Customer resource data system and analysis method in insurance industry

Abstract

The invention relates to the technical field of insurance data processing, in particular to a customer resource data system and analysis method in the insurance industry, and the system comprises a data simplification module which is used for calculating expected values and variances of hidden variables and observation variables of each insurance record based on the insurance records of insurance customer data, the method comprises the following steps: comparing principal component vectors of insurance records, screening the vectors, calculating cosine similarity between the vectors, and integrating the vectors with high similarity to generate a principal component matrix; according to the invention, through deep analysis and optimization processing of the insurance customer data, the data precision and the decision reliability are improved. Through calculation of expected values and variances of hidden variables and observation variables in the insurance records, screening and high-similarity integration are carried out on principal component vectors, a simpler principal component matrix is formed, and the accuracy of data is enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of insurance data processing, and in particular to a customer resource data system and analysis method in the insurance industry. Background Art

[0002] The customer resource data system in the insurance industry aims to build a special customer resource data management system for insurance companies and provide corresponding analysis methods. The main purpose of the system is to help insurance companies effectively integrate and manage scattered customer data. This information can support the company to make more scientific and accurate decisions in product design, marketing, risk management, etc., thereby improving the company's competitiveness and customer service level.

[0003] However, existing technologies have problems such as insufficient data integration accuracy, low processing efficiency, and inaccurate risk assessment. Since key features cannot be effectively distinguished and extracted during data processing, there is a lot of redundant information in the analysis process, which increases data noise and affects the accuracy of subsequent decisions. In terms of the allocation and execution of analytical tasks, existing technologies often use sequential or single processing methods, which leads to inefficient processing and difficulty in adapting to the needs of large-scale data. Therefore, improvements are needed. Summary of the invention

[0004] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a customer resource data system and analysis method in the insurance industry.

[0005] In order to achieve the above-mentioned purpose, the present invention adopts the following technical scheme: A customer resource data system in the insurance industry comprises:

[0006] The data simplification module calculates the expected value and variance of the hidden variables and observed variables of each insurance record based on the insurance records of insurance customer data, compares the principal component vectors of the insurance records, screens the vectors, calculates the cosine similarity between the vectors, integrates the vectors with high similarity, and generates the principal component matrix;

[0007] A feature extraction module, based on the principal component matrix, extracts key features associated with the insurance behavior, calculates the first-order and second-order derivatives of each principal component vector, integrates the derivative values, and then compares the cumulative integral values, selects and marks the key features associated with the insurance behavior, performs hierarchical extraction based on the key features, and generates insurance behavior features;

[0008] A parallel parsing module, based on the characteristics of the insurance behavior, divides the data parsing task into multiple subtasks, assigns the subtasks to multiple parallel processing units, optimizes the parsing process by adjusting the order of the subtasks, and then integrates the parsing results of the subtasks to generate a customer risk mapping set;

[0009] The mapping optimization module constructs a dependency graph of insurance data based on the customer risk mapping set, traverses the nodes in the graph and performs a depth-first search, calculates the similarity between nodes, and hierarchizes and sorts the nodes according to the similarity, then optimizes the node structure, compares the optimized node set with the original data, and generates an optimized risk node set.

[0010] As a further solution of the present invention, the principal component matrix includes eigenvectors, eigenvalues ​​and corresponding weight coefficients of key insurance records, the insurance behavior characteristics include eigenvectors, cumulative integral values ​​and hierarchical annotations of key features, the customer risk mapping set includes risk feature nodes, relationships between nodes and risk levels, and the optimized risk node set includes hierarchical nodes, connection relationships between nodes and node similarities.

[0011] As a further solution of the present invention, the data reduction module includes:

[0012] The variable calculation submodule calculates the expected value and variance of the hidden variables and observed variables of each insurance record based on the insurance customer data, extracts the data distribution of each variable, performs corresponding mathematical operations, obtains the statistical results of each variable, and generates variable statistical results;

[0013] The vector screening submodule, based on the variable statistical results, screens high-weight vectors by comparing the principal component vectors of the insurance records one by one, and performs feature induction and classification to generate a representative vector set;

[0014] The similarity integration submodule calculates the cosine similarity between vectors one by one based on the representative vector set, integrates and merges vectors with high similarity by comparing similarity values, and generates a principal component matrix.

[0015] As a further solution of the present invention, the feature extraction module includes:

[0016] The derivative calculation submodule extracts key features associated with the insurance behavior based on the principal component matrix, calculates the first-order derivative and the second-order derivative of each principal component vector in turn, and accumulates and summarizes the derivative values ​​to generate derivative calculation results;

[0017] The integral comparison submodule performs cumulative partitioning processing on the derivative value based on the derivative calculation result, and screens representative integral segments by comparing the cumulative integral values ​​of the segments to generate a key feature set;

[0018] The hierarchical extraction submodule performs hierarchical processing on the selected key features based on the key feature set, grades the features according to their importance, and extracts the features of each level to generate insurance behavior features.

