Insurance information integration method

Abstract

The invention relates to the technical field of insurance information integration, and discloses an insurance information integration method. The method comprises the following steps: reading annual insurance premium and total claim settlement amount of a customer, calculating and normalizing an extreme value, and eliminating dimensional difference; constructing a customer-insurance policy-risk three-layer edge set, generating a risk label set, calculating a similarity matrix according to an intersection-to-union ratio, and adaptively setting a community merging threshold value by using a global average value; clients are initialized into communities, high-similarity communities are iteratively merged, and the number of groups does not need to be preset; aggregating the average insurance premium and claim settlement of each community, and calculating a risk coverage ratio and homogeneity; and counting cross-community over-threshold similar customer pairs, comparing the cross-community over-threshold similar customer pairs with a mean value to identify potential association, and marking high-premium and high-claim communities. The process unifies and disperses data, eliminates dimensional differences, self-adapts threshold values and groups, generates community portraits, and supports precision marketing and joint risk control.

Description

Technical Field

[0001] This invention relates to the field of insurance information integration technology, specifically to a method for integrating insurance information. Background Technology

[0002] In existing insurance business systems, customer information, policy details, claims records, and risk classifications are often scattered across different business subsystems or third-party platforms, forming multiple isolated data silos. Traditional integration solutions mainly rely on timed batch ETL processes to load tables or files exported from various systems into a data warehouse sequentially, and then connect the data through predefined relational joins or field mappings.

[0003] First, there is a lack of unified standards in the format and field naming of various data sources. For example, different systems may have inconsistent field identifiers, data types, and statistical definitions for "premium amount" and "claims amount," resulting in the need to write a large number of targeted cleaning scripts during the ETL process. These scripts are costly to maintain and prone to failure due to changes in business rules. Second, traditional relational joins can only achieve one-to-one or one-to-many static field matching, making it difficult to handle many-to-many mapping relationships between policies and risk categories, and failing to intuitively express the complex semantic relationships between customers and multiple policies or risk categories. Third, in terms of customer segmentation, existing technologies mostly use binning or threshold segmentation methods based on numerical indicators—for example, dividing customers into several levels according to different ranges of annual premiums or claims rates, or directly applying algorithms such as K-means and hierarchical clustering to cluster numerical features. However, these methods only focus on numerical differences, ignoring the semantic risk classification covered by the customer's policies; and traditional clustering methods often require pre-specifying the number of clusters or manually adjusting the similarity threshold, lacking adaptability to business scenarios and prone to problems such as over-segmentation or under-merging. Furthermore, community-level risk profiles are difficult to generate under the existing model. Even if customers are segmented, only the mean or distribution of each group in terms of a single dimension such as total premium and claims rate can be obtained. It is impossible to deeply characterize the distribution and homogeneity of risk types within the group, let alone automatically discover potential correlations between groups to support cross-group precision marketing or joint risk warning.

[0004] Therefore, this case aims to propose an insurance information integration method based on the customer-policy-risk ternary relationship. It uses a set of customer risk tags as a foundation and a self-developed set operation and similarity measurement algorithm to automatically discover customer communities. Then, it aggregates premiums, claims and risk distribution at the community level, and finally mines potential connections between communities to generate community profiles that can be directly accessed, supporting precise insurance marketing and risk warning. Summary of the Invention

[0005] This invention provides a method for integrating insurance information, which helps to solve the problems mentioned in the background art.

[0006] This invention provides the following technical solution: a method for integrating insurance information, comprising: Raw data acquisition and field definition: Access the customer information database, read all customer records, and count the total number of customers. ; Record the customer identifier as Among them, when hour, Indicates the first One customer; For each customer Read the annual premium field and record it as Read the annual total claims amount field and record it as... ; Calculate the global extremum: , ;in, Minimum premium; Maximum premium; , ;in, Minimum claim amount; Maximum claim amount; Numerical normalization: based on and right Normalization is performed to standardize the annual premium. ;based on and right Normalization is performed to obtain standardized annual claims. ; Establish a multi-layered relationship mapping between customers and risk categories based on customers, their insurance policies, and the risk categories corresponding to those policies; Based on the overlap of risk categories among customers, a similarity matrix between customers is calculated, and a global average similarity threshold is determined. Each customer is treated as an initial community, and communities are gradually merged based on the similarity between customers until the termination condition is met; For customers within each final community, their premium and claims data are aggregated to generate community-level aggregated metrics; Analyze the risk category distribution of each community, and calculate the risk distribution vector and homogeneity index; Assess potential connections between communities and output insurance risk integration results and aggregated indicators.

[0007] Optionally, the numerical normalization specifically includes: For any customer Perform the following steps in sequence: S1. Premium Normalization: ; S2. Claims Normalization: ; S3, Verification Scope: confirm and Is it in Inside; if and Not here Inside, it is set to the most critical endpoint value; if and exist If the result is within the range of S2 or S3, then the value calculated in step S2 or S3 is taken.

