Member integrated management system based on big data analysis

Abstract

The invention belongs to the technical field of member management, and discloses a member integrated management system based on big data analysis. The method comprises the following steps: acquiring member data of insurance members; performing primary processing on the member data to obtain a member combination graph list; constructing a member state updating model based on the member combination diagram list; on the basis of the member state updating model, predicting the overall behavior trend of the member in the future T time periods, and quantifying the overall behavior trend as the risk exposure degree of the corresponding time period; according to the risk exposure degree, a corresponding differentiated marketing strategy is formulated, and fine management of members is realized; a closed loop of member full life cycle management can be constructed, systematization, intelligentization and automation of member management are realized, the labor cost is reduced, and the management efficiency is improved.

Description

Technical Field

[0001] This invention relates to the field of membership management technology, and more specifically, to a membership integrated management system based on big data analysis. Background Technology

[0002] Patent application CN116777633A discloses a data-based financial asset management system, relating to the field of asset management system technology. It includes a data acquisition module, a classification module, a tracking module, a risk management module, and a decision support module. The classification module categorizes assets based on their type and characteristics. The tracking module tracks asset-related data. The risk management module comprehensively analyzes the asset-related data, establishes a risk model, analyzes the investment risk of the assets through the risk model, and formulates corresponding risk control strategies. The data acquisition module collects and obtains necessary data from various asset data sources. The tracking module tracks asset-related data. The risk management module comprehensively analyzes the asset-related data, establishes a risk model, analyzes the risk exposure of the investment portfolio through the risk model, and formulates corresponding risk control strategies, thereby effectively predicting and assessing the risks of financial investment assets and ensuring the security of investment transactions.

[0003] In the insurance sector, current traditional membership management systems rely too heavily on static member information, such as age, gender, and insurance records, neglecting the dynamic changes in member behavior. Members' behavioral decisions are often influenced by complex social networks, which profoundly impact their behavior, but existing systems fail to fully consider these factors. Secondly, there is a lack of quantitative assessment capabilities for members' future behavior. Members' behavioral decisions are dynamically evolving and change over time, but existing systems cannot model and predict these dynamic changes, making it difficult to formulate forward-looking marketing strategies. Furthermore, due to the lack of analysis on the dynamic evolution trends of member behavior, existing systems struggle to develop refined and differentiated marketing strategies. In actual marketing practices, a one-size-fits-all approach is often adopted, promoting the same products and services to all members, ignoring the differentiated needs of different member groups, resulting in low targeting and conversion rates for marketing strategies.

[0004] In view of this, the present invention proposes a membership integrated management system based on big data analysis to solve the above problems. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art and to achieve the above objectives, the present invention provides the following technical solution: a membership integrated management system based on big data analysis, comprising: a data acquisition module, a processing module, a comprehensive evaluation module, and a strategy fitting module; the modules are connected to each other via wired and / or wireless means; The data acquisition module is used to collect member data of insurance members; The processing module is used to perform preliminary processing on member data to obtain a list of member combination charts; The model building module constructs a member status update model based on the member combination diagram list; The comprehensive assessment module, based on the member status update model, predicts the overall behavioral trends of members over the next T time periods and quantifies these trends into the risk exposure levels for the corresponding time periods. Based on these risk exposure levels, corresponding differentiated marketing strategies are developed.

[0006] Furthermore, the membership data includes personal information data, policy data, claims data, health data, geolocation data, and social relationship data.

[0007] Furthermore, the method for preliminary processing of member data includes: By integrating member data, a complete member profile is constructed, defining the types of relationships between members, including family relationships, social relationships, geographical relationships, policy relationships, and claims relationships; based on the relationship types, the corresponding relationships between members are extracted from the member profile; The extracted relationships are represented as an undirected graph G=(V,E), where V is the set of nodes, representing members; E is the set of edges, representing the relationships between members. New relationships are mined from the undirected graph G, and for each new relationship, it is added to the original undirected graph to obtain a complete relationship network. Member characteristics are extracted from the original member data, including demographic characteristics, behavioral characteristics, and risk characteristics. The member characteristics are standardized and discretized to obtain the processed member characteristics. The processed member characteristics and the relationship network are combined to obtain a list of member combination diagrams.

[0008] Furthermore, the methods for integrating member data include: Define the core entities and related entities of the member profile, and determine the relationship between the core entities and related entities; for each core entity, extract the corresponding data records from the member data, and identify several data records of the same core entity in different data types in the member data; identify related entities from the identified data records, and associate the related entities with the core entities to generate a fused view of the corresponding core entities; that is, the complete member profile.

