Real-name information full-link de-identification and security management system and method

By constructing a real-name information link set and restoring the association graph, and combining it with the particle swarm optimization algorithm, the problem of cross-scenario information leakage risk in existing technologies is solved, and dynamic security control and risk suppression of the entire link of real-name information are realized.

CN123116322APending Publication Date: 2026-11-17WUHU SIMBA NETWORK TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611262508.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-19
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing technologies lack the ability to identify risks of information splicing and restoration across fields, nodes, and interfaces in the end-to-end desensitization and security management of real-name information. This makes it difficult to achieve dynamic policy generation and collaborative blocking, resulting in the continued existence of information leakage risks.

Method used

By constructing a set of real-name information links, it is divided into real-name exposure units that have the functions of identity pointing, association supplementation and combination restoration, and a restoration association graph is established. Combined with particle swarm optimization algorithm for iterative optimization, an exposure suppression strategy is generated to collaboratively block cross-scene restoration paths.

Benefits of technology

It enables the overall identification and quantitative analysis of recovery paths across fields, nodes, and interfaces, dynamically generates risk suppression strategies, and improves the pertinence and accuracy of full-link security management of real-name information.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123116322A_ABST
    Figure CN123116322A_ABST
Patent Text Reader

Abstract

This invention discloses a system and method for full-link desensitization and security management of real-name information, relating to the fields of data security and personal information protection. It collects source data, flow data, and access data during the real-name business processing, associates various data execution entities, identifies nodes, and organizes time sequences to construct a real-name information link set. Based on this, it decomposes the real-name information in the link into exposure units and establishes a restoration association graph between different exposure units. Furthermore, it combines current access requests and historical access results to determine the cumulative exposure status, and uses a particle swarm optimization algorithm to jointly optimize the desensitization granularity, field restriction range, path blocking strength, and approval triggering conditions, generating an exposure suppression strategy adapted to the access scenario. This achieves dynamic desensitization of real-name information, restoration risk suppression, and full-link security management.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of data security and personal information protection technology, specifically to a system and method for full-link de-identification and security management of real-name information. Background Technology

[0002] With the widespread application of real-name registration in government, finance, telecommunications, healthcare, and internet services, real-name information continuously flows through multiple stages, including collection, transmission, storage, retrieval, access, and archiving. Existing technologies for anonymization and access control largely rely on single-request, single-field processing, or static permission checks, often only partially restricting the current output content and lacking the ability to jointly identify historical access results, cross-node flow relationships, and cross-interface splicing risks. Consequently, even if a single output content has been anonymized, information exposed at different times, nodes, and interfaces can still be used to restore identity through supplementation, connection, and combination, making traditional solutions ineffective in addressing the risk of continuous, cross-link, and cross-scenario information leakage. Furthermore, existing solutions typically lack a dynamic balancing mechanism between access task requirements, the scope of output availability, and restoration risk constraints, making it difficult to achieve refined management while ensuring the availability of legitimate business operations. Therefore, a full-link anonymization and security management solution for real-name information is needed, capable of dynamically generating policies and collaboratively blocking based on full-link data relationships, historical exposure states, and restoration path risks. Summary of the Invention

[0003] This invention addresses the shortcomings of existing technologies by proposing a full-link de-identification and security management system and method for real-name information. It can accurately identify the risks of information splicing and restoration across fields, nodes, and interfaces, and realize dynamic de-identification and full-link collaborative management of real-name information, thereby improving the accuracy of security protection while ensuring business availability.

[0004] The technical solution to achieve the objective of this invention is as follows:

[0005] On the one hand, the method for full-chain de-identification and security control of real-name information includes the following steps:

[0006] Collect source data, flow data and access data during the real-name business processing, associate various data execution entities, identify nodes and process time sequence, and construct a real-name information link set;

[0007] Based on the real-name information link set, the real-name information in it is split into real-name exposure units, and a restored association graph is established according to the relationship between each real-name exposure unit.

[0008] Based on real-name exposure units and restored correlation graphs, joint analysis is performed on current access requests and historical access results to determine the cumulative exposure status. Particle swarm optimization algorithm is used for iterative optimization to generate corresponding exposure suppression strategies.

[0009] Based on the exposure suppression strategy, link output control is performed on the current access request, and collaborative blocking is implemented on associated paths that have a tendency to recover, generating security control results.

[0010] Furthermore, the real-name exposure unit splitting process includes:

[0011] For all real-name information fields in the real-name information link set, using all real-name subjects as a unified calculation benchmark, calculate the matching range reduction ratio and subject locking probability of each field in individual and combined states.

[0012] Based on the quantitative evaluation results of the narrowing of the matching range and the probability of locking the subject, the real name information field is labeled as the corresponding main function attribute label in the identity pointing function, association supplementation function or combined restoration function. For fields that have multiple function attributes at the same time, the main function attribute label is determined according to the preset priority.

[0013] The real-name information field is split according to the standardized splitting rules corresponding to each main function attribute. The basic exposure risk score of each splitting unit is calculated based on the main function attribute label and the node type to which it belongs, and a real-name exposure unit library is generated.

[0014] Furthermore, the establishment of the reconstructed correlation map includes:

[0015] For all real-name exposure units in the real-name exposure unit library, relationship mining is performed in three scenarios: field supplementation, node connection, and call combination. Based on the quantitative formulas corresponding to each scenario, the relationship strength values ​​of field supplementation, node connection, and call combination are calculated. Invalid relationships that do not reach the effective relationship threshold are removed, and a full relationship library of real-name exposure units is generated.

[0016] Using real-name exposed units in the real-name exposed unit library as graph nodes and valid associations in the full association database as directed weighted edges, the direction of the edges and the corresponding association strength value are assigned according to the logical direction of the restoration path to construct a directed weighted restoration association graph.

[0017] In the directed weighted reconstruction association graph, directed links that meet the reconstruction probability threshold are defined as valid reconstruction paths. All valid reconstruction paths are mined and their attributes are solidified. A global multidimensional index system is established to obtain a fully structured reconstruction association graph.

[0018] Furthermore, iterative optimization using the particle swarm optimization algorithm includes:

[0019] Using cumulative exposure status, access task requirements, output availability range, and recovery risk constraints as boundary conditions, a four-dimensional strategy parameter space is constructed, which includes desensitization granularity, field restriction range, path blocking strength, and approval triggering conditions.

[0020] The parameter combinations in the strategy parameter space are mapped to particle position vectors, and a multi-objective fitness function is constructed with task satisfaction score, risk inhibition rate score and approval efficiency score as objective terms.

[0021] The weights of each objective in the multi-objective fitness function are dynamically adjusted based on the cumulative exposure level. Particle velocity and position are iteratively updated according to the standard particle swarm rules. The globally optimal parameter combination is output to generate an exposure suppression strategy.

[0022] Furthermore, joint analysis is performed on the current access requests and historical access results, including:

[0023] Lock all real-name exposed units involved in the current access request and generate a set of exposed units for the current request; collect the valid real-name exposed units that have been output in the past within the preset backtracking window of the access subject and generate a set of historically output exposed units;

[0024] Based on the restored association graph, a full association analysis is performed on the current request exposure unit set and the historical output exposure unit set to calculate the degree of supplementation and connection between the current access behavior and the historical exposure information.