[0008] Optionally, the step of establishing a multi-layered relationship mapping between customers and risk categories based on their policies and the corresponding risk categories specifically includes: Read the total number of policies and record it as follows. The policy is identified as follows: ; Among them, when hour, For the first Zhang's insurance policy; Read the total number of risk categories, and record it as The risk categories are identified as follows: ; Among them, when hour, For the first Class of risks; Construct the customer policy side set as follows: ; The policy risk category edge set is constructed as follows: ; For each customer Perform the following steps in sequence: S4. Initialize the empty set ;in, For customers A set of risk categories; S5, for all and all implement ; S6. After deduplication Record as customer The complete set of risk categories.

[0009] Optionally, the step of calculating a similarity matrix between customers based on the overlap of risk categories and determining a global average similarity threshold specifically includes: For each different client ,make: ;in, For customers and Similarity of risk categories between them; Constructing a matrix Diagonal elements are set to 1, and off-diagonal elements are set to 1. ;in Let be the set of real numbers. for OK, A set of real matrix columns; The global average similarity is calculated as follows: ;in, This represents the global average similarity.

[0010] Optionally, the step of treating each customer as an initial community and gradually merging communities based on the similarity between customers until a termination condition is met includes: Each customer As a single community Initial number of communities ; For any community Let the average similarity between communities be: ; Repeat the following steps until no community pairs meet the merge criteria: S7. Among all community pairs, select those that meet the following conditions. The pair set; S8. If the set is not empty, then identify the largest. right ; S9, will Merge into a new community, update the community set and the number of communities; S10. Return to calculate the average similarity between communities; S11. If no pair satisfies the condition, terminate the iteration and output the final community set. .

[0011] Optionally, for each customer within a final community, the premium and claims data are aggregated to generate community-level aggregated metrics, specifically including: For each end community Perform the following steps in sequence: S12, Premium Indicator Aggregation: ;in, For the community All The arithmetic mean; S13, Claims Indicator Aggregation: ;in, For the community All The arithmetic mean.

[0012] Optionally, the analysis of the risk category distribution for each community, and the calculation of the risk distribution vector and homogeneity index, specifically includes: For each risk category The risk distribution is calculated as follows: ;in, For the community Includes risk categories Customer ratio; Set the homogeneity index as follows: ;in, Based on distribution vector A measure of homogeneity.

[0013] Optionally, the assessment of potential correlations between communities, outputting insurance risk integration results and aggregated indicators, specifically includes: For any different community To calculate the number of cross-community connections, specifically: ;in, For the community and Between The number of customer pairs; The average number of cross-community connections is calculated as follows: ;in, This represents the average number of connections across all different communities. like If so, then the marker indicates a potential association; like If so, no marking will be performed; Calculate the global average premium as ; Calculate the global claims average as ; For each community : when At that time, it was marked as a high-premium community; when At that time, it was marked as a high-claims community.

[0014] The present invention has the following beneficial effects: 1. Perform a one-time, batch, structured read of all customer, premium, and claims data, with unified definitions for customer identifiers, premiums, and claims. This ensures subsequent processing is based on a consistent and complete customer dataset and indicator fields. It avoids statistical errors caused by inconsistencies in data exported from multiple systems or differences in data definitions. This replaces manual export or piecemeal script extraction, improving data integrity and consistency.

[0015] 2. Min-max normalization is applied to both premiums and claims, and the output is defined as zero when the range is zero, ensuring the completeness of the function's domain. This eliminates the influence of different units, unifying the two indicators to the [0,1] range, facilitating subsequent similarity calculations. It prevents program anomalies or numerical distortions caused by zero denominators in traditional normalization. It is superior to simple standardization methods that do not handle outlier denominators and is also more robust to outliers than Z-score.

[0016] 3. Through customer-policy side aggregation With the policy—risk edge set By combining and operating on the sets, a risk label set for each customer can be directly generated. This system completes many-to-many mapping in one step, eliminating the need for nested loops or complex SQL multi-table joins. It addresses the disconnect between policies and risk categories, building a complete risk profile for each customer. This differs from traditional implementations that require mapping risk item by item, which are inefficient.

[0017] 4. Construct a symmetric matrix Then use the average value of the entire matrix This serves as a merging threshold. It quantifies the degree of risk overlap, automatically adapting the threshold to the data distribution without requiring expert experience for setting. It eliminates subjective threshold selection, preventing over- or under-segmentation of groups. Unlike methods with fixed thresholds or those requiring manual parameter tuning, it achieves adaptive, data-driven results.