[0009] Furthermore, the method for mining new associations from an undirected graph G includes: For each node in the undirected graph Calculate its comprehensive centrality ; ;in, For nodes The set of directly connected neighboring nodes. For nodes Nodes within the set of directly connected neighboring nodes; For nodes The degree; The total number of nodes; The maximum degree of all nodes; For nodes and nodes The strength of the correlation between them; ;in, , , and These are the weight coefficients of the corresponding terms, and their sum is 1; For nodes and nodes Similarity scores between attribute vectors; For nodes and nodes The number of interactions between them For attenuation parameters, Represents distance from node and nodes The time interval between the most recent interaction. For nodes and nodes The shortest path length between them; Sort all nodes from highest to lowest comprehensive centrality to obtain a sorted node sequence. From the sorted node sequence, select the k nodes with the highest comprehensive centrality to construct the core node set C. The remaining nodes are denoted as non-core nodes. For each non-core node, calculate its path distance to all nodes in the core node set C, and take the minimum path distance as the shortest path distance from the corresponding non-core node to C. Assign all non-core nodes with a shortest path distance of 1 to layer 1; assign all non-core nodes with a shortest path distance of 2 to layer 2; and so on, until all non-core nodes are assigned to a layer. Classify the nodes belonging to the same layer to obtain several node groups. For each node group, mine the new associations between nodes within the node group.

[0010] Furthermore, the method for classifying nodes belonging to the same layer includes: For all nodes in the same layer, construct a subnetwork G1=(V1,E1), where V1 is the set of all nodes in the layer, and E1 is the set of edges between nodes in the layer. For each node in V1 Calculate its flow rate value ; ;in Represents nodes The set of connected neighboring nodes. for Nodes within, For nodes Traffic value, For nodes The degree, For nodes Betweenness centrality; and Balance the weights for the flow values; Based on the flow value of nodes, a hierarchical coding tree is constructed using a deterministic weighted circular walk strategy; each node corresponds to a coding node in the hierarchical coding tree, and the weight of the coding node is the flow value of the corresponding node; the hierarchical coding tree is optimized to obtain the final coding tree; In the final coding tree, starting from the root node, for each non-leaf node, calculate the coding length L1 for treating it as an independent cluster and the coding length L2 for treating its child nodes as independent clusters. If L1 is less than L2, then the non-leaf node is considered as an independent cluster; otherwise, its child nodes are considered as independent clusters. Traverse all non-leaf nodes to obtain the final cluster partitioning scheme, with clusters corresponding to node groups, and then obtain several node groups.

[0011] Furthermore, the method for optimizing the hierarchical coding tree includes: Starting from the root node of the hierarchical coding tree, perform a depth-first traversal. For each non-leaf node, check if its child nodes correspond to the same code. If so, mark these child nodes as adjacent identical clusters. For a sequence of nodes marked as adjacent identical clusters, calculate its run length, i.e., the number of consecutive adjacent identical clusters, and create a run-length encoding table containing two columns: run length and corresponding code. The run length increases from 2, and the corresponding code uses either variable-length encoding or fixed-length encoding. Traverse the coding tree, and for each sequence of nodes marked as adjacent identical clusters, replace the original code with the corresponding code from the run-length encoding table. This yields the final coding tree.

[0012] Furthermore, the method for integrating the processed member characteristics and relationship network includes: Create an empty dictionary of nodes, iterate through all nodes in the association network, assign a unique ID as the key and the node itself as the value, and add it to the empty dictionary of nodes. Create an empty dictionary of features, iterate through the processed member features, assign a unique ID as the key and the name of the processed member feature as the value, and add it to the empty dictionary of features. Create an empty list of edges, for each edge, obtain the IDs of the two endpoint nodes in the empty dictionary of nodes, and the association type between the two nodes; encode the IDs and association type of the two nodes into a tuple as the representation of the edge; iterate through the processed member features of the two nodes corresponding to the edge, encode their IDs and feature values ​​into a tuple, add it to the representation of the edge, and add the complete representation of the edge to the empty list of edges, thus obtaining the list of member combination graphs.

[0013] Furthermore, the method for constructing the member status update model includes: The member portfolio graph is represented as a grid, with each node corresponding to a cell in the grid. The topology of the grid is determined based on the relationships between members. The state of a cell is defined as a multi-dimensional vector. The state update rules for cells are defined, including policy purchase intention update rules, claim tendency update rules, and health status update rules. This yields the member state update model. The policy purchase intention update rule is as follows: based on a linear combination of the member's current purchase intention, demographic characteristics, historical insurance frequency, and the average purchase intention of neighboring members, the corresponding member's purchase intention is updated in the next time step; the claims tendency update rule is as follows: based on a linear combination of the member's current claims tendency, health status, risk characteristics, and the average claims tendency of neighboring members, the corresponding member's claims tendency is updated in the next time step. The health status update rule is as follows: based on a linear combination of the member's current health status, the member's age, and the average health status of neighboring members, the corresponding member's health status is updated in the next time step.