[0025] Based on the degree of supplementation, connection relationship and basic exposure risk score of each real-name exposure unit, the cumulative exposure risk score is calculated, the cumulative exposure status level is determined according to the preset level classification rules, and a standardized cumulative exposure status data packet is generated.

[0026] Furthermore, coordinated blocking of associated pathways showing a tendency to relapse is implemented, including:

[0027] For the current access behavior, calculate the incremental restoration risk value, cross-scenario association matching degree and splicing risk coefficient, compare the three core quantitative indicators with the restoration boundary judgment threshold, determine whether the current access behavior has a tendency to supplement restoration across fields, a tendency to connect restoration across nodes or a tendency to splice restoration across interfaces, and mark the corresponding restoration tendency risk level.

[0028] Based on the path blocking strength, the range of high-risk restoration paths that need to be blocked is identified in the restoration association graph. Temporary access permission interception, field call restriction and flow channel circuit breaking are performed on the standard business nodes, service interfaces and data transmission links corresponding to the core association edges within the path.

[0029] After the blocking is completed, the total path recovery probability is verified. Once it is confirmed that the probability has dropped below the preset recovery threshold, the full-link collaborative blocking is completed and the blocking record is fixed.

[0030] Furthermore, link output control is performed on the current access request, including:

[0031] For access behaviors that do not reach the preset restoration boundary, based on the desensitization granularity in the exposure suppression strategy, the corresponding processing of no desensitization, mild desensitization, moderate desensitization, severe desensitization or complete blocking is performed on the real-name exposed unit to be output, and a preliminary desensitization output result is generated;

[0032] For access behaviors that show a tendency to revert to previous behavior, based on the risk level of the tendency to revert to previous behavior and the approval triggering conditions, restrict output, replace output, circuit breaker output or add approval processing for key real-name exposed units in the associated path.

[0033] Perform policy compliance verification, restoration risk verification, and business requirement adaptability verification on all output results. If the verification fails, the control level is upgraded and the handling is re-executed until all verifications pass and a security control result is generated.

[0034] Furthermore, the construction of the real-name information link set includes:

[0035] By combining bypass traffic mirroring collection, system log embedding collection, full audit collection of interface calls, and storage node snapshot collection, we perform full real-time data capture of source data, flow data, and access data, and perform unique hash value storage for each captured data.

[0036] Data cleaning, format normalization, and metadata standardization extraction are performed on all data. The unique root identifier of the real-name subject is used as the binding benchmark. The association processing of the real-name subject is performed on all data subjects through two methods: direct matching and chain association matching.

[0037] The full data is classified according to the standard business node classification system, and the full data is processed according to the time sequence of the business life cycle benchmark time axis to construct a real-name information link set with the full life cycle flow trajectory of real-name information as the core content.

[0038] Furthermore, after generating the security control results, it also includes:

[0039] A standardized operation audit log is generated for the entire process. The operation audit log includes the unique identifier of the access subject, the cumulative exposure status, the result of the recovery tendency judgment, the details of the exposure suppression strategy execution, and the final output content. A unique hash value is used for evidence storage to ensure that it cannot be tampered with.

[0040] The real-name exposure unit of the final output is synchronously updated to the historical output exposure unit set of the corresponding access subject. The call behavior correction coefficient of the corresponding relationship in the restoration association graph is updated according to the call record of this call. The association strength value and the total restoration probability of the effective restoration path are recalculated to complete the dynamic iteration of the graph.

[0041] The characteristics of this access scenario, the effect of policy execution, and the results of risk verification are stored in the policy optimization sample library for continuous iteration of the particle swarm optimization algorithm's optimization model.

[0042] Secondly, the real-name information full-link de-identification and security management system includes a link construction module, a restoration and association module, a policy generation module, and a management execution module:

[0043] Link construction module: used to collect source data, flow data and access data in the real-name business processing process, perform subject association, node identification and time sequence processing of various types of data, and build real-name information link set;

[0044] The restoration association module is used to perform real-name exposure unit splitting on the real-name information link set and establish a restoration association graph based on the association relationship between each real-name exposure unit.

[0045] The strategy generation module is used to perform joint analysis on the current access request and historical access results based on the real-name exposure unit and the restored association map, determine the cumulative exposure status, and use the particle swarm optimization algorithm to iteratively optimize and generate the corresponding exposure suppression strategy.

[0046] Control and Execution Module: Based on the exposure suppression strategy, it performs link output control on the current access request, and at the same time implements coordinated blocking on related paths with a tendency to recover, and generates security control results.

[0047] Compared with the prior art, the advantages of this invention are as follows:

[0048] 1. By breaking down real-name information into real-name exposure units that have identity pointing, association supplementation, and combination restoration functions, and further establishing a restoration association graph that represents field supplementation, node connection, and call combination relationships, it is possible to expand from single-field desensitization to the overall identification and quantitative analysis of restoration paths across fields, nodes, and interfaces. This allows for the early detection of hidden restoration risks that are difficult to cover by traditional static desensitization methods, thereby improving the pertinence and completeness of real-name information restoration prevention and control.

[0049] 2. By combining current access requests with historical access results to determine the cumulative exposure status, and under the constraints of access task requirements, output availability, and recovery risk, the particle swarm optimization algorithm is used to jointly optimize the desensitization granularity, field restriction range, path blocking strength, and approval triggering conditions. This enables the dynamic generation of exposure suppression strategies for different access scenarios, achieving optimal risk suppression while taking into account business availability, and further improving the adaptive capability and execution accuracy of full-link security management of real-name information. Attached Figure Description

[0050] Figure 1 Flowchart of the method for full-link de-identification and security control of real-name information;

[0051] Figure 2 This is a schematic diagram of the splitting and restoration of the association map of the real-name exposure unit in this invention;

[0052] Figure 3 This is a flowchart illustrating the security control execution process in this invention.

[0053] Figure 4 This is a schematic diagram of the structure of the real-name information end-to-end desensitization and security management system in this invention. Detailed Implementation

[0054] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0055] Example 1

[0056] This invention discloses a method for end-to-end de-identification and security management of real-name information, comprising the following steps:

[0057] S1: Collect source data, flow data and access data during the real-name business processing, perform subject association, node identification and time sequence processing of various data, and construct a real-name information link set.

[0058] S2: Based on the real-name information link set, the real-name information involved in the link is split and identified according to its identity pointing role, association supplementation role and combination restoration role to obtain the corresponding real-name exposure unit; based on the association relationship of each real-name exposure unit in field supplementation, node connection and call combination, a restoration association map is established to represent the restoration path between different real-name exposure units.

[0059] S3: Based on the real-name exposure units and the restoration association graph, perform joint analysis on the real-name exposure units involved in the current access request and the real-name exposure units already output in the historical access results to identify the degree of supplementation and connection between the current access behavior and the existing exposure information, and determine the corresponding cumulative exposure status; construct a strategy parameter space based on the cumulative exposure status, access task requirements, output availability range and restoration risk constraints, and use the particle swarm optimization algorithm to jointly optimize the desensitization granularity, field restriction range, path blocking strength and approval triggering conditions to generate an exposure suppression strategy that matches the current access scenario.