[0019] As a further solution of the present invention, the parallel parsing module includes:

[0020] A task division submodule divides the data analysis task into multiple subtasks based on the insurance behavior characteristics, and generates a subtask set by disassembling the task into independent parts and allocating them to multiple parallel processing units;

[0021] A sequence adjustment submodule, based on the subtask set, adjusts the subtask sequence, optimizes the execution sequence of the tasks by analyzing the dependencies between the tasks, and generates an optimized task sequence;

[0022] The result integration submodule integrates the analysis results of the subtasks based on the optimization task sequence, and generates a customer risk mapping set by summarizing and proofreading the results.

[0023] As a further solution of the present invention, the mapping optimization module includes:

[0024] The graph construction submodule constructs a dependency graph of the insurance data based on the customer risk mapping set, forms a complete dependency structure graph by identifying and connecting related data nodes, and generates a dependency graph node set.

[0025] The node hierarchical submodule performs hierarchical processing on the nodes in the dependency graph based on the dependency graph node set, calculates the similarity between the nodes, and sorts and hierarchizes the nodes according to the similarity to generate a hierarchical node set.

[0026] The structure optimization submodule optimizes the node structure in the graph based on the hierarchical node set, adjusts the connection mode of the nodes, and compares the optimized nodes with the original data to generate an optimized risk node set.

[0027] A method for analyzing customer resource data in the insurance industry, the method for analyzing customer resource data in the insurance industry is executed based on the above-mentioned customer resource data system in the insurance industry, and comprises the following steps:

[0028] S1, based on the insurance records of insurance customer data, the principal component analysis algorithm is used to calculate the expected value and variance of the hidden variables and observed variables of each insurance record, extract the principal component vector, screen the representative vector by comparing the differences between the vectors, calculate the cosine similarity between the vectors, and integrate the high-similarity vectors to generate the principal component matrix;

[0029] S2, based on the principal component matrix, using a feature extraction algorithm, calculating the first-order and second-order derivatives of each principal component vector, integrating the derivative values, selecting and marking key features associated with the insurance behavior by comparing the cumulative integral values, extracting the key features in layers, and generating insurance behavior features;

[0030] S3, based on the characteristics of the insurance behavior, adopts a parallel processing algorithm to divide the data parsing task into multiple subtasks and assign them to multiple parallel processing units. By adjusting the order of subtasks, the parsing process is optimized, and the parsing results of the subtasks are integrated to generate a customer risk mapping set. Then, through the graph structure optimization algorithm, a dependency graph of the insurance data is constructed, the nodes in the graph are traversed, the similarity between the nodes is calculated, and the nodes are layered and sorted according to the similarity, the node structure is optimized, and an optimized risk node set is generated.

[0031] Compared with the prior art, the advantages and positive effects of the present invention are:

[0032] In the present invention, the accuracy of data and the reliability of decision-making are improved through in-depth analysis and optimization of insurance customer data. By calculating the expected value and variance of hidden variables and observed variables in the insurance records, the principal component vectors are screened and integrated with high similarity to form a more streamlined principal component matrix, thereby enhancing the accuracy of the data. When extracting key features, the features that are highly correlated with the insurance behavior are accurately identified by comparing the derivative integral and the cumulative integral value, thereby ensuring the accuracy and pertinence of feature extraction. During the data analysis process, the task is divided into multiple subtasks for parallel processing, which optimizes the processing speed, and generates a more accurate customer risk mapping by integrating the analysis results. In the optimization stage, by constructing a dependency graph, calculating the similarity between nodes, and performing stratification and sorting, an optimized risk node set is finally generated. In summary, the present invention improves the efficiency and accuracy of data processing, enabling insurance companies to have stronger competitiveness and responsiveness in product design, marketing, and risk management. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 It is a system flow chart of the present invention. DETAILED DESCRIPTION

[0034] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0035] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, "multiple" means two or more, unless otherwise clearly and specifically defined.

[0036] Example

[0037] See also Figure 1 The present invention provides a technical solution: a customer resource data system in the insurance industry includes:

[0038] The data simplification module calculates the expected value and variance of the hidden variables and observed variables of each insurance record based on the insurance records of insurance customer data, compares the principal component vectors of the insurance records, screens the vectors, calculates the cosine similarity between the vectors, integrates the vectors with high similarity, and generates the principal component matrix;

[0039] The feature extraction module extracts the key features associated with the insurance behavior based on the principal component matrix, calculates the first-order and second-order derivatives of each principal component vector, integrates the derivative values, and then compares the cumulative integral values, selects and marks the key features associated with the insurance behavior, performs hierarchical extraction based on the key features, and generates insurance behavior features;

[0040] The parallel parsing module divides the data parsing task into multiple subtasks based on the characteristics of insurance behavior, assigns the subtasks to multiple parallel processing units, optimizes the parsing process by adjusting the order of the subtasks, and then integrates the parsing results of the subtasks to generate a customer risk mapping set;

[0041] The mapping optimization module builds a dependency graph of insurance data based on the customer risk mapping set, traverses the nodes in the graph and performs a depth-first search, calculates the similarity between nodes, and stratifies and sorts the nodes based on the similarity. It then optimizes the node structure, compares the optimized node set with the original data, and generates an optimized risk node set.

