... Information entropy for salesperson tags Information entropy for customer tags; Normalization is represented as: ; in, This represents the normalized correlation score. and These represent the minimum and maximum values ​​of the adjusted correlation coefficient, respectively. The threshold filtering method considers correlation scores with a normalized value lower than 0.3 as invalid.

[0013] Preferably, in step S5, calculating the matching degree between the customer to be assigned and several salespersons includes: Generate customer feature vectors from the tag values ​​of the customers to be assigned; Generate a salesperson feature vector from the tag values ​​of each salesperson; Using the correlation matrix, calculate the matching score between the customer feature vector and the salesperson feature vector; The matching score is calculated using matrix multiplication. Specifically, the customer feature vector is multiplied by the correlation matrix to obtain an intermediate matrix, and then the intermediate matrix is ​​multiplied by the salesperson feature vector to obtain the final matching degree.

[0014] Preferably, step S5 further includes: screening candidate salespersons based on customer location information, specifically: Using the customer's registered address as the center, set an initial radius range, and filter the salespersons within this range as a candidate set; If there are fewer than 3 salespeople in the candidate set, the radius is expanded until at least 3 candidate salespeople are obtained. Calculate the matching degree for candidate salespersons and select the salesperson with the highest matching degree as the target salesperson; If several sales representatives have the same match score, the radius range is expanded and the match score is recalculated until a unique target sales representative is identified.

[0015] On the other hand, the present invention provides an insurance customer resource optimization allocation system for implementing the above-described insurance customer resource optimization allocation method, including: The data acquisition module is used to acquire customer tag data from multiple customers and salesperson tag data from multiple salespersons. The preprocessing module is used to preprocess and feature-encode the customer tag data and salesperson tag data; The feature analysis module is used to perform feature importance analysis to select key features; The correlation calculation module is used to calculate the correlation matrix between customer tags and salesperson tags; The matching degree calculation module is used to calculate the matching degree between the customer to be assigned and multiple salespersons based on the correlation degree matrix; The allocation module is used to select target salespersons and assign customers based on matching degree.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. It achieves precise and intelligent matching between customers and salespersons, significantly improving service quality and marketing efficiency: This invention breaks through the traditional extensive allocation model that relies on blood ties or geographical location. By constructing a multi-dimensional profile tag system and using an improved correlation algorithm to deeply mine the intrinsic connection between customer needs and salesperson capabilities, customers can be assigned to salespersons who are best suited to serve their specific needs and who best understand their potential needs. This effectively improves the targeting of services, customer satisfaction, and significantly increases the success rate of customer re-development. 2. It drives the shift in insurance customer resource allocation from "experience-driven" to "data-driven," optimizing enterprise resource allocation: This invention bases allocation decisions on objective and quantitative data analysis. By extracting features and performing model calculations on massive amounts of internal and external data, it replaces subjective and one-sided experience-based judgments. This not only reduces allocation bias caused by subjective factors of personnel, but also enables the company's core resources to be allocated to salespeople who can create the greatest value, thereby improving the company's overall resource utilization efficiency and input-output ratio. 3. Improved the automation level and scientific rigor of the customer allocation process, enhancing the adaptability of business operations: This invention provides a complete and automatically executable solution, from data preprocessing and feature analysis to matching degree calculation and final allocation. It automates and simplifies the process, adapts to the complex business environment of dynamic changes in the sales force and continuous updates to customer profiles, and has the ability to continuously learn and optimize. It provides insurance companies with a stable, efficient and scalable operational support method in the face of fierce market competition. Attached Figure Description

[0017] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a schematic diagram of the system structure of the present invention. Detailed Implementation

[0018] The present invention will be further illustrated below with reference to specific embodiments. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. Furthermore, it should be understood that after reading the teachings of this invention, those skilled in the art can make various alterations or modifications to the invention, and these equivalent forms also fall within the scope defined in this application.

[0019] In this invention, terms such as "upper," "lower," "left," "right," "front," "back," "vertical," "horizontal," "side," and "bottom" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. These terms are used only to facilitate the description of the structural relationships of the various components or elements of this invention and do not specifically refer to any component or element in this invention. They should not be construed as limiting the invention.