[0060] S4: Based on the exposure suppression strategy, perform link output control on the output content corresponding to the current access request, and combine the restoration association graph to implement collaborative blocking of association paths with restoration tendency; for access behavior that does not touch the preset restoration boundary, output the desensitization result corresponding to the access task; for access behavior with restoration tendency that forms cross-field supplementation, cross-node connection or cross-interface splicing with historical access results, perform output restriction, output replacement, circuit breaker output or additional approval processing on key real-name exposure units in the association path, and generate the corresponding security control result.

[0061] refer to Figure 1 , Figure 1 A flowchart illustrating the method for end-to-end de-identification and security management of real-name information.

[0062] In step S1, source data, flow data, and access data are collected during the real-name business processing. The data is then processed by associating various data entities, identifying nodes, and organizing time sequences to construct a real-name information link set, including:

[0063] S101: Scope of data collection and full data capture for real-name business across the entire chain.

[0064] The data collection boundaries for the entire lifecycle of real-name registration services are pre-defined, covering all business processes from the initiation of real-name information collection to archiving and destruction. Fixed definitions and collection specifications are clearly defined for three types of collection objects: source data, flow data, and access data. Source data refers to the original basic information of the real-name entity collected at the initiating end of the real-name registration service, including but not limited to legal identity information, biometric information, qualification certificate information, and basic business-related information. Flow data refers to the complete flow trajectory data generated during the transmission, forwarding, temporary storage, and processing of real-name information between business system functional modules, cross-system service interfaces, data platforms, and storage nodes, including but not limited to transmission messages, processing logs, data temporary storage snapshots, interface call parameter fields, and flow approval records. Access data refers to the complete behavioral data corresponding to all query, retrieval, modification, export, comparison, and verification operations performed by all operating entities on real-name information, including but not limited to access entity identifiers, access interface numbers, access timestamps, request parameter fields, return content fields, operation terminal identifiers, permission verification credentials, and operation behavior audit logs.

[0065] By combining bypass traffic mirroring collection, system log tracking collection, full audit collection of interface calls, and storage node snapshot collection, the system performs full and complete real-time capture of source data, flow data, and access data generated during the real-name business processing. Simultaneously, a unique hash value is used to store each captured raw data to lock the original state of the data and prevent data tampering and loss.

[0066] S102: Standardized preprocessing of collected data.

[0067] For the full source data, flow data, and access data captured by S101, data cleaning, format normalization, and metadata standardization extraction are performed sequentially to complete the standardized preprocessing of the collected data.

[0068] Data cleaning removes duplicate and redundant data, invalid null values, abnormally tampered data, and non-business-related interference data from the full collection of data, retaining valid data directly related to real-name business and real-name entities. Format normalization converts structured, semi-structured, and unstructured data from different data sources and storage structures into a unified JSON structured format, applying unified character encoding, field naming, and data type standards to all data fields to eliminate format differences in heterogeneous data. Metadata standardization extraction extracts standardized metadata fields with fixed dimensions from each valid data entry. These metadata fields include, but are not limited to: unique data identifier, data business type, data generation timestamp, real-name entity identifier associated with the data, system node information where the data resides, data operation type, and data lifecycle status, providing a standardized data foundation for subsequent processing.

[0069] S103: Full data subject association processing based on the unique root identifier of the real-name subject.

[0070] A unique root identifier system for real-name entities is pre-defined, using legally unique identity codes such as citizen identification number and unified social credit code as the unique root identifier for real-name entities. The full amount of data that has completed standardized preprocessing in S102 is then subjected to root identifier matching and entity association binding.

[0071] The specific processing flow is as follows: First, for each piece of preprocessed valid data, identify the root identifier field of the real-name subject contained therein. If there is a directly matching valid root identifier in the data, directly bind the data with the corresponding real-name subject to generate a one-to-one subject-data association mapping relationship; if there is no directly matching root identifier in the data, complete the indirect binding of the data with the corresponding real-name subject through the business association field contained in the data; finally, for all the data that has completed the subject binding, group and collect them according to the real-name subject to generate a complete set of subject data with a single real-name subject as the smallest collection unit, ensuring that all source data, flow data, and access data generated by the same real-name subject in the entire business process are completely collected under the corresponding subject data set, without mismatch of ownership or data omission.

[0072] S104: Data node identification processing based on the entire lifecycle of real-name business.

[0073] A standard node classification system is pre-defined for the entire lifecycle of real-name registration services. Based on the full-process division of personal information processing activities in the "Information Security Technology - Provisions on Notification and Consent in Personal Information Processing," this system divides the real-name registration process into eight standard business nodes: information collection node, information transmission node, information storage node, information processing node, information retrieval node, information access node, information archiving node, and information destruction node. Each standard business node is assigned a unique node type code, node permission attributes, and node flow rules. The node permission attributes define the types of operations that the node can perform and the scope of data access, while the node flow rules clarify the triggering conditions and verification mechanisms for data transmission between nodes.

[0074] For each valid piece of data in the complete dataset of the main data corresponding to each real-name entity generated by S103, full-link node attribution labeling is performed. The specific process is as follows: First, identify the data's generation location, system module, interface number, and operation behavior type, match the corresponding standard business node, and assign a unique node type code, unique node identifier, system information, and upstream and downstream relationships to the data. Second, for the flow behavior of the same real-name information between different standard business nodes, label the flow direction, flow triggering conditions, flow data content, and flow timing markers between nodes. Finally, for the complete dataset of the main data of each real-name entity, generate a node topology matrix for the entire business process, clarify the input data, output data, associated nodes, and operation permission boundaries of each standard business node, and complete the standardized node labeling of the entire dataset.

[0075] S105: Time-series processing of full data based on business lifecycle benchmarks.

[0076] Using the initiation time of a single real-name business as the starting point and the completion and archiving time of the business as the ending point, a unique business lifecycle timeline is constructed for each complete real-name business process.

[0077] For all data from the S104 node calibration, a full-process time-series sorting and alignment process is performed. The specific process is as follows: First, the timestamps of all valid data undergo unified time zone conversion and precision calibration, uniformly converting them to millisecond-level timestamps in the UTC+0 time zone to eliminate time deviations and time zone differences between different systems and nodes. Second, according to the calibrated timestamp order, all source data, flow data, and access data under the same real-name entity and the same business process are sorted in ascending order to generate a time-series continuous business data sequence. Third, for abnormal data with time conflicts, time inversions, or time overlaps in the time sequence, time-series logic verification and correction are performed, and abnormal timestamps are calibrated based on the inherent logical sequence of the real-name business process. Finally, a standardized time-series data sequence is generated, which is time-series continuous and logically closed-loop, with each business process as the unit.

[0078] S106: Construction and validity verification of real-name information link set.

[0079] Based on the standardized time-series data sequence generated by S105, which includes subject association, node identification, and time-series organization, a unique corresponding real-name information single link is constructed for each complete real-name business process of each real-name subject. The real-name information single link takes the real-name subject as the sole belonging subject, is based on the standardized time-series data sequence, uses the identified standard business nodes as link nodes, uses the flow relationship between nodes as link connection edges, and uses the real-name information field as the core content of the link. It fully represents the entire lifecycle trajectory of the generation, flow, access, processing, archiving, and destruction of real-name information in the entire process of a single real-name business from initiation to completion.