[0042] The principal component matrix includes the eigenvectors, eigenvalues ​​and corresponding weight coefficients of key insurance records. The insurance behavior characteristics include eigenvectors, cumulative integral values ​​and hierarchical labels of key features. The customer risk mapping set includes risk feature nodes, relationships between nodes and risk levels. The optimized risk node set includes hierarchical nodes, connection relationships between nodes and node similarities.

[0043] In the data simplification module, based on the insurance records of insurance customer data, the data format of each insurance record includes attribute information of multiple dimensions, such as the insured amount, insurance period, customer age, premium payment method, etc. First, the dimensional data is cleaned and normalized to eliminate the noise and outliers in the data, and then the expected value and variance of the hidden variables and observed variables of each insurance record are calculated. This process is achieved through statistical analysis of data distribution, mapping each insurance record to a high-dimensional space, and using the principal component analysis method to reduce the dimension of the high-dimensional data and extract the principal component vector that best represents the difference in insurance records. The principal component vector represents the main variability characteristics of the data. After the principal component is extracted, the principal component vectors are compared to screen out representative principal component vectors, and then the cosine similarity between the vectors is calculated to further identify and integrate vectors with high similarity, and finally generate the principal component matrix. The principal component matrix contains the eigenvectors, eigenvalues ​​and corresponding weight coefficients of key insurance records. The role of the principal component matrix is ​​to simplify complex high-dimensional data into an information set with fewer dimensions but retaining the main characteristics of the data, providing a basis for subsequent feature extraction.

[0044] In the feature extraction module, based on the principal component matrix, the key features related to the insurance behavior are firstly extracted by decomposing and calculating the eigenvectors in the principal component matrix. Each principal component vector represents a certain aspect of the insurance record. Firstly, the first-order derivative and second-order derivative of the principal component vector are calculated to analyze the change rate and acceleration of the insurance record in different dimensions. The cumulative change is accumulated by integral operation to obtain the cumulative integral value of each principal component. Then, the cumulative integral values ​​are compared to screen out the key features that are highly related to the insurance behavior. According to the key features, they are divided into different levels through hierarchical processing to generate insurance behavior features. The insurance behavior features include eigenvectors, cumulative integral values ​​and hierarchical labels of key features. The extraction of insurance behavior features enables the data to be further refined into a feature set that can explain customer behavior, providing detailed input information for risk assessment and behavior prediction.

[0045] In the parallel parsing module, based on the characteristics of insurance behavior, the data parsing task is first divided into multiple subtasks. The subtasks include assessing customer risks, recognizing patterns of insurance behaviors, and predicting potential risks. Each subtask is assigned to multiple parallel processing units by analyzing the data resources and computational complexity required. Then, by analyzing the dependencies and execution order between the subtasks, the execution order of the subtasks is adjusted to maximize the efficiency of parallel execution. During the execution of the subtasks, the processing units will parse different insurance behavior characteristics respectively, and integrate the parsed data into a unified customer risk mapping set. The mapping set describes the customer's risk distribution and correlation through multi-dimensional risk feature nodes and the relationships between nodes. The customer risk mapping set includes risk feature nodes, the relationships between nodes, and risk levels. The generation of this mapping set provides a structured data foundation for subsequent optimization analysis and risk prediction.

[0046] In the mapping optimization module, based on the customer risk mapping set, we first construct a dependency graph of insurance data to display the risk feature nodes and their relationships in the mapping set in the form of a graph. The nodes represent the customer's risk features, and the edges represent the associations between the nodes. We traverse all the nodes in the dependency graph through depth-first search, calculate the similarity between the nodes, and layer and sort the nodes according to the similarity. We cluster the nodes with high similarity together. Then, we optimize the structure of the dependency graph and adjust the connection method of the nodes to make the graph structure more reasonable and compact. Finally, we compare the optimized node set with the original data to generate an optimized risk node set. The optimized risk node set includes layered nodes, connection relationships between nodes, and node similarities. The optimized node set can more effectively reveal the correlation between customer risks and provide insurance companies with more accurate risk management and customer analysis tools.