[0014] Furthermore, the method for predicting the overall behavioral trend of members over the next T time periods and quantifying the overall behavioral trend into the risk exposure level for the corresponding time periods includes: Based on the feature values ​​corresponding to members in the member combination diagram list, assign an initial state to each cell, and iteratively update the state of all cells according to the state update rules. Each iteration represents a time step. Repeat the iteration several times, and record the state of all members after each iteration. Then, divide the cells into segments according to the preset time intervals to obtain the state sequence for the next T time periods. Based on the state sequence of the next T time periods, construct an N-row, D-column member state matrix X, where N is the number of members and D is the dimension of the state; each row of the member state matrix corresponds to the state of a member. The member state matrix X is centered by subtracting the mean of each column, and the covariance matrix of X is calculated. Eigenvalue decomposition is performed on the covariance matrix to obtain D² decomposition values ​​and corresponding D² eigenvectors. The eigenvector corresponding to the largest decomposition value is selected as the projection direction, projecting the original state-related multidimensional vectors onto a 2D / 3D space spanned by the selected eigenvectors. For each time period, the states of all corresponding members are projected onto the spanned 2D / 3D space to obtain a point cloud distribution, which represents the overall behavioral trend. Each point in the point cloud distribution represents a member. A coordinate system is defined, and the point cloud distribution is placed within this system to obtain the coordinates of each point. A reference point is preset in the coordinate system for each time period. Based on the coordinates of each point, the average risk exposure score of all members is calculated. ;in, An index for members; For members The weights; For members The distance from the corresponding point to the reference point; Time period The maximum distance from all points within the area to the reference point; the average risk exposure score is used as the risk exposure level for the corresponding time period; Members are segmented according to their risk exposure levels, and risk exposure threshold ranges are set; members are divided into high-risk, medium-risk, and low-risk levels; differentiated marketing strategies are developed based on the high-risk, medium-risk, and low-risk levels.

[0015] The technical effects and advantages of the membership integrated management system based on big data analysis of this invention are as follows: This invention significantly enhances the sophistication of membership management and the targeting of marketing strategies. By constructing a membership relationship network, it fully considers the complex social relationships among members, making the analysis of factors influencing member behavior more comprehensive and accurate. This helps to gain a deeper understanding of member needs and improve the suitability and personalization of services. Secondly, by modeling the dynamic evolution of member behavior as a grid-based model, it can accurately predict member behavior trends over a future period, providing an important basis for developing forward-looking marketing strategies and improving the foresight and proactivity of marketing decisions. Furthermore, by quantifying the overall behavioral trends of members into risk exposure levels, members are segmented according to their risk exposure levels, laying the foundation for developing differentiated and refined marketing strategies. This allows marketing strategies to more accurately target member groups at different risk levels, improving marketing conversion rates and efficiency. Finally, it helps to build a closed loop for the entire membership lifecycle management, achieving systematization, intelligence, and automation of membership management, reducing labor costs, and improving management efficiency. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the membership integrated management system based on big data analysis according to the present invention; Figure 2 This is a schematic diagram of the membership integration management method based on big data analysis according to the present invention. Detailed Implementation

[0017] 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.

[0018] Example 1 Please see Figure 1 As shown in this embodiment, the membership integration management system based on big data analysis includes: a data acquisition module, a processing module, a comprehensive evaluation module, and a strategy fitting module; the modules are connected via wired and / or wireless means to realize data transmission between modules; The data acquisition module is used to collect member data of insurance members; The processing module is used to perform preliminary processing on member data to obtain a list of member combination charts; The model building module constructs a member status update model based on the member combination diagram list; The comprehensive assessment module, based on the member status update model, predicts the overall behavioral trends of members over the next T time periods and quantifies these trends into the risk exposure levels for the corresponding time periods. Based on the risk exposure levels, it formulates corresponding differentiated marketing strategies to achieve refined member management.

[0019] Member data includes personal information data, policy data, claims data, health data, geolocation data, and social relationship data. Personal information data includes a member's age, gender, marital status, education level, and occupation. Policy data includes information on the member's insurance policies, such as frequency of coverage, insured amount, and policy type. Claims data includes a member's claims history, such as claim frequency, claim amount, and reason for claim. Health data includes a member's health status data, such as body mass index, past medical history, history of genetic diseases, and health habits (exercise, diet, etc.). Geolocation data includes a member's address information, used to determine the geographical relationships between members. Social relationship data includes social relationship data between members, such as communication records, used to determine the social relationships between members.

[0020] It should be noted that all the above-mentioned member data were obtained through legal and compliant means and with the explicit authorization and permission of the data subjects. We strictly abide by relevant laws and regulations, respect and protect personal privacy rights, and ensure that the reasonable use of data complies with legal requirements and ethical standards.

[0021] Methods for initial processing of member data include: By integrating member data, a complete member profile is constructed. Specifically, the core entities (such as customers / members) and related entities are defined, such as personal information, policies, claims, and health. The relationships between the core entities and related entities are determined, such as one-to-one and one-to-many relationships. For each core entity, corresponding data records are extracted from the member data, and several data records of the same core entity in different data types within the member data are identified. Related entities are identified from the identified data records, and related entities are associated with the core entities to generate a fused view of the corresponding core entities; thus, the complete member profile is achieved.