[0018] 5. Initially, each customer is considered a community, and the average similarity function between communities is used. Greedy merge all The algorithm finds the optimal pairs of clusters until no more pairs can be merged. It automatically generates optimal community partitions based on data density and similarity distribution without requiring a preset number of communities. It overcomes the limitations of k-means and other clustering algorithms that require manually specifying the number of clusters, automatically generating the number of communities based on the actual risk structure. Unlike clustering algorithms that require repeated expert adjustments, it achieves zero-parameter tuning for community discovery.

[0019] 6. For each community Calculate the average premium directly Compared with average claims It rapidly quantifies group value and risk costs, generating community-level metrics. It provides community-level comparisons of premiums and claims, supporting the development of segmentation strategies. This is superior to community statistical methods that require manual processing or secondary development.

[0020] 7. Define the risk coverage ratio vector and homogeneity indicators It accurately measures the concentration of community risk. It simultaneously depicts the diversity and concentration of risks, guiding product pricing and risk control priorities. It identifies homogeneous risk groups, preventing concentrated outbreaks of group-wide losses. This is superior to simply reporting a rough profile of the "main risk category."

[0021] 8. Based on cross-community over-threshold similarity customer logs Compared with the average Automatically identifies potential connections between communities and generates unified community profile tables. Reveals deep connections between customer groups, providing specific targets for cross-selling and coordinated risk control. Fills the gap in multi-community collaborative value mining, improving the overall synergistic efficiency of marketing and risk control. Unlike isolated community analysis, it is the first to quantify and automatically label relationships between communities, avoiding reliance on manual experience. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of the process of the present invention. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Example, refer to Figure 1 An insurance information integration method, comprising: Raw data acquisition and field definition: Identify the object of analysis and obtain the raw values ​​to prepare data for subsequent normalization and similarity calculation; Access the customer information database, read all customer records, and count the total number of customers. ; Record the customer identifier as Among them, when hour, Indicates the first One customer; For each customer Read the annual premium field and record it as , indicating customers The total premium paid within the statistical period (e.g., one year), in yuan; the annual total claims amount is read and recorded as... , indicating customers The total amount of claims received within the same statistical period, expressed in "yuan"; key numerical characteristics of each customer are obtained to measure customer value and risk; Calculate the global extremum: , ;in, Minimum premium; Maximum premium; , ;in, Minimum claim amount; Maximum claim amount; Determine the upper and lower bounds required for normalization to ensure subsequent mappings to... scope; Numerical normalization: based on and right Normalization is performed to standardize the annual premium. ;based on and right Normalization is performed to obtain standardized annual claims. ; Establish a multi-layered relationship mapping between customers and risk categories based on customers, their insurance policies, and the risk categories corresponding to those policies; Based on the overlap of risk categories among customers, a similarity matrix between customers is calculated, and a global average similarity threshold is determined. Each customer is treated as an initial community, and communities are gradually merged based on the similarity between customers until the termination condition is met; For customers within each final community, their premium and claims data are aggregated to generate community-level aggregated metrics; Analyze the risk category distribution of each community, and calculate the risk distribution vector and homogeneity index; Assess potential connections between communities and output insurance risk integration results and aggregated indicators.

[0025] By acquiring raw data and defining fields, the issues of scattered and inconsistent insurance data formats were resolved, providing a unified and standardized foundation for subsequent normalization and similarity calculations. This step clarified important fields such as customer, premium, and claims, ensuring that key numerical characteristics for each customer were effectively read and defined. This established a clear data structure for subsequent processing and calculations, avoiding errors caused by data inconsistencies. The global extremum calculation step determined the upper and lower bounds of premium and claims data, resolving the issue of data scale differences between different customers and providing a standardized basis for numerical normalization. This operation ensured that subsequent normalization processing could be performed within the same range, enabling unbiased comparison and calculation of various indicators. It effectively eliminated biases caused by different dimensions of data, thus providing a unified and fair foundation for subsequent calculations. The numerical normalization step resolved the impact of differences in data dimensions, allowing annual premiums and claims amounts to be compared and processed under the same standard. Standardizing premiums and claims to the same range makes the data more comparable, making similarity calculations unrestricted by data scale, laying the foundation for subsequent customer similarity analysis and community segmentation. By establishing a multi-layered relationship mapping between customers and risk categories, the complex relationship management problem between customers, policies, and risks is solved. This step associates each customer with their held policies and related risk category information, forming a comprehensive and multi-layered customer risk profile, making subsequent analysis more refined and accurate. By calculating a similarity matrix between customers, the problem of capturing complex relationships between customers in traditional data processing is solved. Based on the overlapping relationships of risk categories, the similarity matrix provides a quantitative basis for subsequent customer grouping, community merging, and potential connection discovery. The determination of a global average similarity threshold makes the merging process more accurate and avoids the instability caused by subjectively set thresholds. By treating each customer as an initial community and merging them step by step, the flexibility problem of customer group segmentation is solved. Traditional group segmentation often relies on manually set thresholds or overly simplistic numerical comparisons, while the dynamic community merging method allows community segmentation to be automatically performed based on the actual similarity between customers, without the need to pre-set the number of groups, avoiding the problems of over-segmentation or under-merging. By aggregating the premium and claims data of customers within a community, community-level aggregation indicators are generated, solving the problem of how to analyze customer risk and value at the group level. Aggregated metrics for each community effectively showcase its overall characteristics, providing a quantitative basis for subsequent risk control and marketing strategies. By analyzing the risk category distribution of each community and calculating risk distribution vectors and homogeneity indicators, the problem of gaining a deeper understanding of the internal structure of each community is addressed. The distribution of risk categories and homogeneity indicators can help identify areas of concentrated risk within a community, thereby providing data support for risk management and product design.By assessing the potential connections between communities, the system outputs insurance risk integration results and aggregated indicators, solving the challenge of uncovering potential value across different communities. Calculating cross-community connections effectively identifies links between groups, supporting targeted marketing or joint risk assessment across groups and improving the overall effectiveness of insurance business.