[0047] In this embodiment, the data reduction module includes:

[0048] The variable calculation submodule calculates the expected value and variance of the hidden variables and observed variables of each insurance record based on the insurance customer data, extracts the data distribution of each variable, performs corresponding mathematical operations, obtains the statistical results of each variable, and generates variable statistical results;

[0049] The vector screening submodule, based on the variable statistical results, screens high-weight vectors by comparing the principal component vectors of the insurance records one by one, and performs feature induction and classification to generate a representative vector set;

[0050] The similarity integration submodule calculates the cosine similarity between vectors one by one based on the representative vector set, and integrates and merges the vectors with high similarity by comparing the similarity values ​​to generate the principal component matrix.

[0051] In the variable calculation submodule, the insurance records of insurance customer data are processed. The insurance records are stored in a table format. Each row represents the insurance information of a customer, including multiple fields such as age, gender, insurance amount, insurance period, etc. Each field represents an observed variable. In order to understand and process the data more accurately, it is first necessary to calculate the expected value and variance of the hidden variables and observed variables of each insurance record. This process is achieved through statistical methods. By normalizing the historical data of each variable, the dimensional differences between different variables are eliminated, and then the data distribution of each variable is extracted. Specifically, the numerical range of each variable is normalized. The data is divided into rows and a histogram or probability density function is generated to capture the distribution characteristics of the data. The expected value and variance are then calculated based on the distribution characteristics. The mean and variance formulas are used to calculate each variable separately. The generated expected value represents the central trend of the data, while the variance represents the degree of dispersion of the data. After the calculation is completed, the expected value and variance are integrated into a new data table to generate variable statistical results. The purpose of this statistical result is to provide basic data for the subsequent principal component analysis. Through the statistical results, the system can initially understand the characteristic distribution of each insurance record on each variable, thereby laying the foundation for more in-depth analysis.

[0052] In the vector screening submodule, the variable statistical results generated by the variable calculation submodule are further processed. The statistical results include the expected value and variance of each insurance record on different variables. The data is used as the input of principal component analysis. The goal of principal component analysis is to extract several main principal component vectors from high-dimensional data. The vectors can explain most of the variability in the data. In order to achieve this goal, the principal component vectors of each insurance record are first compared one by one. Specifically, by calculating the contribution rate of each principal component vector, the vectors with higher contribution rates are screened out. High-weight vectors usually explain most of the variance in the data and are the most important part of the data structure. Next, the high-weight vectors are feature summarized and classified. Feature summary refers to grouping and classifying vectors with similar features. By comparing the direction and length of the vector, similar vectors are classified into one category and a category label is assigned to each category. Finally, a representative vector set is generated. The generation of this representative vector set indicates that the data has been simplified from a high-dimensional complex space into several sets of important feature vectors. The vector set can provide concise and effective input for further data analysis, significantly reducing the complexity of calculation.

[0053] In the similarity integration submodule, the representative vector set generated by the vector screening submodule is deeply processed. The representative vector set contains high-weight vectors that have been screened and classified. In order to further optimize the data representation, it is first necessary to calculate the cosine similarity between vectors. Cosine similarity is a standard for measuring the similarity between two vectors. The similarity is determined by calculating the cosine value of the angle between the two vectors. In the specific operation, the cosine similarity is calculated for each pair of vectors. This operation is achieved by dividing the inner product of the vector by the product of the vector modulus. The calculation result is a value between -1 and 1. The closer the value is to 1, the more similar the two vectors are. Then the calculated similarity values ​​are compared, and the vectors with higher similarity are integrated and merged. Integration refers to merging multiple similar vectors into a new vector by weighted averaging. The merged new vector retains the main features of the original vector and reduces data redundancy. Finally, the principal component matrix is ​​generated. The principal component matrix contains all the integrated vectors and represents the most important features in the data set. This matrix not only simplifies the data structure, but also provides a clearer input basis for subsequent feature extraction and analysis.

[0054] In this embodiment, the feature extraction module includes:

[0055] The derivative calculation submodule extracts the key features associated with the insurance behavior based on the principal component matrix, calculates the first-order derivative and second-order derivative of each principal component vector in turn, and accumulates and summarizes the derivative values ​​to generate the derivative calculation results;

[0056] The integral comparison submodule performs cumulative partitioning on the derivative values ​​based on the derivative calculation results, and screens representative integral segments by comparing the cumulative integral values ​​of the segments to generate a key feature set;

[0057] The hierarchical extraction submodule performs hierarchical processing on the selected key features based on the key feature set, classifies them according to their importance, and extracts the features of each level to generate insurance behavior features.