[0022] Define the types of relationships that exist between members. These relationships include family relationships, social relationships, geographic relationships, policy relationships, and claims relationships. Based on these relationship types, extract the corresponding relationships between members from the member profile. Family relationships are extracted from personal information, geographic relationships from address information, and policy relationships from policy information. Family relationships include: spouses, parents and children, siblings, etc.; social relationships include: friends, colleagues, classmates, etc.; geographic relationships include: living in the same neighborhood, the same city, etc.; policy relationships include: co-insurers, beneficiaries, etc.; and claims relationships include: people involved in the same claims event, etc.

[0023] The extracted relationships are represented as an undirected graph G=(V, E), where V is the set of nodes, representing members; E is the set of edges, representing the relationships between members. New relationships are mined from the undirected graph G. Specifically, for each node in the undirected graph... Calculate its comprehensive centrality ; ;in, For nodes The set of directly connected neighboring nodes. For nodes Nodes within the set of directly connected neighboring nodes; For nodes The degree of (i.e., the degree of the node) (Number of directly connected edges) The total number of nodes; The maximum degree of all nodes; For nodes and nodes The strength of the correlation between them.

[0024] ;in, , , and These are the weight coefficients of the corresponding terms, and their sum is 1; For nodes and nodes Similarity scores between attribute vectors, such as cosine similarity, Jaccard similarity coefficient, etc. For nodes and nodes The number of interactions (such as communication) between them. The attenuation parameter controls the degree to which the time interval affects the correlation strength. Represents distance from node and nodes The time interval between the most recent interaction. For nodes and nodes The shortest path length between (in an undirected graph G); the attribute vector is a vector composed of the member's attributes, including age, gender, occupation, income level, marital status, insurance status, and claims history; the attributes are encoded, for example, for numerical attributes such as age and income, the original values ​​are used directly; for categorical attributes such as gender and occupation, one-hot encoding or other encoding methods are needed to convert them into numerical vectors; the attribute vector of a node is formed by concatenating the numerical values ​​or vectors corresponding to all attributes of the same node in a certain order.

[0025] Sort all nodes in descending order of their comprehensive centrality to obtain a sorted node sequence. From the sorted node sequence, select the k nodes with the highest comprehensive centrality to construct the core node set C. The remaining nodes are denoted as non-core nodes. For each non-core node, calculate its path distance to all nodes in the core node set C, and take the minimum path distance as the shortest path distance from the corresponding non-core node to C. Assign all non-core nodes whose shortest path distance is equal to 1 to layer 1; assign all non-core nodes whose shortest path distance is equal to 2 to layer 2; and so on, until all non-core nodes are assigned to a certain layer.

[0026] Nodes belonging to the same layer are classified to obtain several node groups; specifically, for all nodes in the same layer, a sub-network G1=(V1,E1) is constructed, where V1 is the set of all nodes in the layer and E1 is the set of edges between nodes in the layer.

[0027] For each node in V1 Calculate its flow rate value ; ;in Represents nodes The set of connected neighboring nodes. for Nodes within, For nodes Traffic value, For nodes The degree, For nodes Betweenness centrality (the number of times a node acts as a mediator or bridge for the shortest path between other pairs of nodes in a subnetwork). and To balance the weights of traffic values, the influence of different factors on traffic values ​​can be flexibly controlled; for example, if more attention is paid to network topology, the weights can be increased. If more attention is paid to the mediating role of nodes in the network, then the... It should be noted that the flow value of a node is equal to 1 (i.e., the initial probability of it being the source node to initiate a random walk), plus the flow contribution of all neighboring nodes transmitted to the corresponding node through the edge corresponding to the edge.

[0028] Based on the node's traffic value, a hierarchical coding tree (Huffman tree) is constructed using a deterministic weighted circular walk strategy; each node corresponds to a coding node in the hierarchical coding tree, and the weight of the coding node is the traffic value of the corresponding node; the hierarchical coding tree is then optimized to obtain the final coding tree.

[0029] Starting from the root node of the hierarchical coding tree, perform a depth-first traversal. For each non-leaf node, check if its child nodes correspond to the same code. If so, mark these child nodes as adjacent identical clusters. For a sequence of nodes marked as adjacent identical clusters, calculate its run length, i.e., the number of consecutive adjacent identical clusters, and create a run-length encoding table containing two columns: run length and corresponding code. The run length increases from 2, and the corresponding code uses variable-length encoding (such as Huffman coding) or fixed-length encoding. Traverse the coding tree, and for each sequence of nodes marked as adjacent identical clusters, replace the original code with the corresponding code from the run-length encoding table. This yields the final coding tree.

[0030] In the final coding tree, starting from the root node, for each non-leaf node, calculate the coding length L1 for treating it as an independent clique and the coding length L2 for treating its child nodes as independent cliques. If L1 is less than L2, then the non-leaf node is considered as an independent clique; otherwise, its child nodes are considered as independent cliques. Traverse all non-leaf nodes to obtain the final clique partitioning scheme (the scheme for dividing the subnetwork into multiple cliques). The cliques correspond to node groups, and then several node groups are obtained.