[0026] In summary, this solution overcomes the problems of data silos, inaccurate customer segmentation, and imprecise risk assessment in traditional insurance information processing through a series of data processing, standardization, analysis, and merging steps. It achieves adaptive community discovery and risk analysis based on customer risk tags, providing innovative technical support for precision marketing and risk management in the insurance industry.

[0027] The numerical normalization specifically includes: Mapping premium and claim figures to a unified scale eliminates dimensional differences and facilitates subsequent similarity measurement; For any customer Perform the following steps in sequence: S1. Premium Normalization: ;Will Mapped to To obtain standardized premium indicators ; S2. Claims Normalization: ;Will Mapped to To obtain standardized premium indicators ; S3, Verification Scope: confirm and Is it in Inside; if and Not here Inside, it is set to the most critical endpoint value; if and exist If the value is within the range, then the value calculated in step S2 or S3 is taken. Ensure that the normalization results meet the expected range and that no outlier values ​​affect subsequent calculations.

[0028] The numerical normalization step resolves the issue of dimensional differences between premium and claim values, providing data with a unified scale for subsequent similarity calculations. Specifically, in step S1, the premium value for each customer is normalized, mapping it to a uniform range (e.g., 0 to 1). This normalization process allows premium values ​​from different customers to be compared under the same standard, eliminating differences in dimensions and scales, and ensuring that subsequent calculations are not distorted or erroneous due to inconsistencies in dimensions. In step S2, claim values ​​are normalized, making claim amounts from different customers comparable on the same scale. This process not only eliminates dimensional differences between data but also ensures consistency in the comparison between claim data and premium data, thus providing a reliable data foundation for subsequent similarity calculations. The normalized data ensures that the weights between different data dimensions are no longer biased due to differences in numerical magnitude, guaranteeing that the features of all customers are evenly included in the subsequent calculation process. The verification range operation in step S3 ensures the validity and usability of the normalized data. In this step, the system verifies whether the normalized data for each customer is within a preset valid range (e.g., 0 to 1). If data is found to be outside this range, it is set as the most critical endpoint value to avoid unreasonable data interference. If the data is within the valid range, the normalized value is used directly. This operation effectively prevents extreme outliers from affecting subsequent calculations, ensuring the consistency and stability of the entire dataset.

[0029] In summary, numerical normalization resolves the issue of inconsistent data dimensions, ensuring that different types of customer data are compared under the same standard, thereby eliminating the impact of differences in data scale. Furthermore, normalization improves the accuracy of subsequent similarity calculations and community segmentation, providing a reliable data foundation for analysis throughout the entire insurance information integration process.

[0030] The process of establishing a multi-layered relationship mapping between customers and risk categories, based on the customer's insurance policies and the corresponding risk categories, specifically includes: Construct a mapping relationship between customers and risk categories to provide a risk vector basis for similarity calculation; Read the total number of policies and record it as follows. The policy is identified as follows: ; Among them, when hour, For the first Zhang's insurance policy; Read the total number of risk categories, and record it as The risk categories are identified as follows: ; Among them, when hour, For the first Class of risks; Define the vertex set range of the edge set, and define the policy and risk entity space; Construct the customer policy side set as follows: ; The policy risk category edge set is constructed as follows: ; The relationship between a customer and a policy, and between a policy and a risk category, is recorded in the form of edge sets. For each customer Perform the following steps in sequence: S4. Initialize the empty set ;in, For customers A set of risk categories; S5, for all and all implement ; S6. After deduplication Record as customer A complete set of risk categories; By mapping each customer to all the risk categories covered by their insurance policies, a set of customer risk labels can be obtained. .