[0058] In the derivative calculation submodule, the key features associated with the insurance behavior are extracted by processing the principal component matrix. The data format of the principal component matrix is ​​usually a two-dimensional matrix, with rows representing different insurance records and columns representing principal component vectors. Each principal component vector represents the characteristic change of the insurance record in one dimension. First, for each principal component vector, its first-order derivative and second-order derivative are calculated in turn. The first-order derivative is used to capture the rate of change of the insurance record in time or other independent variables. The calculation method is to make a difference between the adjacent elements of each principal component vector and then obtain the rate of change. The second-order derivative is used to capture the acceleration of the rate of change, that is, the directionality of the change trend. The second-order derivative is obtained by calculating the rate of change of the first-order derivative. Then, the values ​​of the first-order derivative and the second-order derivative of all principal component vectors are accumulated and summarized. This accumulation process can be achieved by gradually adding the derivative values, and finally generating the derivative calculation result. The result contains the derivative information of all principal component vectors, reveals the dynamic change characteristics of the insurance record in multiple dimensions, and provides the necessary data basis for further integral processing.

[0059] In the integral comparison submodule, the derivative calculation results generated by the derivative calculation submodule are further processed. The format of the derivative calculation results is also a two-dimensional matrix, which contains the derivative values ​​of each insurance record on each principal component vector. First, the derivative values ​​are processed by cumulative partitioning. The cumulative partitioning method is to group the derivative values ​​according to a certain interval and accumulate them one by one to obtain the cumulative integral value of each segment. The specific operation is to accumulate the data of each column in the matrix, divide it into several segments and calculate the integral value of each segment. The integral value can be achieved through numerical integration technology. Using the numerical integration method, the change of the derivative value is accumulated as a comprehensive value. Then, by comparing the cumulative integral values ​​of each segment, representative integral segments are screened out. The segment represents the most significant part of the data change, which is often closely related to the key characteristics of the insurance behavior. Finally, the screened segments are integrated into a new data set to generate a key feature set. This feature set contains the most important feature segments in the insurance behavior, laying a data foundation for subsequent stratified extraction.

[0060] In the hierarchical extraction submodule, the key feature set generated by the integral comparison submodule is hierarchically processed. The data format of the key feature set is usually a multidimensional matrix, and each dimension represents a feature segment. First, the segments are hierarchically processed according to the importance of the features. The importance of the features can be judged by the cumulative integral value of each segment and its contribution to the overall insurance behavior. The specific operation is to sort the feature segments in the key feature set according to the size of the cumulative integral value, and divide them into different levels, such as high, medium, and low, according to the preset threshold. Subsequently, the features in each level are extracted and stored in the form of a matrix, and finally the insurance behavior features are generated. The insurance behavior feature matrix contains the key features after hierarchical processing, which can provide structured input for subsequent data analysis and behavior prediction. Through this hierarchical extraction method, not only can the most significant features in the data be captured, but also the data can be effectively organized and managed according to importance, making the subsequent analysis process more efficient and accurate.

[0061] In this embodiment, the parallel parsing module includes:

[0062] The task division submodule divides the data analysis task into multiple subtasks based on the characteristics of insurance behavior, and generates a subtask set by breaking the task into independent parts and assigning them to multiple parallel processing units;

[0063] The sequence adjustment submodule adjusts the sequence of subtasks based on the subtask set, optimizes the execution sequence of tasks by analyzing the dependencies between tasks, and generates an optimized task sequence;

[0064] The result integration submodule integrates the analysis results of the subtasks based on the optimized task sequence, and generates a customer risk mapping set by summarizing and proofreading the results.

[0065] In the task division submodule, the data parsing task is divided into multiple subtasks by parsing the insurance behavior characteristics. Insurance behavior characteristics usually exist in the form of a multidimensional matrix, and each dimension represents different behavioral characteristics and associated information. First, the characteristics are classified, and the overall data parsing task is decomposed into multiple independent subtasks according to the attributes and complexity of the characteristics. The specific operation is to correspond each subtask to a specific feature dimension or its combination. For example, the analysis task involving characteristics such as customer age and premium payment method is divided into one subtask, and the task involving characteristics such as insured amount and insurance period is divided into another subtask. After the task is divided, the subtasks are assigned to multiple parallel processing units, and a load balancing strategy is adopted to ensure the optimal utilization of computing resources between each processing unit. The subtasks run independently in their respective processing units, without direct dependencies between each other, to generate a subtask set. The generation of such a subtask set allows the system to process multiple data tasks simultaneously in a distributed environment, greatly improving the parsing speed and system throughput, and at the same time providing a basis for the subsequent task sequence adjustment.

[0066] In the order adjustment submodule, the subtask set generated by the task division submodule is further optimized to ensure the optimal order of task execution. The format of the subtask set is multiple independent task collections, each of which contains specific data features and corresponding processing logic. First, by analyzing the dependencies between the subtasks, determine which tasks can be executed in parallel and which tasks need to be executed in sequence. The specific analysis method includes detecting the data dependency chain between tasks. For example, the output of a task may be the input of another task. In this case, the former needs to be arranged in front. According to the dependency relationship, the task scheduling algorithm is used to optimize the execution order of subtasks to ensure the optimal allocation and utilization of computing resources. The optimized task order is rearranged to generate an optimized task order. This order helps to reduce the waiting time and resource competition of the system, improve the overall efficiency of task execution, and lay an orderly foundation for the final result integration.