[0080] All single links of real-name information corresponding to all real-name entities are fully collected and integrated to construct a real-name information link set that provides full coverage, full-link tracking, and full-time closed-loop functionality. Simultaneously, a global multi-dimensional index system is established for this real-name information link set. This system includes a subject root identifier index, a node type index, a time-series interval index, and a business type index. The index system supports millisecond-level accurate retrieval and rapid positioning of single real-name information links, single standard business nodes, and single real-name information fields within the link set. Specifically, the subject root identifier index uses a B+ tree structure, supporting precise matching and range queries of the root identifier; the node type index uses a hash table structure, supporting rapid filtering by node type; the time-series interval index uses a time-segmented index structure, supporting retrieval by time range; and the business type index uses an inverted index structure, supporting related queries by business type.

[0081] A full validity check is performed on the completed real-name information link set. The check dimensions include: the uniqueness of the subject ownership of each real-name information link, the closed-loop logic of node flow, the continuity of the time sequence, and the integrity of data content. Invalid links that fail the check are removed, and supplementary backtracking collection and standardization processing are performed on links with missing data to ensure that the final real-name information link set fully covers the entire life cycle flow trajectory of all real-name information in the real-name business processing process, with no ownership mismatch, no data omission, and no logical conflict.

[0082] In step S2, based on the real-name information link set, the real-name information is split into real-name exposure units, and a restored association graph is established according to the association relationships between each real-name exposure unit, including:

[0083] S201: Full field parsing of real-name information link set and full extraction of real-name information.

[0084] Taking the real-name information link set constructed by S1 as input, it traverses all single links of real-name information corresponding to all real-name subjects in the link set. Based on the node topology relationship matrix of the single link, it performs node-by-node parsing on all data in the eight standard business nodes of information collection, information transmission, information storage, information processing, information retrieval, information access, information archiving, and information destruction. It removes business process control data, system operation log data, and interface status data that are not related to the identity attributes of the real-name subject, and extracts all real-name information fields that are directly related to the real-name subject, can represent the identity attributes of the subject, and can participate in the identification of the subject. It binds the corresponding subject root identifier, the unique identifier of the link to which it belongs, the unique identifier of the node to which it belongs, the field generation timestamp, and the field lifecycle status to each field, and generates a full set of fields for real-name information.

[0085] S202: Classification and standardized decomposition of multi-dimensional role attributes of real-name information.

[0086] For all real-name information fields in the full set of real-name information fields generated by S201, perform splitting and identification processing based on identity pointing function, association supplement function, and combination restoration function to generate the smallest unit set of real-name information.

[0087] During the processing, the total number of all real-name registered entities that meet the business coverage scope... To ensure a unified calculation standard, for any real-name information field , or by A combination of fields The number of real-name entities whose field values ​​in the full entity benchmark database are exactly matched. Based on this, two core quantitative indicators are calculated:

[0088] Matching range narrowing ratio The formula for calculating the convergence effect of the combination of characterizing fields on the subject filtering range is as follows:

[0089] ,

[0090] in, The range of values ​​is The larger the value, the stronger the convergence ability of the field combination to the subject filtering range; this formula is based on the concept of complement in set theory, and quantifies the degree of narrowing of the filtering range by calculating the proportion of unmatched subjects to the total subjects.

[0091] Subject locking probability The ability of a combination of characterizing fields to uniquely identify a real-name entity is calculated using the following formula:

[0092] ,

[0093] in, The range of values ​​is The larger the value, the stronger the ability of the field combination to lock a unique real-name subject; the first product factor in the formula represents the probability of achieving unique locking within the matching subject set, and the second product factor represents the rarity of the matching subject set among all subjects. The identity recognition capability of the field combination is fully quantified through the product of the two terms.

[0094] Based on the above quantitative indicators, a unique primary attribute label is assigned to each real-name information field: a single real-name information field. of A preset matching upper limit is defined as having an identity pointing function. An example value of the preset matching upper limit is 5. This value is based on the industry-standard judgment of the identifiability of personal information. When the number of matched subjects does not exceed 5, the field has a strong identity pointing capability and can be directly used for accurate subject positioning. Real-name information fields with identity pointing function include, but are not limited to: the full number of citizen identity number, the full number of unified social credit code, the full number of valid bank card number, and the full number of online mobile phone number.

[0095] Single real-name information field of Preset matching upper limit and A preset convergence threshold is set, with an exemplary value of 1000. This value corresponds to the subject screening granularity in a medium-sized business scenario. When there are more than 1000 matching subjects, a single field has no actual identity positioning value and can only be used as a supplementary screening dimension or a single real-name information field. Combined with any other field Yes, it is marked as having a supplementary association function; real-name information fields with a supplementary association function include, but are not limited to: subject name, gender, ethnicity, date of birth, registered address, permanent address, employer name, education information, and occupation category;

[0096] Single real-name information field of Preset convergence threshold, but a single real-name information field Combined with 2 or more other fields or The combined restoration function is defined as follows: a 95% convergence rate and a 90% locking probability correspond to a risk level that is close to unique identification after the combination of multiple fields, which matches the risk attributes of the combined restoration fields; the real-name information fields with combined restoration function include, but are not limited to: business processing outlet number, the last 8 digits of the business transaction number, the last 6 digits of the bank card number, the last 4 digits of the mobile phone number, the business processing timestamp, the order payment amount, and the last 4 digits of the main receiving address;

[0097] For composite fields that simultaneously possess multiple functional attributes, the primary functional attribute is labeled according to the priority of "identity pointing function > association supplementary function > combination restoration function," and the secondary functional attributes are labeled simultaneously. After the fields are labeled, standardized splitting is performed: identity pointing function fields are split into legally complete and independent units, and cannot be split into fragments without independent identity pointing capabilities; association supplementary function fields are split into independent attribute dimensions, with each independent attribute dimension being a minimum unit, such as the address field being split into five independent units: province, city, district, street, and house number; combination restoration function fields are split into the minimum fragments that can independently participate in combination restoration, such as the timestamp being split into five independent units: year, month, day, hour, and minute; after the splitting is completed, a set of minimum real-name information units is generated.

[0098] S203: Identification, labeling and attribute solidification of real-name exposed units.

[0099] The minimum set of real-name information units generated by S202 is subjected to validity identification, uniqueness marking, and attribute solidification to generate a real-name exposure unit library. First, invalid units without independent functional attributes, unable to participate in combined restoration, or without exposure risk are removed, and valid units that meet the requirements are retained and confirmed as real-name exposure units. Each real-name exposure unit is assigned a globally unique identifier compatible with the S1 identifier system, with the encoding format "Main Root Identifier - Link Unique Identifier - Node Unique Identifier - Field Unique Code - Unit Sequence Number". A standardized attribute set is solidified for each real-name exposure unit, including unit unique identifier, main root identifier, belonging link unique identifier, belonging node unique identifier, main functional attribute label, secondary functional attribute label, field original value, field data type, field length, generation timestamp, lifecycle status, basic exposure risk score, and basic exposure risk level.