[0031] By constructing customer-policy edge sets and policy-risk category edge sets, and executing steps S4–S6 to generate a complete set of customer risk labels, this method solves the problem of inaccurate correlation between many-to-many relationships between customers and multiple policies and risk categories in traditional methods, achieving a complete mapping of the risks covered by each customer's policies. This multi-layered relationship mapping eliminates information loss caused by data silos or simple field associations, resulting in a complete set of risk labels for each customer. Accurate and comprehensive, this laid a solid foundation for subsequent risk analysis. This was achieved by sequentially initializing the empty set for each customer. traverse all and Edge set merging solves the problems of low efficiency and error-proneness caused by manual multi-table joins and repetitive calculations, enabling automated and batch generation of the complete risk label set. The deduplicated risk label set is free of redundancy, ensuring the purity and accuracy of risk vectors input into the similarity matrix calculation, and avoiding similarity calculation bias caused by redundant labels. Through a multi-layer relation mapping method, the limitations of single-dimensional or numerical clustering in representing risk semantics are overcome, achieving the innovation of using risk labels as customer feature vectors. This mapping method not only enhances the model's ability to perceive customer risk preferences but also enables subsequent similarity measurements based on risk labels to more accurately reflect the degree of risk overlap between customers, thereby significantly improving the business relevance and effectiveness of community discovery and group segmentation.

[0032] By systematizing and streamlining the mapping steps described above, the subjectivity problem of relying on expert experience to set data association rules in existing technologies is solved, realizing a fully data-driven risk mapping process. The mapping rules only rely on existing policy-risk relationships in the database, reducing reliance on external manual intervention, improving the repeatability and maintainability of the integration process, and providing reliable assurance for insurance companies to quickly deploy and iterate.

[0033] The calculation of a similarity matrix between customers based on the overlap of risk categories and the determination of a global average similarity threshold specifically includes: Quantify the degree of overlap in risk categories among customers to generate similarity thresholds for community segmentation; For each different client ,make: ;in, For customers and The similarity of risk categories between the two clients is measured using the intersection-union ratio (IUU) to measure the overlap of their risk sets and generate a symmetric similarity value. Constructing a matrix Diagonal elements are set to 1, and off-diagonal elements are set to 1. ;in Let be the set of real numbers. for OK, A collection of real-valued matrices; centrally stores the similarity between all customer pairs for easy subsequent community calculation and retrieval; The global average similarity is calculated as follows: ;in, The global average similarity is used as the benchmark threshold for community merging.

[0034] Similarity is calculated for each pair of customers based on the crossover ratio (CLO) of risk categories. This process addresses the problem that traditional clustering methods relying solely on premium or claim rate values ​​cannot reflect the semantic overlap of customer risks. It effectively quantifies the degree of correlation among customers at the risk label level, thus providing a more business-relevant similarity basis for community discovery. By combining all... The operation of filling a symmetric matrix with values ​​and setting the diagonal elements to 1 solves the problems of scattered similarity calculation and difficulty in unified management, enabling centralized storage and fast retrieval of similarity results. This provides efficient support for frequent similarity queries in subsequent community merging algorithms. The global average value of all off-diagonal elements in the similarity matrix is ​​calculated. As a step for setting the threshold for community merging, this approach addresses the subjectivity issue of previously requiring experience or repeated manual parameter tuning to specify the threshold. It enables automatic threshold determination based on data distribution characteristics, allowing the community segmentation process to adapt to the current customer risk structure and improving the robustness and repeatability of group segmentation.

[0035] By using the above similarity calculation and threshold determination process, risk semantics and algorithm logic are effectively unified, solving the problem that multi-dimensional data is difficult to integrate into a single feature for clustering. This allows subsequent iterative community merging operations to be executed under the guidance of precise and controllable thresholds, ultimately forming a customer community with high business relevance and strong risk homogeneity, significantly improving the hit rate and efficiency of insurance precision marketing and risk warning.

[0036] The process of treating each customer as an initial community and gradually merging communities based on the similarity between customers until a termination condition is met includes: Based on a similarity threshold, highly similar customers are gradually grouped into communities. Each customer As a single community Initial number of communities Provides the finest-grained community units, facilitating iterative merging; For any community Let the average similarity between communities be: ;Measures the average risk similarity between members within two communities; Repeat the following steps until no community pairs meet the merge criteria: S7. Among all community pairs, select those that meet the following conditions. The pair set; S8. If the set is not empty, then identify the largest. right ; S9, will Merge into a new community, update the community set and the number of communities; S10. Return to calculate the average similarity between communities; S11. If no pair satisfies the condition, terminate the iteration and output the final community set. ; Community detection is achieved by iteratively merging high average similarity community pairs until no more pairs can be merged.

