A method and system for processing value-added rights data based on insurance policy data

Abstract

This invention relates to the field of insurance data processing technology, specifically disclosing a method and system for processing value-added benefits data based on policy data. The method involves collecting data from multiple insurance policies under a customer's name to generate an initial attribute set; constructing each rule in a composite rule base as a rule hyperedge to form a rule hypergraph; performing spectral clustering segmentation on the rule hyperedges in the rule hypergraph to obtain multiple rule clusters and corresponding template bit strings; converting customer attributes into customer bit strings, calculating the Hamming distance with the template bit strings, and filtering candidate rule clusters; constructing a policy association subgraph under the customer's name, performing directed graph traversal verification on the rules in the candidate rule clusters, and outputting value-added benefits that meet the conditions. This invention can reduce the computational overhead of composite rule matching and improve the accuracy of identifying cross-policy combination benefits.

Description

Technical Field

[0001] This invention relates to the field of insurance data processing technology, specifically to a method and system for processing value-added rights data based on policy data. Background Technology

[0002] In the insurance business, value-added benefits are linked to the customer's policy data. In traditional value-added benefit data processing methods, insurance companies pre-set multiple composite rules, each composed of multiple attribute dimensions (such as insurance type, continuous insurance years, and claims records) combined through logical AND-OR relationships. When it is necessary to match benefits for a customer, the system obtains the attribute values ​​of the customer's policies one by one, and sequentially traverses each composite rule to determine whether the rule conditions are met.

[0003] When dealing with complex relationships involving large-scale composite rules and overlapping policyholder and insured identities across multiple policies, how can we avoid the computational waste of traversing all rules one by one while accurately identifying conditions requiring cross-policy combination verification? Existing technologies employ linear scanning of the rule base and multiple database queries, which struggles to balance matching efficiency with accurate capture of relationships. This invention pre-reduces a large number of rules into a small number of candidate clusters through spectral clustering of the rule hypergraph, and then performs fine-grained verification using directional depth-first traversal of the policy association subgraph. This reduces the computational load of matching while ensuring accurate identification of cross-policy combination conditions. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for processing value-added benefits data based on insurance policy data, so as to solve the problems mentioned above.

[0005] The objective of this invention can be achieved through the following technical solutions: A method for processing value-added benefits data based on policy data includes the following steps: S1. Collect data on multiple insurance policies under the customer's name, including: insurance type, continuous insurance period and claims record, and generate the customer's original attribute set based on the data of multiple insurance policies; S2, based on the preset composite rule library, take the multiple attribute dimensions associated with each composite rule as nodes, and construct a rule hyperedge by connecting all the attribute dimensions involved in each rule to form a rule hypergraph; S3, perform spectral clustering segmentation on all rule hyperedges in the rule hypergraph to obtain multiple rule clusters, and generate a low-dimensional attribute combination template and the corresponding template bit string for each rule cluster; S4. Convert the original attribute set into a client bit string according to the attribute dimension encoding order, calculate the Hamming distance between the client bit string and each template bit string, and select the rule clusters with Hamming distance less than the threshold as candidate rule clusters. S5: Construct a subgraph of multiple insurance policies under the customer's name based on the original attribute set. For each composite rule in the candidate rule cluster, perform a directed graph traversal in the subgraph to verify whether the policy attributes meet the rule conditions. Output the value-added benefits corresponding to the rules that meet the conditions as the processing result.

[0006] As a further aspect of the present invention: S2 specifically includes: Extract all attribute dimensions contained in each composite rule in the preset composite rule library, and assign a unique position identifier to each attribute dimension; For each composite rule, generate a binary bit string with a length equal to the total number of attribute dimensions. Set 1 to the corresponding position of the attribute dimension involved in the rule and set 0 to the other positions. Use the binary bit string as the rule superedge corresponding to the rule. Aggregate all rule hyperedges according to their position identifiers to form a rule hypergraph with attribute dimensions as nodes and binary bit strings as hyperedge connections.

[0007] As a further aspect of the present invention: the formation process of the regular hypergraph is as follows: For each attribute dimension, a corresponding attribute node is created based on its location identifier. All attribute nodes form a node set. Traverse all rule hyperedges. For each rule hyperedge's binary bit string, scan the position markers that are set to 1, retrieve the corresponding attribute nodes from the node set, and form the associated node group of the rule hyperedge. Using each rule hyperedge as a connection, all attribute nodes in the associated node group are aggregated into one through the hyperedge, generating a many-to-many mapping structure between attribute nodes and hyperedges, thus obtaining a rule hypergraph.