[0100] The basic exposure risk score is obtained through quantitative calculation and is denoted as . The calculation formula is:

[0101] ,

[0102] in, The probability of locking the subject for a single field of this real-name exposed unit; To represent the smallest field entity corresponding to a single independent real-name exposure unit, that is, the smallest unit of a single piece of real-name information that has independent attributes after standardization and splitting, used to characterize the field entity object when a single information unit exists independently; The weights of the role attributes are set based on the statistical results of the contribution of different role attribute fields in identity restoration. The role weight of identity pointing is 1.0, the role weight of association supplementation is 0.6, the role weight of combined restoration is 0.2, and the role weight of identity pointing is increased to 1.2 in the government affairs scenario. The node risk weights are set based on the statistical results of data leakage risks for different nodes. The weights for information access nodes and information call nodes are 1.0, for information transmission nodes and information processing nodes are 0.7, for information storage nodes and information collection nodes are 0.5, and for information archiving nodes and information destruction nodes are 0.3. The coefficient 100 in the formula is used to convert the probability value of 0%-100% into a standardized score range of 0-100, which is consistent with the magnitude of the node risk score in the latter half, ensuring that the weight ratio of the two indicators conforms to the preset risk assessment logic. The coefficient 20 is used to limit the value range of the node risk weight to 0-20, ensuring that the weight ratio of the core risk item, namely the probability of subject locking and the function attribute, is not less than 80%, which is in line with the industry consensus that "the identity pointing ability of the field itself is the core source of risk" in the risk of exposure of real-name information and the statistical conclusions of information leakage event samples.

[0103] Risk levels are determined based on baseline exposure risk scores, according to the following rules: High risk, Medium risk The risk level is low. This three-level classification standard is aligned with the personal information risk classification specification in the network security level protection. A score of 80 or above corresponds to high identification risk, a score of 40 or below corresponds to low identification risk, and the middle range is medium risk. After all unit attributes are solidified, a real-name exposure unit library is generated.

[0104] S204: Real-name exposed unit association mining and quantitative calibration.

[0105] For all units in the real-name exposure unit library generated by S203, perform full-scale relational mining and intensity quantification for three scenarios: field supplementation, node connection, and call combination, and generate a full-scale relational library of real-name exposure units.

[0106] The specific processing procedure is as follows:

[0107] Field supplementary association identification: For real-name exposure units belonging to the same subject root identifier, the same real-name information single link, and the same standard business node, pairwise combination matching is performed to form a dual-field real-name exposure unit combination. After satisfying the combination Unit pairs with a pre-defined complementary field mapping relationship are identified as having a supplementary field association relationship, and the strength value of the supplementary field association is specified. The calculation formula is:

[0108] ,

[0109] in, The larger the value, the stronger the correlation between the supplementary fields, and the stronger the ability to restore identity through the combination of two units; the subject locking probability directly represents the ability to lock a unique subject after combination, and is the core determinant of the correlation strength, assigned a core weight of 0.6; the matching range reduction ratio only represents the convergence effect of the filtering range, assigned an auxiliary weight of 0.4, and this ratio is optimized based on test data of the real name information field combination restoration scenario; the effective correlation threshold is set to an example of 0.4. A threshold of 30 points corresponds to a field combination that has basic identity screening capabilities. Associations below this value do not contribute substantially to identity recovery and can be removed as invalid associations. Invalid associations below the threshold are removed, while valid associations are retained and their attributes are solidified.

[0110] Node connection relationship identification: For real-name exposed units belonging to the same root identifier, the same single link of real-name information, and upstream and downstream related standard business nodes, unit pairs that have a traceable correspondence and can restore complete real-name information fields or fill in missing field fragments are identified as having a node connection relationship; first, the traceability matching degree is calculated. This characterizes the degree of content source matching between downstream units and upstream source units, and is calculated using the following formula:

[0111] ,

[0112] in, The content character matching length between the downstream unit and the upstream source unit is the structured field, which is the number of overlapping characters in the field value. The processed field is the length of the content fragment that can be traced and mapped. The total character length of the upstream source unit; The range of values ​​is Node connection strength value The calculation formula is:

[0113] ,

[0114] in, The higher the value, the stronger the node connection and the stronger the ability to restore identity through the combination of upstream and downstream units; the source matching degree directly determines whether upstream and downstream units can be restored to the complete original fields, and is the core judgment indicator of node connection and association, assigned a core weight of 0.7; the subject locking probability is only used as an auxiliary verification indicator of the restoration ability after combination, assigned an auxiliary weight of 0.3, which is in line with the security control logic that "field fragments flowing from the same source have the highest restoration priority" in the process of real-name information flow; the effective association threshold is set to an example of 0.7. A threshold of 20 points corresponds to upstream and downstream node field fragments having basic traceable splicing capabilities. When the value is lower than this, the fragments cannot form an effective restoration connection, invalid associations can be eliminated, invalid associations below the threshold can be eliminated, valid associations can be retained, and association attributes can be solidified.

[0115] Calling combined association identification: For real-name exposure units belonging to the same subject root identifier but belonging to different real-name information single links / standard business nodes / different business scenarios, for each unit... A 50% single-field locking probability is a common dividing line between high-risk and ordinary fields. When a single field falls below this value, it lacks independent identification capability and requires cross-scenario combination to recover risk. of or Units where the same accessing entity retrieves historical records within a preset time window are identified as having a call combination association. A 90% lock probability and a 95% convergence ratio correspond to a high-risk state where cross-scenario combinations are nearly uniquely identifiable, and are the core risk thresholds for determining call combination associations. The preset time window is 30 days as an example. The call combination association strength value... The calculation formula is:

[0116] ,

[0117] in, The larger the value, the stronger the combination association, and the stronger the ability to restore identity through cross-scene unit combination; This is a correction factor for historical access behavior. If units within a unit group are accessed sequentially by the same accessing subject within a preset time window... An example value of 1.2 is used, with a maximum correction limit of 1.5, when there is no corresponding historical access behavior. The base correction coefficient of 1.0 reflects the amplifying effect of continuous historical retrieval behavior on the recovery risk, while the upper limit of 1.5 is used to avoid over-correction leading to distortion of risk assessment, which meets the robustness requirements of risk quantification. The subject locking probability directly represents the risk level of uniquely identifying the real-name subject after cross-scenario combination, and is the core determinant of the strength of the call combination association, so it is assigned a core weight of 0.7. The matching range reduction ratio only represents the convergence effect of the screening range and cannot directly reflect the recovery risk of unique locking, so it is assigned an auxiliary weight of 0.3. This weight ratio is based on the judgment rules of personal information identifiability in the national cybersecurity level protection system and the risk assessment specifications for cross-scenario information leakage. The effective association threshold is set as follows: A threshold of 40 points corresponds to a medium to high identity restoration capability after cross-scene combination. If the score is lower than this value, there is no substantial recognition risk in cross-scene splicing, and it can be removed as an invalid association. Invalid associations below the threshold are removed, valid associations are retained, and association attributes are solidified.

[0118] For all valid associations that have been calibrated, assign a unique association identifier, solidify core attributes such as association type, unique identifier of the associated unit, association strength value, matching range reduction ratio, subject locking probability, and generation timestamp, and generate a full association database of real-name exposed units.

[0119] S205: Reconstructing the structured framework and establishing an indexing system for the reconstructed association map.

[0120] Using the real-name exposure unit library generated by S203 as the graph nodes and the full association database generated by S204 as the graph's directed weighted edges, a directed weighted restoration association graph representing the identity restoration path between real-name exposure units is constructed.