[0031] For each node group, new associations between nodes within the group are mined. Mining can be performed using association rule mining, anomaly detection, and other methods based on node attribute information, topological structure information, etc. For the mined new associations, they are added to the original undirected graph to obtain the complete association network.

[0032] Member characteristics are extracted from the original member data. These characteristics include demographic features, behavioral features, and risk features. Specifically, demographic features are extracted from members' personal information, including age, gender, marital status, education level, and occupation. Behavioral features are extracted from members' policy information, claims information, and health data, including frequency of insurance purchases, frequency of claims, health habits (such as exercise and diet), and consumption habits. Risk features are extracted from members' claims information and health data, including medical history, pre-existing conditions, and hereditary diseases.

[0033] The member features are standardized and discretized to obtain the processed member features. The processed member features and the association network are then integrated. Specifically, an empty node dictionary is created to store nodes and their corresponding IDs. All nodes (members) in the association network are traversed, and each node is assigned a unique ID as the key and the node itself as the value, which is then added to the empty node dictionary. An empty feature dictionary is also created to store the processed member features and their corresponding IDs. The processed member features are traversed, and each processed member feature is assigned a unique ID as the key and the name of the processed member feature as the value, which is then added to the empty feature dictionary.

[0034] Create an empty edge list to traverse each edge (the relationship between members) in the association network. For each edge, obtain the corresponding numbers of the two endpoint nodes in the node empty dictionary, as well as the relationship type between the two nodes (such as family relationship, geographical location relationship, etc.). Encode the numbers of the two nodes and the relationship type into a tuple as the edge representation. Traverse the processed member features of the two nodes corresponding to the edge, and encode their numbers and feature values ​​(the feature value encoding is obtained from the standardization and discretization processes) into a tuple, add it to the edge representation, and add the complete representation of the edge (including node number, relationship type, and feature value) to the empty edge list, thus obtaining the member combination graph list.

[0035] Each element in the member combination graph list is a tuple, representing an edge and its related information. The first element of the tuple is a binary tuple, representing the node numbers of the two endpoints of the edge. The second element of the tuple is the association type number. Subsequent elements are a series of binary tuples, where the first element is the feature number and the second element is the corresponding feature value, representing the feature information of the two nodes corresponding to the edge.

[0036] For example, suppose there are two nodes, 1 and 2, that have a family relationship. Node 1 has an age characteristic of 30, and node 2 has an age characteristic of 25. Then, the edge would be represented in the edge set array as follows: ((1,2),1,(0,30),(0,25)); where (1,2) represents the node number, 1 represents the family relationship type number, and (0,30) and (0,25) represent the age characteristics of node 1 and node 2 (assuming the age characteristic number is 0) and their corresponding values, respectively.

[0037] The methods for constructing a member status update model include: The member portfolio graph is represented as a grid, with each node (member) corresponding to a cell in the grid. The topology of the grid is determined by the relationships (edges) between members. The state of a cell is defined as a multi-dimensional vector, which includes the following dimensions: policy purchase intention (a real number between 0 and 1), claim tendency (a real number between 0 and 1), and health status (a real number between 0 and 1, where 0 represents the worst and 1 represents the best).

[0038] The defined state update rules for the grid include policy purchase intention update rules, claim tendency update rules, and health status update rules; the state update rules should be able to simulate the dynamic evolution of member behavior, such as policy purchase, claims, and changes in health status; thus, a member state update model is obtained.

[0039] The policy purchase intention update rule is as follows: based on a linear combination of the member's current purchase intention, demographic characteristics (such as age, income, etc.), historical insurance frequency, and the average purchase intention of neighboring members (with correlation) (the influence of neighbors), the corresponding member's purchase intention in the next time step is updated.

[0040] The claim tendency update rule is as follows: based on a linear combination of a member’s current claim tendency, health status, risk characteristics (such as pre-existing conditions, hereditary diseases, etc.) and the average claim tendency of neighboring members, the corresponding member’s claim tendency is updated in the next time step.

[0041] The health status update rule is as follows: based on a linear combination of the member's current health status, the member's age, and the average health status of neighboring members, the corresponding member's health status is updated in the next time step.

[0042] It should be noted that, based on members' historical insurance purchase data, the insurance purchase frequency within a certain period is calculated and mapped to a range of 0-1 as a quantitative value of the current purchase intention; for each member, a set of neighboring members with which they are related is found, and the average purchase intention of all members in the neighboring member set is calculated as the average purchase intention of that member's neighbors; based on members' historical claims data, the claims frequency within a certain period is calculated and mapped to a range of 0-1 as a quantitative value of the current claims tendency, or a machine learning model is used to predict the claims tendency score based on member characteristics.

[0043] Key indicators, such as body mass index, are extracted from member health data. These key indicators are then comprehensively considered and mapped to a score between 0 and 1, serving as a quantitative value for health status. Similar to the average purchase intention, the average claim tendency is calculated for each member's neighbor set. Similarly, the average health status is calculated for each member's neighbor set. The above quantification methods can be adjusted according to actual circumstances, and the linear combinations are all quantified using methods such as machine learning.