[0008] As a further aspect of the present invention: S3 specifically includes: Calculate the number of attribute dimensions shared between each pair of rule hyperedges in the rule hypergraph, and use the number of shared dimensions as the similarity to construct a hyperedge similarity matrix; Transform the hyperedge similarity matrix into a Laplacian matrix, and take the eigenvector corresponding to the second smallest eigenvalue of the Laplacian matrix as the segmentation basis; All rule hyperedges are segmented according to the ascending order of the values ​​of each component in the feature vector to obtain multiple rule clusters; Extract the common attribute dimensions of all rule superedges within each rule cluster, combine the common attribute dimensions into a low-dimensional attribute combination template for the rule cluster, and generate the corresponding template bit string based on the value of the attribute dimension in the template.

[0009] As a further aspect of the present invention: the construction of the hyperedge similarity matrix specifically includes: Extract the binary string of each rule superedge. The length of the binary string is equal to the total number of attribute dimensions. For any two rule superedges, perform a bitwise logical AND operation on their binary bit strings to obtain the overlapping bit string; Count the number of positions with a value of one in the overlapping bit string, use the number of positions as the number of shared attribute dimensions between the two rule super edges, and use the number of shared dimensions as the similarity value. Fill the matrix with the similarity values ​​calculated for all paired rule hyperedges, according to rows and columns, to obtain the hyperedge similarity matrix.

[0010] As a further aspect of the present invention: S4 specifically includes: Get the actual values ​​of each attribute dimension in the original attribute set, and judge whether each attribute dimension satisfies the value conditions in the corresponding rule cluster template according to the attribute dimension encoding order. If it satisfies, assign 1 to the corresponding bit; otherwise, assign 0 to generate the client bit string. After aligning the customer bit string with each template bit string, compare them bit by bit. Whenever the corresponding bit values ​​are different, accumulate the difference count to obtain the Hamming distance corresponding to the template bit string. Compare all Hamming distances with a preset threshold, extract template bit strings whose Hamming distance is less than the threshold, and mark the rule cluster to which the template bit string belongs as a candidate rule cluster.

[0011] As a further aspect of the present invention: the calculation process of the Hamming distance is as follows: The client bit string is divided into multiple consecutive bit segments of fixed length, and each bit segment corresponds to a set of sequentially arranged position identifiers. Perform the same partitioning on each template bit string to obtain the corresponding template bit segment. Align the client bit segment with the template bit segment and then perform a bitwise XOR operation to obtain the difference bit segment. A lookup table is pre-established to map each possible difference segment to the number of bits with a value of one. The difference count in each difference segment is obtained by looking up the table. The difference counts of each difference segment are accumulated to obtain the Hamming distance.

[0012] As a further aspect of the present invention: S5 specifically includes: Extract the unique identifier of each policy and the relationship between policies from the original attribute set, and construct an initial subgraph with policies as nodes and relationships as edges; Obtain the attribute dimension sequence and connection order contained in each composite rule in the candidate rule cluster. Perform a depth-first path search along the connection order in the initial subgraph. When each attribute dimension node is reached, verify whether the actual attribute value of the policy matches the rule conditions. If all attribute dimension nodes in the path search are successfully matched and the connection order is complete, the composite rule is deemed to meet the conditions; otherwise, the current search is stopped and the verification of the next rule is switched.

[0013] As a further aspect of the present invention: the construction of the initial subgraph specifically includes: Extract the unique identifier of each policy, as well as the policyholder's ID number and the insured's ID number, from the original attribute set; Iterate through all policies. If two policies have the same policyholder ID number, create a policyholder-related edge between the two policies. If the insured's ID number of one policy is the same as the policyholder's ID number of another policy, create a mutual insurance-related edge between the two policies. Treat all insurance policies as nodes, merge and deduplicate all edges related to the same policyholder and mutual insurance, and form an initial undirected subgraph between policy nodes.

[0014] A value-added benefits data processing system based on policy data, comprising: The policy data collection module is used to collect data on multiple insurance policies under a customer's name, including: insurance type, continuous insurance period and claims record, and generate the customer's original attribute set based on the data of multiple insurance policies; The rule hypergraph construction module, based on a preset composite rule library, uses the multiple attribute dimensions associated with each composite rule as nodes, and constructs a rule hyperedge by connecting all the attribute dimensions involved in each rule, thus forming a rule hypergraph; The rule cluster segmentation and template generation module performs spectral clustering segmentation on all rule hyperedges in the rule hypergraph to obtain multiple rule clusters, and generates a low-dimensional attribute combination template and the corresponding template bit string for each rule cluster; The customer bit string matching and filtering module converts the original attribute set into customer bit strings according to the attribute dimension encoding order, calculates the Hamming distance between the customer bit string and each template bit string, and filters out the rule clusters with Hamming distance less than the threshold as candidate rule clusters. The subgraph traversal and rights verification module constructs a subgraph of multiple insurance policies under a customer's name based on the original attribute set. For each composite rule in the candidate rule cluster, it performs a directed graph traversal in the subgraph to verify whether the policy attributes meet the rule conditions. The value-added rights corresponding to the rules that meet the conditions are output as the processing results.