[0067] In the result integration submodule, the data results are integrated by adjusting the order of the optimized tasks generated by the order adjustment submodule. The execution results of each subtask are usually output in a structured data format, including risk feature values, classification labels, correlation scores, etc. First, the parsing results of the subtasks are integrated in the optimized order. The specific operation is to merge the results of each subtask in sequence to ensure the coherence and consistency of the data. During the integration process, the matching degree between the results of each task is checked to eliminate possible conflicts and redundant information. Through data cleaning and standardization technology, the format and unit of the result data are unified to ensure the accuracy and consistency of the data. Finally, all the integrated data are summarized into a complete customer risk mapping set, which is stored in a multidimensional data structure and contains all the parsed risk features and their relationships. The purpose of generating a customer risk mapping set is to provide insurance companies with a comprehensive risk analysis and prediction basis, thereby supporting more accurate risk management and customer stratification.

[0068] In this embodiment, the mapping optimization module includes:

[0069] The graph construction submodule builds the dependency graph of insurance data based on the customer risk mapping set, forms a complete dependency structure graph by identifying and connecting related data nodes, and generates a dependency graph node set.

[0070] The node stratification submodule performs stratification processing on the nodes in the graph based on the dependency graph node set, calculates the similarity between nodes, and sorts and stratifies the nodes according to the similarity to generate a stratified node set.

[0071] The structural optimization submodule optimizes the node structure in the graph based on the hierarchical node set, adjusts the connection method of the nodes, and compares the optimized nodes with the original data to generate an optimized risk node set.

[0072] In the graph construction submodule, the dependency graph of insurance data is constructed by processing the customer risk mapping set. The customer risk mapping set is usually stored in the form of a multidimensional data set, which contains the risk characteristics of each customer, the correlation between the characteristics, and the risk level information. When constructing the dependency graph, it is first necessary to extract each risk characteristic node from the risk mapping set and analyze the correlation between the nodes. Specifically, by identifying the mutual dependence between each node and other nodes, such as the causal relationship between risk characteristic A and characteristic B, the degree of mutual influence, etc., the correlation is converted into an edge in the graph. In this process, the correlation between each pair of nodes is calculated, and whether there is a direct connection between the nodes is determined based on the correlation. If so, an edge is added to the graph so that the relationship between the nodes can be clearly expressed. All nodes and their connection relationships are combined into a complete dependency structure graph, which fully presents the complex relationship between customer risk characteristics, and the position and connection mode of each node reflect its importance in the entire risk network. Finally, a dependency graph node set is generated, which includes all identified nodes and their positions and connection relationships in the dependency graph. This node set provides a basis for subsequent node hierarchical processing, so that the graph structure can be further optimized.

[0073] In the node stratification submodule, the nodes in the graph are stratified by deeply analyzing the dependency graph node set. Each node in the dependency graph node set represents a risk feature, and the edges between nodes represent the correlation between features. First, for each node, the level in the graph is determined by calculating its similarity with other nodes. The specific operation is to use cosine similarity to calculate the similarity between nodes. Cosine similarity measures the similarity by calculating the cosine value of the angle between the feature vectors of two nodes. The higher the similarity value, the more similar the features of the two nodes. Subsequently, according to the calculated similarity, the nodes are sorted, and the nodes with high similarity are classified into the same level, and stratified according to the similarity. Nodes at higher levels usually have stronger influence or more significant risk features. After stratification, a stratified node set is generated, which includes all nodes that have been stratified and their corresponding level information. This stratified node set makes the important nodes in the graph structure stand out, provides a stratified basis for subsequent structural optimization, and facilitates further optimization of the connection method between nodes.

[0074] In the structure optimization submodule, the node structure in the graph is optimized by analyzing and adjusting the hierarchical node set. The hierarchical node set provides the hierarchical information of the nodes and the corresponding connection relationship. The key to optimization is to adjust the connection mode of the nodes to make the graph structure more compact and efficient. In the specific operation, the connection mode of each node is first analyzed, and the redundant or lengthy connections are identified by checking the number of connections between nodes and the length of the connection path. For those redundant connections, the optimization algorithm is used to remove or reorganize them, and at the same time, the connection between important nodes and other nodes is increased to ensure their core position in the graph. Then, the connection mode of the nodes is reconfigured to shorten the path between nodes, reduce the complexity of information transmission, and improve the robustness of the network. Finally, the optimized node structure is compared with the original data to ensure that the optimized graph structure not only retains the important information in the original data, but also has stronger parsing ability and lower complexity. Finally, the optimized risk node set is generated, which includes all optimized nodes and their new connection modes. This optimized risk node set can more effectively represent the correlation between customer risks and provide a more accurate and efficient data basis for subsequent risk prediction and analysis.