[0044] Based on the feature values ​​corresponding to members in the member combination diagram list, assign an initial state to each cell. According to the state update rules, iterate and update the state of all cells. Each iteration represents a time step (such as one month or one year). Repeat the iteration several times. After each iteration, record the state of all members. Then, divide the cells into segments according to the preset time intervals to obtain the state sequence for the next T time periods.

[0045] Based on the state sequence over T future time periods, construct an N-row, D-column member state matrix X, where N is the number of members and D is the dimension of the state. Each row of the member state matrix corresponds to the state of one member. Center the member state matrix X by subtracting the mean of each column and calculate the covariance matrix of the member state matrix X. Perform eigenvalue decomposition on the covariance matrix to obtain D2 decomposition values ​​and corresponding D2 eigenvectors. Select the eigenvector corresponding to the largest decomposition value as the projection direction and project the original state-corresponding multidimensional vectors onto the 2D / 3D space spanned by the selected eigenvectors.

[0046] For each time period, the status of all members is projected onto Zhang Cheng's 2D / 3D space to obtain a point cloud distribution, which represents the overall behavioral trend. Each point in the point cloud distribution represents a member. A coordinate system (two-dimensional / three-dimensional) is defined, and the point cloud distribution is placed in the defined coordinate system to obtain the coordinates of each point.

[0047] Pre-set reference points in the coordinate system for each time period. Based on the coordinates of each point, the average risk exposure score of all members is calculated. ;in, An index for members; For members Specifically, the weighting is calculated by determining the customer lifetime value or customer value score for each member and using it as the weight. Customer lifetime value is calculated based on a member's historical data (such as the amount of insurance premium, the amount of claims, etc.) and expected future returns. For members The distance from the corresponding point to the reference point; Time period The maximum distance from all points within the area to the reference point; the average risk exposure score is used as the risk exposure level for the corresponding time period.

[0048] Members will be segmented based on their risk exposure levels, with threshold ranges set for each level. Members will be divided into high-risk, medium-risk, and low-risk tiers. Differentiated marketing strategies will be developed based on these tiers. Specifically, high-risk tiers correspond to higher risk exposure levels and require corresponding marketing strategies to mitigate risk; low-risk tiers correspond to lower risk exposure levels and require marketing strategies to promote growth. For high-risk tiers, service quality will be improved, customer relationship maintenance strengthened, product mix optimized, risk protection products provided, health management activities conducted, and member health status improved. Precision marketing will be implemented to increase customer loyalty.

[0049] For medium-risk customers, we will strengthen customer education, raise their risk awareness, recommend suitable insurance product combinations, carry out healthy lifestyle promotion activities, implement targeted marketing activities, and stabilize customer relationships. For low-risk customers, we will optimize product structure, provide value-added services, carry out customer retention activities, enhance customer loyalty, implement developmental marketing, explore potential customer needs, promote healthy living concepts, and cultivate healthy habits. Track changes in member status, evaluate the effectiveness of marketing strategies, dynamically adjust and optimize marketing strategies based on feedback, establish a marketing strategy optimization mechanism, and continuously improve marketing methods.

[0050] This embodiment significantly enhances the sophistication of membership management and the targeting of marketing strategies. By constructing a membership relationship network, it fully considers the complex social relationships among members, making the analysis of factors influencing member behavior more comprehensive and accurate. This helps to gain a deeper understanding of member needs and improve the suitability and personalization of services. Secondly, by modeling the dynamic evolution of member behavior as a grid-based model, it can accurately predict member behavior trends over a future period, providing an important basis for developing forward-looking marketing strategies and improving the foresight and proactivity of marketing decisions. Furthermore, by quantifying the overall behavioral trends of members into risk exposure levels, members are segmented according to their risk exposure levels, laying the foundation for developing differentiated and refined marketing strategies. This allows marketing strategies to more accurately target member groups at different risk levels, improving marketing conversion rates and efficiency. Finally, it helps to build a closed loop for the entire membership lifecycle management, achieving systematization, intelligence, and automation of membership management, reducing labor costs, and improving management efficiency.

[0051] Example 2 Please see Figure 2 As shown, the parts not described in detail in this embodiment are described in Embodiment 1. A membership integration management method based on big data analysis is provided, including: Step 1: Collect insurance member data; Step 2: Perform preliminary processing on the member data to obtain a list of member combination charts; Step 3: Based on the member combination diagram list, construct a member status update model; Step 4: Based on the member status update model, predict the overall behavioral trend of members in the next T time periods, and quantify the overall behavioral trend into the risk exposure level of the corresponding time period; formulate corresponding differentiated marketing strategies based on the risk exposure level.

[0052] Example 3 This embodiment discloses an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the operation mode of the membership integration management method based on big data analysis described above.