[0015] The beneficial effects of this invention are: (1) By constructing the rules in the composite rule base into a rule hypergraph and performing spectral clustering segmentation, the original large number of composite rules that need to be matched one by one are reduced to a small number of candidate rule clusters, so that the online matching stage only needs to perform fine verification on the filtered rules, thereby reducing the computing resources and processing time required for each rights matching.

[0016] (2) By constructing a subgraph of the relationship between policyholders and insured persons among multiple policies under a customer’s name, and performing a depth-first directional traversal according to the attribute dimension sequence of the composite rule, the combination conditions across policies can be accurately identified, avoiding omissions or misjudgments of rights and interests due to ignoring the relationship between policies. Attached Figure Description

[0017] The invention will now be further described with reference to the accompanying drawings.

[0018] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a system block diagram of the present invention. Detailed Implementation

[0019] 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.

[0020] Please see Figure 1 As shown, this invention is a method for processing value-added benefits data based on policy data, including the following steps: S1. Collect data on multiple insurance policies under the customer's name, including: insurance type, continuous insurance period and claims record, and generate the customer's original attribute set based on the data of multiple insurance policies; S2, based on the preset composite rule library, take the multiple attribute dimensions associated with each composite rule as nodes, and construct a rule hyperedge by connecting all the attribute dimensions involved in each rule to form a rule hypergraph; S3, perform spectral clustering segmentation on all rule hyperedges in the rule hypergraph to obtain multiple rule clusters, and generate a low-dimensional attribute combination template and the corresponding template bit string for each rule cluster; S4. Convert the original attribute set into a client bit string according to the attribute dimension encoding order, calculate the Hamming distance between the client bit string and each template bit string, and select the rule clusters with Hamming distance less than the threshold as candidate rule clusters. S5: Construct a subgraph of multiple insurance policies under the customer's name based on the original attribute set. For each composite rule in the candidate rule cluster, perform a directed graph traversal in the subgraph to verify whether the policy attributes meet the rule conditions. Output the value-added benefits corresponding to the rules that meet the conditions as the processing result.

[0021] In S1, data on multiple insurance policies under a customer's name is collected, including: insurance type, continuous insurance period, and claims history. Based on this data, a set of the customer's original attributes is generated, specifically including: The insurance company's core underwriting system database is used daily to extract data from all valid multi-type insurance policies under a customer's name using Structured Query Language (SQuery Language). Specific extracted fields include: Insurance Type (identifying whether the policy belongs to critical illness, medical, auto, or accident insurance); Continuous Coverage Period (recording the number of years the policy has been continuously renewed from its initial effective date to the current year); and Claims Record (reflecting whether and how many claims have been made in the most recent three full policy years). This extraction process relies on the insurance company's existing data extraction tools, deployed on a data warehouse server and connected to a read-only copy of the core underwriting system's database to ensure no impact on online transaction performance. After extraction, the above fields from all policies under the same customer's name are merged and summarized using the customer's unique identifier as the primary key. Based on the merged data, each insurance type, the corresponding continuous insurance period, and the corresponding claims status (claims paid or no claims paid) are converted into attribute entries in a unified format according to a preset attribute coding table. All attribute entries together form the customer's original attribute set. This original attribute set is stored in the temporary cache of the in-memory database in the form of key-value pairs for subsequent processing.

[0022] In S2, based on a pre-defined composite rule base, multiple attribute dimensions associated with each composite rule are used as nodes. A rule hyperedge is constructed by connecting all the attribute dimensions involved in each rule, forming a rule hypergraph, which specifically includes: A composite rule base is pre-established in the insurance company's rights and interests rule server, and this rule base is stored in the form of database tables. Each composite rule contains a logical combination of at least two attribute dimensions, such as performing a logical AND operation between "car insurance continuous insurance years equal to 3 years" and "health insurance claim record equal to no claims". All unique attribute dimensions are extracted from this composite rule base. The attribute dimensions include the specific values ​​of the insurance type, the specific numerical range of the continuous insurance years, and the status value of the claim record, etc.

[0023] Each extracted attribute dimension is assigned a unique position identifier, which is an integer sequence number starting from 1 and incrementing. In the specific assignment, the attribute dimensions are numbered sequentially according to their business category (pre-insurance type, renewal period, subsequent claims record) to ensure that each attribute dimension corresponds to a fixed sequence number. For example, "car insurance continuous insurance period equal to 3 years" is assigned sequence number 5, and "health insurance claims record equal to no claims" is assigned sequence number 12.