[0075] The present invention provides a method for analyzing customer resource data in the insurance industry, comprising the following steps:

[0076] S1, based on the insurance records of insurance customer data, the principal component analysis algorithm is used to calculate the expected value and variance of the hidden variables and observed variables of each insurance record, extract the principal component vector, screen the representative vector by comparing the differences between the vectors, calculate the cosine similarity between the vectors, and integrate the high-similarity vectors to generate the principal component matrix;

[0077] S2, based on the principal component matrix, adopts the feature extraction algorithm to calculate the first-order and second-order derivatives of each principal component vector, integrates the derivative values, selects and marks the key features associated with the insurance behavior through the comparison of the cumulative integral values, extracts the key features in layers, and generates the insurance behavior features;

[0078] S3, based on the characteristics of insurance behavior, adopts a parallel processing algorithm to divide the data parsing task into multiple subtasks and assign them to multiple parallel processing units. By adjusting the order of subtasks, the parsing process is optimized, the parsing results of the subtasks are integrated, and a customer risk mapping set is generated. Then, through the graph structure optimization algorithm, a dependency graph of insurance data is constructed, the nodes in the graph are traversed, the similarity between nodes is calculated, and the nodes are layered and sorted according to the similarity, the node structure is optimized, and an optimized risk node set is generated.

[0079] The method for analyzing customer resource data in the insurance industry provided by the present invention firstly adopts the principal component analysis algorithm to perform dimension reduction processing on the insurance customer data, which can effectively extract the representative principal component vector, greatly reduce the complexity and redundancy of the data, and improve the efficiency and accuracy of data processing. By calculating the cosine similarity between vectors and integrating the high similarity vectors to generate the principal component matrix, the data can be further compressed and simplified while maintaining the core information, providing a solid foundation for subsequent feature extraction. On this basis, the feature extraction algorithm calculates the first-order and second-order derivatives of the principal component vectors, and integrates and cumulatively compares them, accurately identifies the key features associated with the insurance behavior, and structures the key features through hierarchical extraction, so that the analysis of the insurance behavior is more targeted and hierarchical, and the accuracy and depth of risk assessment are improved. Subsequently, the parallel processing algorithm effectively decomposes the complex data parsing task into multiple subtasks, and parallel processing improves the speed of data parsing. At the same time, by adjusting the task order to optimize the parsing process, the coordination between the subtasks and the consistency of the results are ensured, and the customer risk mapping set finally integrated provides a comprehensive risk view. The dependency graph and optimized risk node set constructed by the graph structure optimization algorithm clearly present and optimize the relationship between various risk features, which helps insurance companies better understand and manage customer risks. It not only improves the efficiency and accuracy of data analysis, but also enhances the ability of risk prediction and management, allowing insurance companies to more accurately stratify customers, develop personalized insurance plans, and make more scientific decisions in risk control, ultimately gaining greater advantages in the fierce market competition.

[0080] It should be understood that, although each step in the flow chart of each embodiment of the present invention is shown in sequence according to the indication of the arrow, these steps are not necessarily performed in sequence according to the order indicated by the arrow. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be performed in other orders. Moreover, at least a portion of the steps in each embodiment may include a plurality of sub-steps or a plurality of stages, and these sub-steps or stages are not necessarily performed at the same time, but can be performed at different times, and the execution order of these sub-steps or stages is not necessarily performed in sequence, but can be performed in turn or alternately with at least a portion of other steps or sub-steps or stages of other steps.

[0081] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0082] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A customer resource data system in the insurance industry, characterized in that: The system comprises: The data simplification module calculates the expected value and variance of the hidden variables and observed variables of each insurance record based on the insurance records of insurance customer data, compares the principal component vectors of the insurance records, screens the vectors, calculates the cosine similarity between the vectors, integrates the vectors with high similarity, and generates the principal component matrix; A feature extraction module, based on the principal component matrix, extracts key features associated with the insurance behavior, calculates the first-order and second-order derivatives of each principal component vector, integrates the derivative values, and then compares the cumulative integral values, selects and marks the key features associated with the insurance behavior, performs hierarchical extraction based on the key features, and generates insurance behavior features; A parallel parsing module, based on the characteristics of the insurance behavior, divides the data parsing task into multiple subtasks, assigns the subtasks to multiple parallel processing units, optimizes the parsing process by adjusting the order of the subtasks, and then integrates the parsing results of the subtasks to generate a customer risk mapping set; The mapping optimization module constructs a dependency graph of insurance data based on the customer risk mapping set, traverses the nodes in the graph and performs a depth-first search, calculates the similarity between nodes, and hierarchizes and sorts the nodes according to the similarity, then optimizes the node structure, compares the optimized node set with the original data, and generates an optimized risk node set.