[0121] First, the graph node layer is constructed: each real-name exposed unit is treated as a unique graph node, and the node attributes are fully mapped to the standardized attribute set of the real-name exposed unit, ensuring a one-to-one correspondence between nodes and real-name exposed units. Second, the graph edge layer is constructed: each effective association is treated as a directed weighted edge, with the direction of the edge pointing from the preceding unit node to the following unit node in the restoration path, fully representing the logical flow of identity restoration. The edge weight is the association strength value of the corresponding association, and the edge attributes are fully mapped to the fixed attributes of the association. Third, the restoration path layer is constructed: the graph is constructed by... Each real-name exposed unit node passes through A directed link formed by connecting a series of directed edges satisfies the following condition: total link length... Furthermore, the total recovery probability after combining all nodes of the link is greater than or equal to the preset recovery threshold. The preset recovery threshold is 90% for example. A 90% recovery probability corresponds to a high-risk level that is close to unique identification. Paths that reach this threshold have the ability to recover the identity and need to be included in the effective recovery path management and defined as an effective real-name information recovery path.

[0122] Quantitative calculation of the total restoration probability of the restoration path, for a directed link The node sequence is The corresponding directed edge sequence is Its total recovery probability is denoted as The calculation formula is:

[0123] ,

[0124] in, The probability of locking the subject is the combination of fields composed of all nodes within the path; For the first The correlation strength value of a directed edge is divided by 100 to convert it into a weight coefficient of 0 to 1; This is the chain multiplication operator, representing the product of all edge weights; The range of values ​​is .

[0125] All valid restoration paths are mined, each path is assigned a unique identifier, and core parameters such as path node composition, edge composition, total restoration probability, path length, association type, and cross-link / cross-node attributes are solidified to form a full restoration path library. Finally, a global multi-dimensional index system is established for the restored association graph, including unit unique identifier index, main root identifier index, association type index, restoration probability interval index, and link unique identifier index, to obtain a fully structured restored association graph.

[0126] refer to Figure 2 , Figure 2 A schematic diagram of the construction of the association map for the splitting and restoration of the real-name exposure unit.

[0127] Figure 2 The diagram illustrates the process of extracting real-name information fields from the real-name information link set, then splitting and identifying them based on their identity pointing function, association supplementation function, and combined restoration function to form real-name exposure units, and further establishing a restoration association graph around three types of association relationships: field supplementation, node connection, and call combination. In the diagram, nodes represent real-name exposure units, directed edges represent restoration association relationships between different real-name exposure units, and links represent effective restoration paths.

[0128] It should be noted that, Figure 2 This is merely a structured illustration to help understand the relationship between the real-name exposed unit and the restoration path, and does not represent the specific values ​​of the actual graph size, number of nodes, and number of edges.

[0129] In step S3, based on the real-name exposure unit and the restoration correlation map, a joint analysis is performed on the current access request and historical access results to determine the cumulative exposure status. Then, a particle swarm optimization algorithm is used for iterative optimization to generate the corresponding exposure suppression strategy, including:

[0130] S301: Parsing of current access requests and collection of historical access data.

[0131] Receive the current access request and complete full-dimensional structured parsing, extracting core attributes such as the unique identifier of the access subject, the access task requirement level, the real-name information field requested, and the identifier of the associated link / node. Match it with the real-name exposure unit library generated by S203, identify all real-name exposure units involved in the current request, and generate the current request exposure unit set. , This represents the total number of real-name exposed units involved in the current request.

[0132] Using the unique identifier of the accessing entity as the search key, the system aggregates valid real-name exposed units that have been output within a preset 90-day retrospective window. This retrospective window covers a quarterly business access cycle, enabling the complete capture of continuous query behavior across months. This avoids the risk of missed detections caused by scattered queries over long periods. After removing invalid units whose lifecycles have ended, a set of historically output exposed units is generated. , The total number of historically valid real-name exposed units is used to synchronously collect the associated attributes of the corresponding access behaviors.

[0133] S302: Joint correlation analysis of current and historical exposure units.

[0134] Using the restored correlation map constructed by S205 as the sole analytical benchmark, for and Perform a full joint correlation analysis to quantitatively identify the extent to which current access behavior supplements and connects with existing exposed information:

[0135] Full set of association matching: Traverse and restore the association graph, identify the effective association edges between two sets of units and the effective restoration paths that can be formed together, extract the association type, association strength, and total path restoration probability, and generate a cross-set association matrix.

[0136] Quantitative calculation of supplementation level: Supplementation level The improvement in the current requesting unit's ability to lock onto subjects and narrow the filtering scope of historically exposed information is represented by the following formula:

[0137] ,

[0138] ,

[0139] in, To reduce the percentage increment for matching range, This reduces the matching range of the historically output exposed unit set by a certain percentage. The ratio for narrowing the matching range between the current unit and the entire set of historical units strictly follows the global calculation formula for the ratio of narrowing the matching range in S202; the weights of 0.6 and 0.4 correspond to the priority of subject locking ability and range convergence ability, respectively, which are consistent with the full text calculation system. The value ranges from 0% to 100%. The higher the value, the greater the increase in both replenishment capacity and risk.

[0140] Quantitative calculation of connection relationships: Connection relationships The integrity and risk level of the effective recovery path that can be formed by the two sets of units are characterized by the following formula:

[0141] ,

[0142] in, The total number of valid restoration paths formed by the two groups of units. The first set represents the combination of the currently requested set of exposed units and the historically output set of exposed units. An effective recovery path This represents the total recovery probability corresponding to this valid recovery path. It is used to quantify the probability level of achieving unique identity recovery after combining all real-name exposed units within this path. Its calculation logic strictly follows the global calculation formula of S205. This represents the theoretical maximum number of restoration paths. The value ranges from 0% to 100%. The higher the value, the greater the continuity and the greater the feasibility of identity restoration.

[0143] Consolidation of analysis results: Identify high-risk associated unit pairs and a list of effective recovery paths, generate standardized joint analysis results, and provide quantitative basis for determining cumulative exposure status.

[0144] S303: Quantitative calculation and level determination of cumulative exposure status.

[0145] Based on the joint analysis results and combined with the basic risk attributes of the units, a full quantitative calculation and standardized classification of the cumulative exposure status were completed:

[0146] Cumulative exposure risk score calculation: Cumulative exposure risk score The formula for calculating the overall risk level of the information that the accessing subject can obtain after the current access is executed is as follows:

[0147] ,

[0148] ,

[0149] ,

[0150] ,

[0151] in, The average baseline exposure risk score, For a single real-name exposed unit The basic exposure risk score strictly follows the calculation formula of S203; This refers to the complete set of the current unit and all historical units. The probability of locking the entire set of entities; The weighting is set based on relevant regulations on personal information protection and practical experience in managing real-name registration services, representing the maximum risk increment. The value ranges from 0 to 100; the higher the value, the higher the cumulative exposure risk.

[0152] Cumulative Exposure Status Classification: Extremely High Risk Cumulative Exposure Status: or High-risk cumulative exposure status: or Cumulative exposure status at medium risk: or Low-risk cumulative exposure status: and This four-level classification standard aligns with the data security classification protection requirements. A subject locking probability of 95 corresponds to an extremely high risk that can be uniquely identified, while a probability below 30 corresponds to a low-risk state with no substantial identification risk. The middle range is evenly divided according to the risk gradient.

[0153] Cumulative exposure status solidification: Generate a standardized cumulative exposure status data package containing risk level, cumulative exposure risk score, probability of locking all subjects, and a list of high-risk recovery paths.