[0024] For each composite rule in the composite rule base, a binary bit string of length equal to the total number of attribute dimensions is generated. Each bit in this binary bit string corresponds to the position identifier of an attribute dimension. For all attribute dimensions involved in the rule, the corresponding position in the binary bit string for that dimension is assigned a value of 1; for attribute dimensions not involved in the rule, the corresponding position is assigned a value of 0. Thus, each composite rule is converted into a binary bit string consisting of 0s and 1s, and this binary bit string is used as the rule's superedge.

[0025] To aggregate all rule hyperedges into a rule hypergraph, we first create an attribute node for the location identifier of each attribute dimension. Specifically, we iterate through the complete set of location identifiers, and for each integer index, create a node object named after that index. All these node objects together form a node set. Each node in this set represents a specific attribute dimension; for example, node number 5 represents "car insurance continuous insurance coverage period equal to 3 years".

[0026] Traverse all rule-based hyperedges. For each rule-based hyperedge's binary string, scan each bit sequentially from beginning to end. When a bit is found to be 1, record its corresponding position identifier, and then retrieve the node with the same position identifier from the node set. Combine all the nodes obtained from scanning the rule-based hyperedge to form the associated node group for that rule-based hyperedge. For example, an associated node group for a rule-based hyperedge might contain the node with index 5 and the node with index 12.

[0027] Using each rule hyperedge as a connecting link, all attribute nodes in the associated node group of that rule hyperedge are aggregated into a whole through that hyperedge. Specifically, a bidirectional mapping table is established in computer memory. The left side of this table records the number of each rule hyperedge, and the right side records the set of node indices in the associated node group of that hyperedge; it also records which rule hyperedges each node is associated with. This generates a many-to-many mapping structure between attribute nodes and hyperedges, which is the required rule hypergraph. This rule hypergraph is subsequently used for rule cluster segmentation.

[0028] In S3, spectral clustering is performed on all regular hyperedges in the regular hypergraph to obtain multiple regular clusters. A low-dimensional attribute combination template and a corresponding template bit string are generated for each regular cluster, specifically including: Obtain the binary representation of all rule hyperedges in the rule hypergraph. The length of the binary representation of each rule hyperedge is equal to the total number of attribute dimensions, denoted as . For any two rule-defined superedges, let the first one be denoted as . The binary string of the superedge of the rule is , No. The binary string of the superedge of the rule is .Will and Perform a bitwise AND operation to obtain an overlapping bit string. Then, the overlapping bit strings were counted. The number of binary bits that are 1 in the middle is denoted as . .Should This refers to the number of attribute dimensions shared between the two rule hyperedges. As the first Rule 1 and the first edge Similarity values ​​between rule super edges The above calculation process can be expressed using the following mathematical formula: ; in, Indicates the first The first rule in the binary string of the super-edge The value of the bit (0 or 1). Similarly, The AND operator represents the logical AND operation, and the summation symbol represents the summation over all... The bits are accumulated, and the final value is the number of shared attribute dimensions.

[0029] Repeat the above calculation for all pairs of regular hyperedges to obtain the similarity value between each pair of regular hyperedges. Assume the total number of regular hyperedges is... Then construct a OK Let the square matrix of columns be denoted as the similarity matrix. .matrix The first in Line 1 The column element is Because similarity has symmetry, that is... equal And diagonal elements This represents the number of shared attribute dimensions between the rule's hyperedge and itself, and its value is equal to the number of 1s in the binary string of that rule's hyperedge. In actual calculations, it follows... From 1 to , From 1 to The order, each Fill in the corresponding positions in the matrix to complete the construction of the hyperedge similarity matrix.

[0030] Obtain the similarity matrix Then, it is transformed into a Laplace matrix. First, the degree matrix is ​​calculated. The matrix is ​​related to For diagonal square matrices of the same order, the first element on the diagonal is... element Equal to the similarity matrix The The sum of all elements in the row, i.e. All off-diagonal elements are 0. Then, the Laplace matrix is ​​calculated using the following formula. : ; in, Too A square matrix of order 1, the first order of 2 Line 1 Column elements exist equal Time equals minus ,exist Not equal to Time equals negative .Should The matrix is ​​the Laplace matrix of the regular hyperedge.

[0031] Laplace matrix Eigenvalue decomposition is performed to find all eigenvalues ​​and their corresponding eigenvectors. In practice, numerical computation methods such as the QR algorithm or the power iteration method can be used to solve for matrix eigenvalues ​​on a computer. The second smallest eigenvalue is found from all eigenvalues ​​(i.e., the second eigenvalue after sorting from smallest to largest), and this eigenvalue is denoted as eigenvalue . .extract The corresponding feature vector is a feature vector containing A column vector of features, each component corresponding to a segmentation coefficient of a regular hyperedge. The features in this vector... The component values ​​are arranged in ascending order from smallest to largest, and the original number of the rule superedge corresponding to each component is recorded.