2. The customer resource data system in the insurance industry according to claim 1, characterized in that: The principal component matrix includes the eigenvectors, eigenvalues ​​and corresponding weight coefficients of key insurance records. Including feature vectors, cumulative integral values ​​and hierarchical annotations of key features, the customer risk mapping set includes risk feature nodes, relationships between nodes and risk levels, and the optimized risk node set includes hierarchical nodes, connection relationships between nodes and node similarities.

3. The customer resource data system in the insurance industry according to claim 1, characterized in that: The data reduction module comprises: The variable calculation submodule calculates the expected value and variance of the hidden variables and observed variables of each insurance record based on the insurance customer data, extracts the data distribution of each variable, performs corresponding mathematical operations, obtains the statistical results of each variable, and generates variable statistical results; The vector screening submodule, based on the variable statistical results, screens high-weight vectors by comparing the principal component vectors of the insurance records one by one, and performs feature induction and classification to generate a representative vector set; The similarity integration submodule calculates the cosine similarity between vectors one by one based on the representative vector set, integrates and merges vectors with high similarity by comparing similarity values, and generates a principal component matrix.

4. The customer resource data system in the insurance industry according to claim 1, characterized in that: The feature extraction module comprises: The derivative calculation submodule extracts key features associated with the insurance behavior based on the principal component matrix, calculates the first-order derivative and the second-order derivative of each principal component vector in turn, and accumulates and summarizes the derivative values ​​to generate derivative calculation results; The integral comparison submodule performs cumulative partitioning processing on the derivative value based on the derivative calculation result, and screens representative integral segments by comparing the cumulative integral values ​​of the segments to generate a key feature set; The hierarchical extraction submodule performs hierarchical processing on the selected key features based on the key feature set, grades the features according to their importance, and extracts the features of each level to generate insurance behavior features.

5. The customer resource data system in the insurance industry according to claim 1, characterized in that: The parallel parsing module includes: A task division submodule divides the data analysis task into multiple subtasks based on the insurance behavior characteristics, and generates a subtask set by disassembling the task into independent parts and allocating them to multiple parallel processing units; A sequence adjustment submodule, based on the subtask set, adjusts the subtask sequence, optimizes the execution sequence of the tasks by analyzing the dependencies between the tasks, and generates an optimized task sequence; The result integration submodule integrates the analysis results of the subtasks based on the optimization task sequence, and generates a customer risk mapping set by summarizing and proofreading the results.

6. The customer resource data system in the insurance industry according to claim 1, characterized in that: The mapping optimization module includes: The graph construction submodule constructs a dependency graph of the insurance data based on the customer risk mapping set, forms a complete dependency structure graph by identifying and connecting related data nodes, and generates a dependency graph node set.

7. The customer resource data system in the insurance industry according to claim 6, characterized in that: Also includes: The node hierarchical submodule performs hierarchical processing on the nodes in the dependency graph based on the dependency graph node set, calculates the similarity between the nodes, and sorts and hierarchizes the nodes according to the similarity to generate a hierarchical node set.

8. The customer resource data system in the insurance industry according to claim 7, characterized in that: Also includes: The structure optimization submodule optimizes the node structure in the graph based on the hierarchical node set, adjusts the connection mode of the nodes, and compares the optimized nodes with the original data to generate an optimized risk node set.

9. A method for analyzing customer resource data in the insurance industry, characterized in that: The customer resource data system in the insurance industry according to any one of claims 1 to 8 is implemented, comprising the following steps: Based on the insurance records of insurance customer data, the principal component analysis algorithm is used to calculate the expected value and variance of the hidden variables and observed variables of each insurance record, extract the principal component vector, screen the representative vectors by comparing the differences between the vectors, calculate the cosine similarity between the vectors, and integrate the high-similarity vectors to generate the principal component matrix; Based on the principal component matrix, a feature extraction algorithm is used to calculate the first-order and second-order derivatives of each principal component vector, and the derivative values ​​are integrated. By comparing the cumulative integral values, key features associated with the insurance behavior are selected and marked, and the key features are extracted in layers to generate insurance behavior features; Based on the characteristics of the insurance behavior, a parallel processing algorithm is used to divide the data parsing task into multiple subtasks and assign them to multiple parallel processing units. The parsing process is optimized by adjusting the order of the subtasks, and the parsing results of the subtasks are integrated to generate a customer risk mapping set. Subsequently, a graph structure optimization algorithm is used to construct a dependency graph of the insurance data, traverse the nodes in the graph, calculate the similarity between the nodes, and layer and sort the nodes according to the similarity to optimize the node structure and generate an optimized risk node set.

Images