[0154] S304: Construction of the strategy parameter space under multiple constraints.

[0155] Using cumulative exposure status as the core constraint and combining it with boundary conditions across four dimensions, a standardized parameter space for the exposure suppression strategy is constructed:

[0156] Constraint Boundary Setting: Four key hard constraint boundaries are clearly defined: Cumulative Exposure Status Constraint: The higher the risk level, the stricter the control constraints; Access Task Requirement Constraint: Based on four levels—legal verification, business processing, statistical query, and auxiliary support—minimum task satisfaction thresholds are set at 95%, 85%, 70%, and 50%, respectively. This threshold system matches the legal requirements and functional dependencies of various business types. Legal verification requires ensuring the complete output of core fields, while auxiliary support has the lowest requirement for field completeness, with corresponding thresholds decreasing progressively; Output Usability Scope Constraint: Clearly define the lists of prohibited, restricted, and permitted output units. Identity-oriented high-risk units are included in the restricted output list by default; Recovery Risk Constraint: Set a security red line where the maximum allowed subject locking probability after policy execution is ≤50%. A 50% subject locking probability is a general security dividing line for personal information identifiability. Below this value, it is impossible to accurately locate a specific subject through output information, which meets the basic data security control objectives.

[0157] Strategy Parameter Dimension Definition: The four optimization parameters completely correspond to the core parameters of the original invention, defining a standardized value space:

[0158] Desensitization particle size Discrete integers 1-5, corresponding to five control levels: no desensitization, mild desensitization, moderate desensitization, severe desensitization, and complete blocking.

[0159] Field Limitation Range A continuous value of 0-1 represents the proportion of the number of allowed output units to the total number of currently requested units; the closer to 0, the stricter the restriction.

[0160] Path blocking strength Discrete integers 1-4, corresponding to four levels: no blockage, mild blockage, moderate blockage, and severe blockage.

[0161] Approval Triggering Conditions The value is 0-100, representing the cumulative exposure risk threshold that triggers approval. The lower the value, the stricter the control.

[0162] Compliance parameter space generation: Based on constraint boundaries, invalid parameter values ​​are pruned to form a four-dimensional standardized strategy parameter space. .

[0163] S305: Multi-parameter joint optimization based on particle swarm optimization algorithm.

[0164] With strategy parameter space To determine the feasible solution domain, a particle swarm optimization algorithm is used to find the globally optimal parameter combination that balances task satisfaction, risk mitigation, and approval efficiency.

[0165] Algorithm initialization: Map each set of parameters to a particle position vector. The initial population size is set to 30, the maximum number of iterations to 100, the inertia weight to 0.7, and the individual / global learning factor to 1.5. An initial population is generated within the feasible solution domain, and the optimal positions of individuals are initialized simultaneously. with the global optimal position .

[0166] Constructing a multi-objective fitness function The fitness value is used to evaluate the quality of the parameter combination corresponding to each particle position. A higher fitness value indicates a better parameter combination. The calculation formula is:

[0167] ,

[0168] in, The task satisfaction score is calculated as the proportion of valid fields that can be output after the parameter combination is executed, relative to the core fields required by the task. The risk suppression rate score is calculated by dividing the cumulative subject lock probability (combined with the historical output content after the exposure suppression strategy is implemented) by the cumulative subject lock probability (combined with the current request content and historical output content before the strategy is implemented), subtracting this ratio from 1, and then multiplying by 100 to obtain a standardized quantitative score in the range of 0 to 100. The approval efficiency score is calculated based on the approval trigger probability corresponding to the parameter combination. The lower the approval trigger probability, the higher the efficiency score. The values ​​range from 0 to 100. The risk level is dynamically adjusted based on cumulative exposure risk level. The adjustment rules correspond exactly to the cumulative exposure status level classification in S303. The higher the risk, the greater the risk suppression weight. The specific classification configuration is as follows: Under the extremely high risk cumulative exposure status, in order to prioritize the risk suppression effect of identity restoration, it is set to an example of... , , Under conditions of high-risk cumulative exposure, in order to balance the suppression of core risks and the satisfaction of basic business needs, an exemplary setting is provided. , , To balance business needs adaptation, risk control and mitigation, and approval efficiency under medium-risk cumulative exposure conditions, an example is set as follows: , , In a low-risk, cumulatively exposed state, priority should be given to ensuring the efficiency and ease of approval for legitimate business access. An example setting is... , , .

[0169] The particle velocity and position are iteratively updated according to the standard particle swarm optimization rules. Each iteration verifies parameter boundaries and updates the individual / global optimal position. The maximum number of iterations is reached, or the global optimal fitness changes by less than [a certain value] for 10 consecutive generations. The iteration terminates when the optimal position is reached, and the optimal parameter combination corresponding to the globally optimal position is output.

[0170] S306: Generation and effectiveness verification of exposure suppression strategies.

[0171] The optimal parameter combination is converted into control rules that match the access scenario, forming an initial exposure suppression strategy that includes four modules: de-identification execution rules, field output control rules, restoration path collaborative blocking rules, and approval trigger execution rules. By binding the initial exposure suppression strategy to the corresponding access request, subject identifier, and cumulative exposure status data packet, a standardized exposure suppression strategy is finally generated and synchronously stored in the operation audit log, providing a unique execution basis for subsequent link output control.

[0172] In step S4, based on the exposure suppression strategy, link output control is performed on the current access request, and coordinated blocking is implemented on associated paths that show a tendency to recover, generating security control results, including:

[0173] S401: Exposure suppression strategy analysis and execution baseline initialization.

[0174] The standardized exposure suppression strategy generated by S3 is used as the sole basis for execution. The execution baseline is fully initialized by combining the current access request, the real-name exposure unit library, and the restored association graph.

[0175] First, a comprehensive structured analysis of the exposure suppression strategy is performed, breaking down and solidifying four core execution rules: de-identification execution rules, field output control rules, restoration path collaborative blocking rules, and approval-triggered execution rules. The optimal de-identification granularity matching the current access request is then extracted. Optimal field limit range Optimal path blocking strength Optimal approval trigger conditions The document explicitly allows the output of lists of entities with exposed real names, restricts the output of lists of entities with exposed real names, prohibits the output of lists of entities with exposed real names, and lists of entities with blocked high-risk recovery paths.

[0176] Secondly, the execution context baseline is bound, including the set of units exposed by the current request. Historically exposed unit set The data includes: unique identifier of the accessing entity, root identifier of the corresponding real-name entity, cumulative exposed status data packets, all nodes, edges and valid restoration path data related to the real-name entity in the restored association graph.

[0177] Finally, the initialization execution verification benchmark includes the minimum threshold for task satisfaction corresponding to the access task requirements, the maximum allowable recovery risk threshold preset by the policy, and the preset recovery boundary judgment threshold. The maximum allowable recovery risk threshold is 50% for example. The recovery boundary judgment threshold includes the incremental recovery risk value threshold, the cross-scenario association matching degree threshold, and the splicing risk coefficient threshold, which are 10%, 30%, and 50% for example, respectively. The three thresholds correspond to the minimum risk trigger line for incremental risk, path coverage, and splicing feasibility, respectively. 10% incremental risk is an acceptable range of safety fluctuations, and 30% matching degree and 50% splicing coefficient are the dividing values ​​of no substantial recovery risk, providing a standardized benchmark for subsequent full-process control execution.