[0032] Based on the ascending order of the feature vector components, all regular hyperedges are segmented. The segmentation point is selected based on the abrupt change in difference between adjacent components. Specifically, the difference between any two adjacent components in the ascending sequence is calculated, and the position with the largest difference is selected as the segmentation point, dividing the sequence into left and right parts, each corresponding to a regular cluster. When multiple regular clusters are needed, the above segmentation process can be repeated for each sub-cluster until the number of regular hyperedges within the cluster is less than a preset threshold (e.g., 50) or the interval between feature values ​​is less than 0.1. Finally, multiple regular clusters are obtained, each containing a set of regular hyperedges with high similarity.

[0033] For each rule cluster, extract the attribute dimensions common to all rule hyperedges within the cluster. Specifically, take the binary bit string of the first rule hyperedge in the cluster as the initial common bit string, and then perform a bitwise logical AND operation with the binary bit strings of the remaining rule hyperedges in the cluster sequentially, updating the common bit string after each operation. After iterating through all hyperedges, the positions where the value is 1 in the final binary bit string represent the attribute dimensions commonly associated with all rule hyperedges in the cluster. Combine these common attribute dimensions according to their position identifiers in ascending order to form a low-dimensional attribute combination template. Finally, based on the value condition code corresponding to each attribute dimension in the template (e.g., a specific code for insurance type, a range code for the number of years), generate a template bit string corresponding to the template. The length of this template bit string is the same as the total number of original attribute dimensions, but it only sets 1 at the positions of the dimensions included in the template, and sets 0 at the remaining positions, for rapid comparison in subsequent steps.

[0034] In S4, the original attribute set is converted into client bit strings according to the attribute dimension encoding order. The Hamming distance between the client bit string and each template bit string is calculated, and rule clusters with Hamming distances less than a threshold are selected as candidate rule clusters. Specifically, these include: Retrieve the original attribute set generated in S1. This set stores the attribute values ​​corresponding to each type of insurance under the customer's name in key-value pairs, including the specific category of the insurance type, the value of the continuous insurance period, and the status of the claims record. Simultaneously, retrieve the template bit string corresponding to each rule cluster generated in S3, as well as the attribute dimension value conditions corresponding to each template bit string. Following the pre-determined attribute dimension encoding order, which is completely consistent with the sequence number of the assigned position identifier in S2, traverse each attribute dimension sequentially.

[0035] For the currently traversed attribute dimension, retrieve the actual value corresponding to that dimension from the original attribute set. Compare this actual value with the value condition of that attribute dimension in the rule cluster template corresponding to the template bit string. For example, if the attribute dimension is "continuous insurance coverage period equals 3 years", then determine whether the customer's actual continuous insurance coverage period is equal to 3; if they are equal, the dimension meets the condition; otherwise, it does not. When the condition is met, assign the number 1 to the corresponding bit in the customer bit string; when the condition is not met, assign the number 0. After performing the above judgment and assignment operations on all attribute dimensions in sequence, a binary bit string with a length equal to the total number of attribute dimensions is obtained. This binary bit string is used as the customer bit string for that customer.

[0036] Align the client bit string with the template bit string of the first rule cluster, i.e., align the starting bits of the two bit strings, ensuring a one-to-one correspondence. Then, starting from the first bit, compare the values ​​at the same position in the client bit string and the template bit string bit by bit. Whenever the two values ​​at a certain bit are different, increment the difference counter by 1. After comparing all bits, the final value of the difference counter is the Hamming distance corresponding to that template bit string. Repeat the above process to calculate the Hamming distance between the client bit string and the template bit string of each rule cluster.

[0037] To accelerate the calculation of Hamming distance, a segmented lookup table method is used. First, the client bit string is divided into multiple consecutive bit segments of a fixed length, with each segment being 8 bits long. Since the total length may not be a multiple of 8, the last bit segment with less than 8 bits is treated as having the actual number of bits. Each bit segment corresponds to a set of sequentially arranged position identifiers. Then, the same division is performed on each template bit string to obtain the corresponding set of template bit segments. After aligning the client bit segments with the corresponding template bit segments, a bitwise XOR operation is performed: if two values ​​at the same position are the same, the result is 0; otherwise, the result is 1. This yields a new binary bit string, called the difference bit segment.

[0038] A pre-built mapping lookup table records the correspondence between all possible 8-bit binary strings (from 00000000 to 11111111) and the bits with a value of 1 within them. For example, the binary string 00000001 corresponds to a bit with a value of 1, and 00000011 corresponds to a bit with a value of 2. This lookup table is generated once and stored in memory during system startup. For each difference segment, the binary value of the difference segment is used as an index to quickly retrieve the bits with a value of 1 in the lookup table; this bit count is the difference count for that segment. For the last segment, which is less than 8 bits long, a lookup table of the corresponding length is also pre-built. The difference counts obtained from looking up each segment are accumulated, and the accumulated result is the Hamming distance between the client bit string and the current template bit string.