[0037] By initializing each customer as a single community, the problem of traditional clustering methods requiring pre-specified group size is solved, effectively establishing the finest-grained community units. This operation allows the algorithm to explore from the minimum granularity without external parameters, adaptively determining the number of communities and avoiding over-segmentation or under-merging that might result from manual setting. This is achieved by defining an average similarity function between communities. This approach addresses the bias inherent in traditional methods that only compare representative community points. The function comprehensively considers the risk similarity among all members of two communities, accurately measuring the overall correlation between groups and providing a scientific basis for subsequent merging. By screening all community pairs that meet the criteria… Step S7 addresses the problem of low efficiency in judging massive community pairs by retaining only candidate merge pairs with an adaptive threshold, significantly reducing the search space for merge decisions and improving algorithm efficiency. Step S8 solves the problem of difficulty in making decisions when multiple community pairs simultaneously meet the merge conditions by selecting the community pair with the highest average similarity from the candidate pair set. This greedy strategy prioritizes merging the two most similar groups, improving the cohesion and risk homogeneity of the merged community and ensuring that the community structure is more in line with business risk tendencies. Step S9 solves the problem of unclear group boundaries caused by static allocation by merging selected community pairs into a new community. Dynamic merging ensures that the internal member relationships of the community can be re-evaluated after each iteration, allowing the group to grow continuously and the structure to become clearer, thereby generating a high-quality customer community. Step S10 solves the problem of outdated similarity of residual communities after merging by recalculating the similarity of all community pairs by returning to the previous step. This operation triggers a global similarity update with each merge, reflecting changes in the community structure in real time, ensuring that subsequent merges are based on the latest group relationships, and enhancing the stability and accuracy of the algorithm. By terminating the iteration when no community pairs meet the merging conditions (step S11), the risk of excessive merging potentially destroying the natural community structure is addressed. This termination mechanism, based on the self-consistency judgment of an adaptive threshold, can automatically determine that a community has reached its optimal partitioning state, avoiding repeated manual verification and ensuring that the final number of communities is highly consistent with the inherent structure of the data.

[0038] This iterative merging process solves a series of problems such as the need to preset the number of clusters in clustering, ignoring risk semantics, low merging efficiency, and distortion of group structure through fine-grained initialization, adaptive thresholds, greedy parallel merging, and dynamic updates. It ultimately generates a highly homogeneous and business-related customer community, providing a solid data foundation and decision support for precision marketing, personalized pricing, and joint risk warning.

[0039] For each customer within a final community, their premium and claims data are aggregated to generate community-level aggregated metrics, specifically including: Calculate the average standardized premium and claims for customers in each community to characterize community-level features; For each end community Perform the following steps in sequence: S12, Premium Indicator Aggregation: ;in, For the community All The arithmetic mean is used to obtain the community-level premium level, which is used to distinguish high-premium groups. S13, Claims Indicator Aggregation: ;in, For the community All The arithmetic mean is used to obtain the community-level claims level, which is used to distinguish high-claims groups.

[0040] Step S12, premium index aggregation, solves the problem of scattered and difficult-to-understand premium data at the single customer level, effectively aggregating the standardized premium values ​​of all customers in the community into a single average. This operation eliminates individual noise from premium differences among each customer, making premium levels at the community level readily apparent and facilitating comparisons of premium contributions across different communities on a uniform scale. Simultaneously, it provides a quantitative basis for the subsequent automatic identification of high-premium communities and tiered marketing strategies, achieving a smooth transition from the "customer" to the "community" dimension. Through the aggregation of claims indicators in step S13, the problem of individual customer claims fluctuations misleading the overall community assessment is resolved, effectively aggregating standardized claims values ​​for all customers within the community into a single average. This aggregation operation, on the one hand, eliminates the extreme impact of individual large claims on overall risk assessment, and on the other hand, provides a clear benchmark for distinguishing high-claim groups; the quantification of community-level claims levels enables risk control departments to quickly locate high-claim communities and carry out targeted risk management and pricing adjustments, thereby improving risk management efficiency.

[0041] By aggregating community-level metrics in two steps (S12 / S13), the problem of "overly granular customer dimensions making it difficult for businesses to grasp group profiles" was solved, forming a dual-metric system that reflects both the overall value of the community and internal differences. Ultimately, the aggregated community premium and claims metrics not only simplified the data structure but also supported the accurate identification of communities with "high premiums but low claims" or "high claims with potential high risk," significantly improving the decision-making efficiency of insurance precision marketing and risk warning.