[0053] Since the electronic device described in this embodiment is the electronic device used to implement the membership integration management method based on big data analysis in the embodiments of this application, those skilled in the art can understand the specific implementation method and various variations of the electronic device in this embodiment based on the membership integration management method based on big data analysis described in the embodiments of this application. Therefore, how the electronic device implements the method in the embodiments of this application will not be described in detail here. Any electronic device used by those skilled in the art to implement the membership integration management method based on big data analysis in the embodiments of this application falls within the scope of protection of this application.

[0054] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0055] The above description is merely a preferred embodiment of the present invention, and the scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for users of ordinary technical skills, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A member integrated management system based on big data analysis, characterized in that: include: Data acquisition module, processing module, comprehensive evaluation module and strategy fitting module; each module is connected by wired and / or wireless means; Data collection module, used to collect member data of insurance members; The processing module is used to preliminarily process the member data and obtain a member combination chart list; Model building module, based on the member combination chart list, builds a member status update model; The comprehensive assessment module predicts the overall behavior trend of members in the next T time periods based on the member status update model, and quantifies the overall behavior trend into the risk exposure degree of the corresponding time period; Develop corresponding differentiated marketing strategies based on the degree of risk exposure.

2. The member integrated management system based on big data analysis according to claim 1 is characterized in that: The member data includes personal information data, insurance policy data, claims data, health data, geographic location data and social relationship data.

3. The member integrated management system based on big data analysis according to claim 2 is characterized in that: The methods of preliminary processing of member data include: Integrate member data to build a complete member portrait and define the types of relationships between members, including family relationships, social relationships, geographic location relationships, policy relationships, and claims relationships; based on the relationship types, extract the relationships between corresponding members from the member portrait; The extracted association relationships are represented as an undirected graph G=(V, E), where V is a set of nodes, which represent members; E is a set of edges, which represent the association relationships between members; new association relationships are mined from the undirected graph G, and the mined new association relationships are added to the original undirected graph to obtain a complete association network; member characteristics are extracted from the original member data, and the member characteristics include demographic characteristics, behavioral characteristics, and risk characteristics; Standardize and discretize member characteristics; obtain processed member characteristics; integrate the processed member characteristics and the association relationship network to obtain a member combination map list.

4. The member integrated management system based on big data analysis according to claim 3 is characterized in that: The method of integrating member data includes: Define the core entities and related entities of the member portrait, and determine the association relationship between the core entities and related entities; for each core entity, extract the corresponding data records from the member data, and identify several data records of the same core entity in different data types in the member data; identify related entities from the identified data records, and associate the related entities with the core entities to generate a fusion view of the corresponding core entity; that is, a complete member portrait.

5. The member integrated management system based on big data analysis according to claim 4 is characterized in that: The method of mining new association relationships from the undirected graph G includes: For each node in an undirected graph , calculate its comprehensive centrality ; ;in, For the node The set of directly connected neighbor nodes, For the node A node in the set of directly connected neighbor nodes; For Node The degree of is the total number of nodes; is the maximum degree of all nodes; For Node and nodes The strength of the association between ;in, , , and is the weight coefficient of the corresponding item, and the sum is 1; For Node and nodes The similarity score between the attribute vectors of ; For Node and nodes The number of interactions between is the attenuation parameter, Indicates the distance node and nodes The time interval between the last interaction behavior, For Node and nodes The shortest path length between All nodes are sorted from large to small according to their comprehensive centrality to obtain a sorted node sequence; from the sorted node sequence, the first k nodes with the highest comprehensive centrality are selected to construct a core node set C; the remaining nodes are recorded as non-core nodes; for each non-core node, the path distance between it and all nodes in the core node set C is calculated, and the minimum value of the path distance is taken as the shortest path distance from the corresponding non-core node to C, and all non-core nodes corresponding to the shortest path distance equal to 1 are divided into the first layer; all non-core nodes corresponding to the shortest path distance equal to 2 are divided into the second layer; and so on, until all non-core nodes are assigned to a certain layer; nodes belonging to the same layer are classified to obtain several node groups; for each node group, new association relationships between nodes in the node group are mined.

6. The member integrated management system based on big data analysis according to claim 5 is characterized in that: The method of classifying nodes belonging to the same layer includes: For all nodes in the same layer, construct a subnetwork G1=(V1, E1), where V1 is the set of all nodes in the layer, and E1 is the set of edges between nodes in the layer; For each node in V1 , calculate its flow value ; ;in Representation and Node The set of connected neighbor nodes, for The nodes inside For Node The flow value, For Node The degree of For Node Betweenness centrality of ; and Balance weights for flow values; Based on the traffic value of the node, a deterministic weighted circular walk strategy is used to construct a hierarchical coding tree; each node corresponds to a coding node in the hierarchical coding tree, and the weight of the coding node is the traffic value of the corresponding node; the hierarchical coding tree is optimized to obtain the final coding tree; On the final coding tree, starting from the root node, for each non-leaf node, calculate the coding length L1 of treating it as an independent cluster, and the coding length L2 of treating its child nodes as independent clusters; if L1 is less than L2, then the non-leaf node is regarded as an independent cluster; otherwise, its child nodes are regarded as independent clusters; traverse all non-leaf nodes to obtain the final cluster division scheme, the clusters correspond to node groups, and then several node groups are obtained.