[0039] A distance threshold of 5 is pre-defined. Each calculated Hamming distance is compared to this threshold. If the Hamming distance corresponding to a template bit string is less than 5, the rule cluster to which the template bit string belongs is marked as a candidate rule cluster; if the Hamming distance is greater than or equal to 5, the rule cluster is discarded. After filtering all rule clusters, a set of candidate rule clusters is obtained. The composite rules in these candidate rule clusters are closely similar to the actual attributes of the customer. Subsequent steps only require fine-tuning the rules within these candidate rule clusters, thus significantly reducing the number of rules that need to be traversed.

[0040] In S5, a subgraph is constructed based on the original attribute set for multiple policies under the customer's name. For each composite rule in the candidate rule cluster, a directed graph traversal is performed in the subgraph to verify whether the policy attributes meet the rule conditions. The value-added benefits corresponding to the rules that meet the conditions are output as the processing result, specifically including: From the original attribute set generated by S1, extract the unique identifier for each policy under the customer's name; this unique identifier is the policy number. Simultaneously, extract the policyholder's ID number and the insured's ID number for each policy; the ID number uses the national ID card number. Treat each policy as a node, and the policy number as the node's unique identifier. Then, iterate through the policyholder's ID number and the insured's ID number for all policies under the customer's name to construct the relationships between the policies.

[0041] Compare all policies pairwise. For any two policies, if the policyholder's ID number is the same, create a policyholder-related edge between them, indicating that both policies are held by the same policyholder. For any two policies, if the insured's ID number of the first policy is the same as the policyholder's ID number of the second policy, create a cross-insurance edge between them, indicating that the insured of the first policy is also the policyholder of the second policy. After traversing all the edges, a set of related edges is obtained.

[0042] Merge all policy nodes with the previously created edges linking to the same policyholder and those linking to each other, and remove duplicate edges (i.e., retain only one edge between two identical policy nodes). Since the relationships are symmetrical, all resulting edges are treated as undirected edges, meaning the direction is not distinguished. This ultimately forms an initial undirected subgraph composed of policy nodes and edges. This subgraph reflects the overlapping relationships between different policies under a customer's name based on the identities of the policyholder and the insured.

[0043] From the candidate rule clusters selected by S4, each composite rule is extracted sequentially. Each composite rule contains a sequence of attribute dimensions and the connection order between these dimensions. For example, the attribute dimension sequence might be "car insurance continuous insurance period equals 3 years" followed by "health insurance claim record equals no claims". Based on this connection order, a path search is performed in the initial undirected subgraph using a depth-first approach. Specifically, starting from any policy node in the subgraph, the system checks whether the policy meets the value condition of the first attribute dimension required by the rule; if it does, it moves along the associated edge to the next adjacent policy node and continues matching the next attribute dimension in the rule.

[0044] During the path search, upon reaching each policy node, based on the attribute dimension that needs to be matched (e.g., insurance type, continuous insurance years, or claims history), the actual value of the corresponding dimension for that policy is retrieved from the original attribute set and compared with the value in the rule conditions. If the actual value matches the rule conditions, the search continues along the connection order to the next dimension node; if they do not match, the current path search is immediately stopped, and the search backtracks to the previous step to try other branch paths. If all branch paths fail to match all dimensions, the current composite rule is deemed not to meet the conditions.

[0045] When all attribute dimension nodes of a composite rule are successfully matched and the rule is fully traversed in the required connection order, the composite rule is deemed to meet the conditions. The corresponding value-added benefits item for this composite rule is recorded, such as "Travel Health Protection Package". After performing the above verification on each composite rule in all candidate rule clusters, the value-added benefits corresponding to all satisfying composite rules are output as the final processing result. The output result is indexed by customer and includes information such as benefit name, validity period, and usage frequency for subsequent benefit distribution or notification.

[0046] Please see Figure 2 As shown, a value-added benefits data processing system based on policy data includes: The policy data collection module is used to collect data on multiple insurance policies under a customer's name, including: insurance type, continuous insurance period and claims record, and generate the customer's original attribute set based on the data of multiple insurance policies; The rule hypergraph construction module, based on a preset composite rule library, uses the multiple attribute dimensions associated with each composite rule as nodes, and constructs a rule hyperedge by connecting all the attribute dimensions involved in each rule, thus forming a rule hypergraph; The rule cluster segmentation and template generation module performs spectral clustering segmentation on all rule hyperedges in the rule hypergraph to obtain multiple rule clusters, and generates a low-dimensional attribute combination template and the corresponding template bit string for each rule cluster; The customer bit string matching and filtering module converts the original attribute set into customer bit strings according to the attribute dimension encoding order, calculates the Hamming distance between the customer bit string and each template bit string, and filters out the rule clusters with Hamming distance less than the threshold as candidate rule clusters. The subgraph traversal and rights verification module constructs a subgraph of multiple insurance policies under a customer's name based on the original attribute set. For each composite rule in the candidate rule cluster, it performs a directed graph traversal in the subgraph to verify whether the policy attributes meet the rule conditions. The value-added rights corresponding to the rules that meet the conditions are output as the processing results.