[0020] Example: like Figure 1 As shown, this embodiment provides a method for optimizing the allocation of insurance customer resources based on big data profile tag matching, including: Step S1: Obtain customer tag data for a number of customers and salesperson tag data for a number of salespersons. Customer tag data includes, but is not limited to: customer's occupation, marital status, education level, age group, policy type, and customer star rating. Salesperson tag data includes, but is not limited to: the number of salespersons with different occupations, marital statuses, education levels, age groups, policy types, and average customer star ratings. Step S2: Preprocess and feature encode the customer tag data and salesperson tag data from step S1, converting non-numerical tags into numerical features; Step S3: Based on the preprocessed numerical features, perform feature importance analysis and select key features; Step S4: Based on key features, calculate the correlation matrix between customer tags and salesperson tags, where the correlation matrix is ​​calculated by combining the improved Pearson correlation coefficient with information entropy weight adjustment; Step S5: Based on the correlation matrix, calculate the matching degree between the customer to be assigned and several salespersons; Step S6: Based on the matching degree, select the target salesperson from several salespersons and assign the customers to be assigned to the target salesperson.

[0021] This embodiment takes Company A (a large life insurance company) as an example to provide a detailed explanation of a method for optimizing the allocation of insurance customer resources through big data profile tag matching.

[0022] Step S1: Company A first extracts data from its core business system, CRM system, and human resources system, including customer data and salesperson data; Specifically, the acquisition of customer data involves extracting raw information from the 10,000 "orphan policy" customers who need to be reassigned, as well as from the active customers under all the company's current agents (used for model training). Taking customer "C1001" as an example, their raw data is as follows: "Customer ID: C1001; Occupation: Engineer; Marital Status: Married; Education: Master's Degree; Age: 38; Last Valid Policy Type: Whole Life Insurance; Customer Rating: 4 stars (based on an annual premium of 15,000 yuan); Annual Income: 350,000 yuan; Gender: Male; Residence: No. 123, Z Road, Y District, X City". The acquisition of salesperson data specifically involves extracting their personal attributes and aggregated data of their historical clients from all salespersons with potential clients. For example, salespersons "A2001" and "A2002" are described as follows: "Salesperson ID: A2001; Most frequent occupation of their clients (max_count(client occupation)): Business owner; Most frequent marital status of their clients: Married; Most frequent education level of their clients: Bachelor's degree; Most frequent age group of their clients: 35-45 years old; Most frequent policy type of their clients (max_count(client policy type)): Annuity insurance; Average star rating of their clients (max_count(client star rating)): 3.8; Gender: Male." Salesperson ID: A2002; Most frequent occupation of clients under their name (max_count(client occupation)): Engineer; Most frequent marital status of clients under their name: Married; Most frequent education level of clients under their name: Master's degree; Most frequent age group of clients under their name: 30-40 years old; Most frequent policy type of clients under their name (max_count(client policy type)): Whole life insurance; Average star rating of clients under their name (max_count(client star rating)): 4.2; Gender: Male.

[0023] Step S2: Company A encodes the above-mentioned original label data: Customer C1001 label encoding: "Occupation: Engineer → One-hot encoding. In the 'Occupation' feature vector, the position corresponding to 'Engineer' is 1, and the positions of other occupations are 0. For the sake of simplicity, we assume that 'Engineer' is encoded as the value 3." Marital status: Married → One-hot encoding, assuming 'married' is encoded as the value 1; Education level: Master's degree → one-hot encoding, assuming 'Master' is encoded as the value 4; Age range: 38 years old → ordinal code, falling into the '36-45 years old' range, mapped to the ordinal value 3; Policy type: Whole life insurance → numerical mapping, assuming 'whole life insurance' is mapped to 2; Customer rating: 4 stars → Numericalized, directly obtained; Income level: 350,000 yuan → ordinal code, falling into the '200,000-500,000' range, mapped to the ordinal value 4; Gender: Male → ordinal code, mapped to 0; Salesperson tagging: Salesperson A2001: max_count(customer occupation): Business owner → code is 5. Marital status (mode): Married → Code 1 Education level (mode): Bachelor's degree → Code 3 Age group (mode): 35-45 years old → coded as 3. max_count(customer policy type): Annuity insurance → code is 3 max_count(customer star rating): 3.8 → Numericalized, set to 3.8. Gender: Male → Coded as 0; Salesperson A2002: (Coding rules are the same as A2001) max_count(customer occupation): Engineer → code is 3. Marital status: Married → Code 1 Education level: Master's degree → Code 4 Age group: 30-40 years old → Coded as 2 max_count(customer policy type): Whole life insurance → code is 2. max_count(customer star rating): 4.2 → Get 4.2. Gender: Male → Coded as 0.