[0042] The analysis of risk category distribution in each community, and the calculation of risk distribution vectors and homogeneity indicators, specifically include: Quantify the distribution characteristics and uniformity of risk categories within the community; For each risk category The risk distribution is calculated as follows: ;in, For the community Includes risk categories The proportion of customers; measuring the coverage of risk categories in the community, used for comparison between profiles; Set the homogeneity index as follows: ;in, Based on distribution vector Homogeneity measure, when A value close to 1 indicates that the risk categories of customers within the community are highly concentrated and homogeneous.

[0043] By employing risk distribution vector calculation steps, the problem of unclear risk structure within the community is resolved, effectively quantifying the coverage ratio of each risk category in the community into a vector. In this step, for each community Each risk category calculate This operation upgrades community profiling from a "single indicator" to a "multi-dimensional distribution," intuitively showcasing the coverage and relative importance of various risks within a community. It solves the problem that relying solely on average claims rates or total premiums cannot capture risk diversity, making comparisons of risk structures across different communities readily apparent. Through homogeneity indicator calculation steps, it addresses the difficulty in measuring the uniformity of risk categories within a community, effectively generating homogeneous indicators. .when When the value approaches 1, it indicates a high concentration of risk labels among customers within the community; when... When the value approaches its minimum, it indicates that the risk categories are evenly and diversely distributed. This indicator not only quantifies the degree of risk concentration but also provides an intuitive basis for identifying "high-risk homogeneous groups," assisting risk control departments in quickly identifying target groups that require focused monitoring or higher premium pricing. Through a comprehensive analysis of both risk distribution and homogeneity indicators, it solves the problem of biased judgments arising from relying on a single risk indicator, achieving a three-dimensional portrayal of the overall community risk landscape. and This two-dimensional approach reveals both a community's preference for specific risk categories and assesses the concentration or dispersion of its risk structure. This dual analytical framework overcomes the limitations of single distribution or homogeneity indicators, resulting in a more comprehensive and comparable risk profile, which facilitates the development of more precise differentiated products and risk control strategies.

[0044] The quantitative analysis described above significantly enhances insurance institutions' ability to understand the risk characteristics of customer groups, effectively supporting precise pricing and differentiated services. By incorporating risk distribution vectors and homogeneity indicators into community profiles, institutions can adjust rates, design coverage amounts or supplementary insurance products at the community level. They can also implement stricter underwriting measures or risk reserve provisions for highly homogeneous, high-risk communities, reducing payout volatility and concentrated outbreak risks, ultimately achieving the optimal balance between risk and return.

[0045] The assessment of potential connections between communities outputs insurance risk integration results and aggregated indicators, specifically including: Identify potential business connections between communities and generate community profile tables that can be used by business systems; For any different community To calculate the number of cross-community connections, specifically: ;in, For the community and Between The number of customer pairs; the number of customer pairs with risk similarity exceeding a threshold between different communities, reflecting the strength of community connections; The average number of cross-community connections is calculated as follows: ;in, The average number of connections across all different communities is used to determine a global baseline for potential associations. like If so, then the marker indicates a potential association; like If so, no marking will be performed; Identify highly interactive or highly similar community pairs to provide a basis for cross-community marketing or risk control; Calculate the global average premium as ; Calculate the global average claims amount ; For each community : when At that time, it was marked as a high-premium community; when At that time, it was marked as a high-claims community; Label each community with its premium and claims levels to facilitate business selection.

[0046] By calculating the number of cross-community connections, the problem of quantifying connections between different communities is solved. This step counts the number of customer pairs in each community that meet the risk similarity threshold. This effectively reflects the degree of risk overlap and customer interaction intensity between communities, thus providing accurate data support for subsequent identification of potential business collaborations or risk propagation paths. By calculating the average cross-community connection threshold, the subjective problem of manually setting community association thresholds is solved. This step uses the average number of connections across all communities. As a global benchmark, the criteria for determining potential connections between communities adapt to the current data distribution, improving the consistency and repeatability of connection identification and reducing parameter tuning costs during business deployment. The potential connection marking step solves the problem of the difficulty in intuitively discovering high-value connections between communities. When a community... Automatically marking communities as "potentially related" enables rapid filtering of highly interactive or similar community pairs. This allows cross-community marketing teams and risk control departments to quickly identify community combinations with collaborative development or joint early warning value, improving the efficiency and accuracy of marketing and risk control decisions. The global premium and claims average calculation steps address the lack of a unified benchmark for community tagging. This step calculates the global average standardized premium. With claims This allows subsequent determinations of "high premium" or "high claims" labels for each community to be based on the statistical characteristics of the data itself, rather than empirical thresholds, ensuring the objectivity and fairness of the labeling system and simplifying the community screening and tiered management process. The community labeling step also solves the problem of inconvenient integration between community profiles and business systems. After applying the "high premium" and "high claims" labels to each community sequentially, the generated community profile table contains a complete list of related communities and hierarchical indicators, which can be directly accessed by marketing, risk control, or product design systems. This achieves seamless integration from data analysis to business actions, significantly shortening business response time and improving operational efficiency.