[0047] The working principle of this invention is as follows: Data on multiple insurance policies under a customer's name is collected from the insurance company's core underwriting system, including policy type, continuous insurance period, and claims records, generating an original attribute set. Based on a preset composite rule library, the attribute dimensions associated with each composite rule are used as nodes. A rule hyperedge is constructed by connecting all the attribute dimensions involved in each rule, forming a rule hypergraph. Then, spectral clustering is performed on all rule hyperedges in the rule hypergraph to obtain multiple rule clusters, and a low-dimensional attribute combination template and corresponding template bit string are generated for each rule cluster. Next, the original attribute set is converted into customer bit strings according to the attribute dimension encoding order. The Hamming distance between the customer bit string and each template bit string is calculated, and rule clusters with a Hamming distance less than a threshold are selected as candidate rule clusters. A subgraph of multiple insurance policies under the customer's name is constructed based on the original attribute set. A directed graph traversal is performed on each composite rule in the candidate rule cluster within the subgraph to verify whether the policy attributes meet the rule conditions. The value-added benefits corresponding to the rules that meet the conditions are output as the processing result.

[0048] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A method for processing value-added benefits data based on policy data, characterized in that, Includes the following steps: S1. Collect data on multiple insurance policies under the customer's name, including: insurance type, continuous insurance period and claims record, and generate the customer's original attribute set based on the data of multiple insurance policies; S2, based on the preset composite rule library, take the multiple attribute dimensions associated with each composite rule as nodes, and construct a rule hyperedge by connecting all the attribute dimensions involved in each rule to form a rule hypergraph; S3, perform spectral clustering segmentation on all rule hyperedges in the rule hypergraph to obtain multiple rule clusters, and generate a low-dimensional attribute combination template and the corresponding template bit string for each rule cluster; S4. Convert the original attribute set into a client bit string according to the attribute dimension encoding order, calculate the Hamming distance between the client bit string and each template bit string, and select the rule clusters with Hamming distance less than the threshold as candidate rule clusters. S5: Construct a subgraph of multiple insurance policies under the customer's name based on the original attribute set. For each composite rule in the candidate rule cluster, perform a directed graph traversal in the subgraph to verify whether the policy attributes meet the rule conditions. Output the value-added benefits corresponding to the rules that meet the conditions as the processing result.

2. The method for processing value-added rights data based on policy data according to claim 1, characterized in that, S2 specifically includes: Extract all attribute dimensions contained in each composite rule in the preset composite rule library, and assign a unique position identifier to each attribute dimension; For each composite rule, generate a binary bit string with a length equal to the total number of attribute dimensions. Set 1 to the corresponding position of the attribute dimension involved in the rule and set 0 to the other positions. Use the binary bit string as the rule superedge corresponding to the rule. Aggregate all rule hyperedges according to their position identifiers to form a rule hypergraph with attribute dimensions as nodes and binary bit strings as hyperedge connections.

3. The method for processing value-added rights data based on policy data according to claim 2, characterized in that, The formation process of the regular hypergraph is as follows: For each attribute dimension, a corresponding attribute node is created based on its location identifier. All attribute nodes form a node set. Traverse all rule hyperedges. For each rule hyperedge's binary bit string, scan the position markers that are set to 1, retrieve the corresponding attribute nodes from the node set, and form the associated node group of the rule hyperedge. Using each rule hyperedge as a connection, all attribute nodes in the associated node group are aggregated into one through the hyperedge, generating a many-to-many mapping structure between attribute nodes and hyperedges, thus obtaining a rule hypergraph.

4. The method for processing value-added rights data based on policy data according to claim 1, characterized in that, S3 specifically includes: Calculate the number of attribute dimensions shared between each pair of rule hyperedges in the rule hypergraph, and use the number of shared dimensions as the similarity to construct a hyperedge similarity matrix; Transform the hyperedge similarity matrix into a Laplacian matrix, and take the eigenvector corresponding to the second smallest eigenvalue of the Laplacian matrix as the segmentation basis; All rule hyperedges are segmented according to the ascending order of the values ​​of each component in the feature vector to obtain multiple rule clusters; Extract the common attribute dimensions of all rule superedges within each rule cluster, combine the common attribute dimensions into a low-dimensional attribute combination template for the rule cluster, and generate the corresponding template bit string based on the value of the attribute dimension in the template.