[0024] Step S3: Company A has a large number of initial labels, but not all labels contribute significantly to the matching degree calculation. To reduce dimensionality and remove noise, Company A uses Principal Component Analysis (PCA) to perform feature importance analysis. 1. Construct a feature matrix: Construct a massive feature matrix from the initial coded tags (9 dimensions in total: occupation, marital status, education, age group, policy type, customer star rating, income level, gender, and place of residence) of tens of thousands of historical customers and salespersons; 2. Perform PCA analysis: "import pandas as pd; from sklearn.decomposition import PCA; # df is a DataFrame containing all initial labels; pca=PCA(n_components=6); # Six principal components were selected to retain more than 90% of the variance contribution rate; principal_components=pca.fit_transform(df); client_data=pd.DataFrame(data=principal_components, columns=['PC1', 'PC2', 'PC3', 'PC4', 'PC5', 'PC6']);"; 3. Screening key features: By analyzing the loadings of each principal component, Company A found that the six labels "occupation", "marital status", "education level", "age group", "policy type" and "customer star rating" had the highest contribution. Therefore, it was decided that these six key features would be used for both customers and agents in the subsequent correlation calculation. Income level, gender and place of residence were not considered in this round of modeling because of their low variance contribution.

[0025] Step S4: Company A uses historical data on the tag relationships between salespersons and their well-served customers to calculate a general "attribute tag correlation matrix": 1. Construct a joint feature matrix: Company A selected 10,000 high-performing salespersons and their core customers with stable service relationships and high satisfaction. Each row of data represents a "salesperson-core customer" pairing sample, which includes the salesperson's 6 key tag codes and the core customer's 6 key tag codes. 2. Calculate the basic Pearson correlation coefficient: For these 6 salesperson tags and 6 customer tags, calculate the Pearson correlation coefficient for each pair; For example, calculate the correlation coefficient between "salesperson - max_count(customer occupation)" and "customer - occupation": The calculation formula is: ; in, Indicates the first The "max_count(customer occupation)" code value for each salesperson; Indicates the first Each core customer has a corresponding "professional" code value; This represents the average value of the "max_count(customer occupation)" code for all salespersons; This represents the average of the "profession" code values ​​for all core customers. Indicates the sample size (10000); 3. Introducing Entropy Weighting: To consider the information content and distinguishability of different tags, Company A calculated the information entropy of each tag using the following formula: ; in, This indicates that the label is number 1. The probability of each value appearing in the sample. This indicates the number of different values ​​that the label can take; Information entropy The higher the value, the more dispersed the value distribution of the tag, the greater the amount of information it contains, and the higher its importance in matching should be. 4. Calculate the weighted adjusted correlation coefficient: For each calculated Pearson correlation coefficient... The weighted correction is performed using the information entropy of the two corresponding labels, and the calculation formula is as follows: ; in, The Pearson correlation coefficient is... Information entropy for salesperson tags Information entropy for customer tags, It is the harmonic weight, used to balance the entropy influence of the two labels; 5. Generate a relevance matrix: Create a 6x6 matrix with row headings representing customer key tags and column headings representing salesperson key tags. Each cell... Enter the corresponding weighted adjusted correlation. Then, the matrix is ​​normalized so that all values ​​fall within the range of [0,1], and a threshold of 0.3 is set. The correlation degree below 0.3 is set to 0, which is considered to be too weak to be ignored. Company A ultimately obtained an attribute tag correlation pattern matrix.