7. The member integrated management system based on big data analysis according to claim 6 is characterized in that: The method of optimizing the hierarchical coding tree includes: Starting from the root node of the hierarchical coding tree, a depth-first traversal is performed. For each non-leaf node, its child nodes are checked to see if they correspond to the same code. If so, these child nodes are marked as adjacent identical clusters. For the sequence of nodes marked as adjacent identical clusters, the run length, that is, the number of consecutive adjacent identical clusters, is calculated, and a run coding table is created, which contains two columns, namely the run length and the corresponding code. The run length increases from 2, and the corresponding code uses variable-length coding or fixed-length coding. The coding tree is traversed, and for each sequence of nodes marked as adjacent identical clusters, the original code is replaced with the corresponding code in the run coding table. The final coding tree is obtained.

8. The member integrated management system based on big data analysis according to claim 7 is characterized in that: The method of integrating the processed member characteristics and the associated relationship network includes: Create an empty node dictionary, traverse all nodes in the association network, assign a unique number to each node as the key, and the node itself as the value, and add it to the empty node dictionary; create an empty feature dictionary, traverse the processed member features, assign a unique number to each processed member feature as the key, and the name of the processed member feature as the value, and add it to the empty feature dictionary; create an empty edge list, for each edge, get the corresponding numbers of the two endpoint nodes in the empty node dictionary, and the type of association relationship between the two nodes; encode the numbers and the type of association relationship of the two nodes into a tuple as the representation of the edge; traverse the processed member features of the two nodes corresponding to the edge, encode their numbers and feature values ​​as a tuple, add it to the representation of the edge, and add the complete representation of the edge to the empty edge list, so as to obtain the member combination graph list.

9. The member integrated management system based on big data analysis according to claim 8, characterized in that: The method of constructing the member status update model includes: The member combination graph list is represented as a grid, each node corresponds to a grid in the grid, and the topological structure of the grid is determined according to the association relationship between members; the state of the grid is defined as a multidimensional vector; the state update rules of the grid are defined, including the policy purchase intention update rule, the claim tendency update rule and the health status update rule; that is, the member state update model is obtained; The update rule for the policy purchase intention is: based on the linear combination of the member's current purchase intention, demographic characteristics, historical insurance frequency, and the average purchase intention of neighboring members, the purchase intention of the corresponding member at the next time step is updated; the update rule for the claim tendency is: based on the linear combination of the member's current claim tendency, health status, risk characteristics, and the average claim tendency of neighboring members, the claim tendency of the corresponding member at the next time step is updated; The health status update rule is: based on the linear combination of the member's current health status, the member's age, and the average health status of neighbor members, update the health status of the corresponding member in the next time step.

10. The member integrated management system based on big data analysis according to claim 9, characterized in that: The method of predicting the overall behavior trend of members in the next T time periods and quantifying the overall behavior trend into the risk exposure degree of the corresponding time period includes: According to the characteristic values ​​corresponding to the members in the member combination map list, an initial state is assigned to each grid, and the states of all grids are iteratively updated according to the state update rule. Each iteration represents a time step. The iteration is repeated several times. After each iteration, the states of all members are recorded. The states are segmented according to the preset time intervals. The state sequence of the next T time periods is obtained. Based on the state sequence of the next T time periods, construct a member state matrix X with N rows and D columns, where N is the number of members and D is the dimension of the state. Each row of the member state matrix corresponds to the state of a member. Center the member status matrix X, subtract the mean of each column from the column, calculate the covariance matrix of the member status matrix X, perform eigenvalue decomposition on the covariance matrix, obtain D2 decomposition values ​​and corresponding D2 eigenvectors, select the eigenvector corresponding to the largest decomposition value as the projection direction, and project the multidimensional vector corresponding to the original state into the 2D / 3D space spanned by the selected eigenvectors; for each time period, project the states of all corresponding members into the spanned 2D / 3D space to obtain the point cloud distribution, which is the corresponding overall behavior trend; each point in the point cloud distribution represents a member; define a coordinate system, and place the point cloud distribution in the defined coordinate system to obtain the coordinate point corresponding to each point; preset a reference point in the coordinate system, for each time period , based on the coordinate points corresponding to each point, calculate the average risk exposure score of all members ;in, An index of members; For Members The weight of For Members The distance from the corresponding point to the reference point; For time period The maximum distance from all points in the range to the reference point; the average risk exposure score is used as the risk exposure level for the corresponding time period; Members are stratified according to their risk exposure levels and a threshold range for risk exposure levels is set; members are divided into high-risk, medium-risk and low-risk levels; differentiated marketing strategies are formulated based on the divided high-risk, medium-risk and low-risk levels.

Images