5. The method for processing value-added rights data based on policy data according to claim 4, characterized in that, The construction of the hyperedge similarity matrix specifically includes: Extract the binary string of each rule superedge. The length of the binary string is equal to the total number of attribute dimensions. For any two rule superedges, perform a bitwise logical AND operation on their binary bit strings to obtain the overlapping bit string; Count the number of positions with a value of one in the overlapping bit string, use the number of positions as the number of shared attribute dimensions between the two rule super edges, and use the number of shared dimensions as the similarity value. Fill the matrix with the similarity values ​​calculated for all paired rule hyperedges, according to rows and columns, to obtain the hyperedge similarity matrix.

6. The method for processing value-added rights data based on policy data according to claim 1, characterized in that, S4 specifically includes: Get the actual values ​​of each attribute dimension in the original attribute set, and judge whether each attribute dimension satisfies the value conditions in the corresponding rule cluster template according to the attribute dimension encoding order. If it satisfies, assign 1 to the corresponding bit; otherwise, assign 0 to generate the client bit string. After aligning the customer bit string with each template bit string, compare them bit by bit. Whenever the corresponding bit values ​​are different, accumulate the difference count to obtain the Hamming distance corresponding to the template bit string. Compare all Hamming distances with a preset threshold, extract template bit strings whose Hamming distance is less than the threshold, and mark the rule cluster to which the template bit string belongs as a candidate rule cluster.

7. A method for processing value-added rights data based on policy data according to claim 6, characterized in that, The calculation process for the Hamming distance is as follows: The client bit string is divided into multiple consecutive bit segments of fixed length, and each bit segment corresponds to a set of sequentially arranged position identifiers. Perform the same partitioning on each template bit string to obtain the corresponding template bit segment. Align the client bit segment with the template bit segment and then perform a bitwise XOR operation to obtain the difference bit segment. A lookup table is pre-established to map each possible difference segment to the number of bits with a value of one. The difference count in each difference segment is obtained by looking up the table. The difference counts of each difference segment are accumulated to obtain the Hamming distance.

8. The method for processing value-added rights data based on policy data according to claim 1, characterized in that, S5 specifically includes: Extract the unique identifier of each policy and the relationship between policies from the original attribute set, and construct an initial subgraph with policies as nodes and relationships as edges; Obtain the attribute dimension sequence and connection order contained in each composite rule in the candidate rule cluster. Perform a depth-first path search along the connection order in the initial subgraph. When each attribute dimension node is reached, verify whether the actual attribute value of the policy matches the rule conditions. If all attribute dimension nodes in the path search are successfully matched and the connection order is complete, the composite rule is deemed to meet the conditions; otherwise, the current search is stopped and the verification of the next rule is switched.

9. A method for processing value-added rights data based on policy data according to claim 8, characterized in that, The construction of the initial subgraph specifically includes: Extract the unique identifier of each policy, as well as the policyholder's ID number and the insured's ID number, from the original attribute set; Iterate through all policies. If two policies have the same policyholder ID number, create a policyholder-related edge between the two policies. If the insured's ID number of one policy is the same as the policyholder's ID number of another policy, create a mutual insurance-related edge between the two policies. Treat all insurance policies as nodes, merge and deduplicate all edges related to the same policyholder and mutual insurance, and form an initial undirected subgraph between policy nodes.

10. A value-added benefits data processing system based on policy data, characterized in that, For execution, a method for processing value-added rights data based on policy data as described in any one of claims 1-9, comprising: The policy data collection module is used to collect data on multiple insurance policies under a customer's name, including: insurance type, continuous insurance period and claims record, and generate the customer's original attribute set based on the data of multiple insurance policies; The rule hypergraph construction module, based on a preset composite rule library, uses the multiple attribute dimensions associated with each composite rule as nodes, and constructs a rule hyperedge by connecting all the attribute dimensions involved in each rule, thus forming a rule hypergraph; The rule cluster segmentation and template generation module performs spectral clustering segmentation on all rule hyperedges in the rule hypergraph to obtain multiple rule clusters, and generates a low-dimensional attribute combination template and the corresponding template bit string for each rule cluster; The customer bit string matching and filtering module converts the original attribute set into customer bit strings according to the attribute dimension encoding order, calculates the Hamming distance between the customer bit string and each template bit string, and filters out the rule clusters with Hamming distance less than the threshold as candidate rule clusters. The subgraph traversal and rights verification module constructs a subgraph of multiple insurance policies under a customer's name based on the original attribute set. For each composite rule in the candidate rule cluster, it performs a directed graph traversal in the subgraph to verify whether the policy attributes meet the rule conditions. The value-added rights corresponding to the rules that meet the conditions are output as the processing results.

Images