[0026] Step S5: Company A needs to assign customer C1001 to the most suitable salesperson: 1. Generate customer feature vector: After customer C1001 is coded and key feature is selected, its feature vector is generated. for: (These correspond to: occupation, marital status, education level, age group, policy type, and customer rating, respectively). 2. Generate salesperson feature vectors: Feature vector of salesperson A2001: ; Feature vector of salesperson A2002: ; 3. Calculate the matching degree: (1) Multiply the customer vector with the correlation matrix to obtain an intermediate vector. This intermediate vector can be understood as the "projection of customer tags on the salesperson tag space". Then, perform a dot product between this intermediate vector and the salesperson vector to obtain the final matching score. (2) Calculate the matching degree between customer C1001 and salesperson A2002: intermediate vector , This calculation involves the multiplication of a vector and a matrix, and the result... It is a 1x6 vector; Final match (Dot product with the transpose of salesperson A2002); (3) Similarly, calculate the matching degree with A2001; After calculation, we can obtain: Matching degree (C1001, A2001) = 0.65; Matching degree (C1001, A2002) = 0.92; 4. In order to ensure the convenience of service, Company A can first take the address of customer C1001, "No. 123, Z Road, Y District, X City", as the center, set an initial radius (such as 5 kilometers), and filter out all candidate salespersons within this range (assuming it includes A2001, A2002, and A2003). Then, the above profile matching degree was calculated only for these three candidate salespersons, and the results showed that A2002 had the highest matching degree (0.92); 5. The system automatically assigns customer C1001 to the salesperson A2002 with the highest matching degree. This is because A2002's historical customer group characteristics (engineer, highly educated, holding whole life insurance, high star rating) are highly consistent with customer C1001's personal profile, indicating that A2002 can better understand C1001's needs and provide more professional and appropriate services.

[0027] Step S6: Based on the matching degree, select a target salesperson from several salespersons and assign the customers to be assigned to the target salesperson.

[0028] like Figure 2 As shown in the figure, this embodiment also provides an insurance customer resource optimization and allocation system, including: The data acquisition module is used to acquire customer tag data from multiple customers and salesperson tag data from multiple salespersons. The preprocessing module is used to preprocess and feature-encode the customer tag data and salesperson tag data; The feature analysis module is used to perform feature importance analysis to select key features; The correlation calculation module is used to calculate the correlation matrix between customer tags and salesperson tags; The matching degree calculation module is used to calculate the matching degree between the customer to be assigned and multiple salespersons based on the correlation degree matrix; The allocation module is used to select target salespersons and assign customers based on matching degree.

[0029] The above is a detailed description of the preferred embodiments of the present invention. However, the present invention is not limited to the embodiments described. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention. All such equivalent modifications or substitutions are included within the scope defined by the claims of this application.

Claims

1. An insurance customer resource optimization method, characterized by, The method comprises the following steps: Step S1: obtaining customer label data of a plurality of customers and agent label data of a plurality of agents, wherein the customer label data comprises but is not limited to the occupation, marital status, education, age range, policy type and customer star level of the customers, and the agent label data comprises but is not limited to the mode of the occupation, the mode of the marital status, the mode of the education, the mode of the age range, the mode of the policy type and the mean value of the customer star level of the agents; Step S2: preprocessing and feature coding the customer label data and the agent label data in step S1, and converting the non-numeric labels into numeric features; Step S3: performing feature importance analysis based on the preprocessed numeric features, and selecting key features; Step S4: calculating the correlation matrix between the customer label and the agent label based on the key features, wherein the correlation matrix is calculated by using the improved Pearson correlation coefficient combined with information entropy weight adjustment; Step S5: calculating the matching degree between the to-be-assigned customer and the plurality of agents based on the correlation matrix; Step S6: selecting a target agent from the plurality of agents according to the matching degree, and assigning the to-be-assigned customer to the target agent.

2. The method of claim 1, wherein, In step S2, the customer label data and the agent label data are preprocessed and feature coded, specifically: For the customer label data, the occupation, marital status, education and residence are encoded using one-hot encoding to generate a binary feature vector; the age range and income level are encoded using ordinal encoding, wherein the age range is mapped to ordinal values according to the age interval, and the income level is mapped to ordinal values according to the annual income range; the policy type and customer star level are encoded using numeric mapping, wherein the policy type is mapped to integer values according to the classification of the insurance type, and the customer star level is directly mapped to integer values; For the agent label data, the occupation, marital status, education, age range and policy type of the customers of the agents are aggregated and encoded using the mode, and the customer star level is aggregated and encoded using the mean value.