[0047] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0048] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for integrating insurance information, characterized in that, include: Raw data acquisition and field definition: Access the customer information database, read all customer records, and count the total number of customers. ; Record the customer identifier as Among them, when hour, Indicates the first One customer; For each customer Read the annual premium field and record it as Read the annual total claims amount field and record it as... ; Calculate the global extremum: , ;in, Minimum premium; Maximum premium; , ;in, Minimum claim amount; Maximum claim amount; Numerical normalization: based on and right Normalization is performed to standardize the annual premium. ;based on and right Normalization is performed to obtain standardized annual claims. ; Establish a multi-layered relationship mapping between customers and risk categories based on customers, their insurance policies, and the risk categories corresponding to those policies; Based on the overlap of risk categories among customers, a similarity matrix between customers is calculated, and a global average similarity threshold is determined. Each customer is treated as an initial community, and communities are gradually merged based on the similarity between customers until the termination condition is met; For customers within each final community, their premium and claims data are aggregated to generate community-level aggregated metrics; Analyze the risk category distribution of each community, and calculate the risk distribution vector and homogeneity index; Assess potential connections between communities and output insurance risk integration results and aggregated indicators.

2. The insurance information integration method according to claim 1, characterized in that, The numerical normalization specifically includes: For any customer Perform the following steps in sequence: S1. Premium Normalization: ； S2. Claims Normalization: ； S3, Verification Scope: confirm and Is it in Inside; if and Not here Inside, it is set to the most critical endpoint value; if and exist If the result is within the range of S2 or S3, then the value calculated in step S2 or S3 is taken.

3. The insurance information integration method according to claim 2, characterized in that, The process of establishing a multi-layered relationship mapping between customers and risk categories based on their insurance policies and the corresponding risk categories includes: Read the total number of policies and record it as follows. The policy is identified as follows: ; Among them, when hour, For the first Zhang's insurance policy; Read the total number of risk categories, and record it as The risk categories are identified as follows: ; Among them, when hour, For the first Class of risks; Construct the customer policy side set as follows: ; The policy risk category edge set is constructed as follows: ; For each customer Perform the following steps in sequence: S4. Initialize the empty set ;in, For customers A set of risk categories; S5, for all and all implement ; S6. After deduplication Record as customer The complete set of risk categories.

4. The insurance information integration method according to claim 3, characterized in that, The calculation of a similarity matrix between customers based on the overlap of risk categories and the determination of a global average similarity threshold specifically includes: For each different client ,make: ;in, For customers and Similarity of risk categories between them; Constructing a matrix Diagonal elements are set to 1, and off-diagonal elements are set to 1. ;in Let be the set of real numbers. for OK, A set of real matrix columns; The global average similarity is calculated as follows: ;in, This represents the global average similarity.

5. The insurance information integration method according to claim 4, characterized in that, The process of treating each customer as an initial community and gradually merging communities based on the similarity between customers until a termination condition is met includes: Each customer As a single community Initial number of communities ; For any community Let the average similarity between communities be: ； Repeat the following steps until no community pairs meet the merge criteria: S7. Among all community pairs, select those that meet the following conditions. The pair set; S8. If the set is not empty, then identify the largest. right ; S9, will Merge into a new community, update the community set and the number of communities; S10. Return to calculate the average similarity between communities; S11. If no pair satisfies the condition, terminate the iteration and output the final community set. .

6. The insurance information integration method according to claim 5, characterized in that, For each customer within a final community, their premium and claims data are aggregated to generate community-level aggregated metrics, specifically including: For each end community Perform the following steps in sequence: S12, Premium Indicator Aggregation: ;in, For the community All The arithmetic mean; S13, Claims Indicator Aggregation: ;in, For the community All The arithmetic mean.

7. The insurance information integration method according to claim 6, characterized in that, The analysis of risk category distribution in each community, and the calculation of risk distribution vectors and homogeneity indicators, specifically include: For each risk category The risk distribution is calculated as follows: ;in, For the community Includes risk categories Customer ratio; Set the homogeneity index as follows: ;in, Based on distribution vector A measure of homogeneity.

8. The insurance information integration method according to claim 7, characterized in that, The assessment of potential connections between communities outputs insurance risk integration results and aggregated indicators, specifically including: For any different community Calculate the number of cross-community connections, specifically: ;in, For the community and Between The number of customer pairs; The average number of cross-community connections is calculated as follows: ;in, This represents the average number of connections across all different communities. like If so, then the marker indicates a potential association; like If so, no marking will be performed; Calculate the global average premium as ; Calculate the global average claims amount ; For each community : when At that time, it was marked as a high-premium community; when At that time, it was marked as a high-claims community.

Images