3. The method of claim 1, wherein, In step S3, the feature importance analysis adopts principal component analysis for dimension reduction, comprising the following steps: Step S31: constructing the preprocessed label data into a feature matrix; Step S32: performing principal component analysis on the feature matrix, and selecting the principal component with the highest contribution rate as the key feature; The principal component analysis is represented as: ； wherein indicates the selection principal components, is an integer determined according to the characteristic contribution rate; The key features include but are not limited to the occupation, marital status, education, age range, policy type and customer star level.

4. The method of claim 1, wherein, In step S4, the calculation of the correlation matrix comprises: Constructing a customer-agent joint feature matrix, wherein each row represents an agent, and the columns include the agent label aggregation value and the label mean value of the customers under the agent; For each pair of customer label and agent label, calculating the Pearson correlation coefficient; Calculating the information entropy of each label, and using the information entropy to weight and adjust the Pearson correlation coefficient to obtain the adjusted correlation degree; Normalizing the adjusted correlation degree value to the interval [0, 1], and applying threshold filtering for weak correlation.

5. The method of claim 4, wherein, The Pearson correlation coefficient is represented as: ； wherein, represents a sample value of the th salesperson label, represents a sample value of the th customer label, represents a sample mean of the salesperson label, represents a sample mean of the customer label, represents a sample number.

6. The method of claim 4, wherein, The information entropy is represented as: ； wherein, denotes the probability of the label value to occur in the sample, denotes the number of different values of the label value; The information entropy is used to measure the distinguishability of the label, and the higher the entropy value, the greater the feature distinguishability.

7. The method of claim 4, wherein, The weighted adjusted correlation degree is represented as: ； wherein, is the Pearson correlation coefficient, is the information entropy of the salesperson label, is the information entropy of the customer label; The normalization processing is represented as: ； wherein, represents the normalized correlation value, and respectively represent the minimum and maximum of the adjusted correlation value; The threshold filtering is to regard the correlation degree lower than 0.3 after normalization as invalid.

8. The method of claim 1, wherein, In step S5, the matching degree between the to-be-assigned customer and the several salesmen includes: The label value of the to-be-assigned customer is generated into a customer feature vector; The label value of each salesman is generated into a salesman feature vector; The matching score between the customer feature vector and the salesman feature vector is calculated using the correlation degree matrix. The matching score is calculated by matrix multiplication, specifically: the customer feature vector is multiplied by the correlation degree matrix to obtain an intermediate matrix, and then the intermediate matrix is multiplied by the salesman feature vector to obtain the final matching degree.

9. The method of claim 8, wherein, Step S5 further includes: screening candidate salesmen based on customer location information, specifically: Taking the customer registration address as the center, an initial radius range is set, and salesmen in the range are screened as a candidate set; If the number of salesmen in the candidate set is less than 3, the radius range is expanded until at least 3 candidate salesmen are obtained; The matching degree of the candidate salesmen is calculated, and the salesman with the highest matching degree is selected as the target salesman; If the matching degrees of the several salesmen are the same, the radius range is continuously expanded, and the matching degree is recalculated until a unique target salesman is determined.

10. An insurance customer resource optimization system for implementing an insurance customer resource optimization method as claimed in any of claims 1 to 9, characterized by It includes: A data acquisition module is configured to acquire customer label data of a plurality of customers and salesman label data of a plurality of salesmen; A preprocessing module is configured to preprocess and feature encode the customer label data and the salesman label data; A feature analysis module is configured to perform feature importance analysis to select key features; A correlation degree calculation module is configured to calculate a correlation degree matrix between customer labels and salesman labels; A matching degree calculation module is configured to calculate a matching degree between a to-be-assigned customer and a plurality of salesmen based on the correlation degree matrix; An assignment module is configured to select a target salesman according to the matching degree and assign the customer.

Images