[0178] S402: Pre-identification of access behavior restoration tendency based on restored association graph.

[0179] Based on the restoration association map data initialized by S401, and combined with the cumulative exposure status data packet, a full-quantitative identification and judgment of the restoration tendency of the current access behavior is performed to clarify the type and risk level of the restoration tendency.

[0180] The first step is to calculate the core indicators for determining the tendency to restore health.

[0181] For the current access request, three core judgment indicators are calculated, and the calculation formulas for all indicators are completely consistent with the global calculation rules for the entire text:

[0182] Incremental recovery risk value : Represents the increased risk of identity restoration after the current request is executed; the calculation formula is as follows. ;in, The subject locking probability of the historically exposed unit set. The probability of locking the current unit with the entire set of historical units;

[0183] Cross-scene correlation matching degree : Represents the effective recovery path coverage that can be formed between the current request unit and the historical output unit, and is calculated using the following formula: ,in, This represents the total number of valid recovery paths, including the current request unit. This represents the total number of valid recovery paths corresponding to this real-name entity.

[0184] splicing risk factor : Characterizes the feasibility of information splicing and restoration across fields, nodes, and interfaces; the calculation formula is as follows. ,in, To supplement the degree, For consistency, the global calculation formula of S302 is used.

[0185] The second step is to determine the type of restoration tendency and verify the boundaries.

[0186] Based on three core indicators and combined with the restoration boundary judgment threshold, a classification judgment of restoration tendency is performed:

[0187] Determination of no tendency to recover: simultaneously satisfying , , If the cumulative probability of the subject being locked after the strategy is executed does not exceed the maximum allowed recovery risk threshold, it is determined to be a normal access behavior that has not touched the preset recovery boundary;

[0188] Cross-field supplementation and restoration tendency determination: satisfies A 70% supplementation level corresponds to the current field being able to significantly improve the identity recognition capability of historical information. It already has the substantial risk of identity restoration through field complementarity. Furthermore, the current request unit and the historical output unit belong to the same standard business node but different field dimensions. An effective restoration path can be formed through field complementarity. It is determined that there is a tendency for cross-field supplementation restoration.

[0189] Cross-node connection restoration tendency determination: satisfies 70% of the connection relationships correspond to upstream and downstream node information that can form a highly complete restoration path. There is a risk of splicing complete information through flow tracing. Furthermore, the current request unit and the historical output unit belong to the same real-name information single link and upstream and downstream related standard business nodes. An effective restoration path can be formed through node flow tracing. It is determined that there is a tendency for cross-node connection restoration.

[0190] Cross-interface splicing and restoration tendency determination: satisfies A 60% cross-scenario correlation matching rate corresponds to the current access covering most effective recovery paths, which already has a high risk of cross-interface splicing to achieve identity recovery. Furthermore, the current request unit and the historical output unit belong to different real-name information single links, different business interfaces, and different business scenarios. They can form effective recovery paths through cross-scenario call combinations, which indicates a tendency for cross-interface splicing recovery.

[0191] The third step is to solidify the risk level of the restoration tendency.

[0192] For access behaviors that show a tendency to revert to previous status, the risk level of the tendency to revert to previous status is simultaneously determined based on the cumulative exposure status level: extremely high risk cumulative exposure status corresponds to extremely high risk of reverting to previous status, high risk cumulative exposure status corresponds to high risk of reverting to previous status, and medium-low risk cumulative exposure status corresponds to medium risk of reverting to previous status. A standardized reverting to previous status determination data package is generated to provide a basis for subsequent hierarchical control and handling.

[0193] S403: Link output control and desensitization execution in scenarios without restoration tendency.

[0194] For normal access behavior that S402 determines does not touch the preset restoration boundary and has no tendency to restore, strictly follow the exposure suppression strategy parsed by S401 and perform full-link output control and standardized desensitization processing.

[0195] The first step is to verify the compliance of the output fields.

[0196] Based on field output control rules and optimal field limit range The set of units exposed for the current request A second verification is performed to remove real-name exposed units from the prohibited output list. Access task requirement matching verification is performed on units in the restricted output list. Only units directly related to the core requirements of the task are retained, and finally, a list of compliant real-name exposed units to be output is locked.

[0197] The second step involves implementing tiered desensitization processing.

[0198] Based on optimal desensitization granularity For each real-name exposed unit in the compliance output list, standardized desensitization processing is performed that precisely matches the unit's function attributes and risk level. The processing rules are completely consistent with the full-text desensitization granularity definition.

[0199] when In other words, without desensitization, only prohibited output units are removed, and the original complete values ​​are output for the remaining units. This is only applicable to legal verification tasks and low-risk cumulative exposure scenarios.

[0200] when This is called mild desensitization. It outputs the original value for the combined and complex real-name exposure unit, performs masking processing on the associated supplementary unit, and retains only the minimum fragment necessary for business for the identity pointing unit. For example, the first 3 and last 4 digits of the mobile phone number are retained, and the rest of the characters are replaced with *.

[0201] when This is called moderate desensitization. Irreversible masking is performed on all identity-pointing units, only legally necessary fragments such as address code and date of birth code are retained, synonym generalization replacement is performed on related supplementary units, such as generalizing the detailed address to the city of origin, and non-core fragment masking is performed on combined complex units.

[0202] when This is called heavy desensitization, which performs irreversible salted hashing or synonymous anonymization replacement on all real-name exposed units. The output results have no direct identity reference and only retain statistical dimension attributes that can satisfy statistical query tasks.

[0203] when This means completely blocking the output of any real-name exposed unit content and only returning a business status code without identity attributes.

[0204] The third step is to conduct a second verification of the compliance of the de-identification results.

[0205] For the output content that has completed the de-identification process, three mandatory checks are performed: First, check whether the de-identified content meets the field restriction range requirements and whether there is any output content that exceeds the range; second, check whether the cumulative subject locking probability after merging the de-identified content with the historical output content does not exceed the maximum allowed restoration risk threshold; third, check whether the task satisfaction is not lower than the minimum threshold corresponding to the access task requirements; after all checks pass, a preliminary de-identification output result is generated.

[0206] S404: Collaborative blocking and tiered management and control of scenarios with a tendency to relapse.

[0207] For access behaviors identified by S402 as having a tendency to revert, the exposure suppression strategy analyzed by S401 is strictly followed. The associated paths are blocked in conjunction with the reversion association map. Furthermore, based on the type of reversion tendency and the risk level, key real-name exposed units are subject to graded control and disposal.

[0208] The first step is to identify key real-name exposed units and core associated paths.

[0209] Based on the restored correlation map and the data package for determining restoration tendency, the target for further action is identified.

[0210] Definition of key real-name exposure units: In the restoration association graph, the core real-name exposure units that rank in the top 20% of the association strength value within the corresponding effective restoration path and contribute more than 30% to the total restoration probability of the path. The top 20% of high association strength units cover more than 80% of the restoration contribution in the path. The units corresponding to the 30% contribution threshold have an irreplaceable core role in the path, are key targets for blocking and control, and are core nodes that rely on identity restoration.

[0211] Core related path locking: combining optimal path blocking strength Identify a list of high-risk recovery paths that require coordinated intervention: Only lock in extremely high-risk paths with a total recovery probability ≥ 95%; Lock in extremely high-risk paths with a total recovery probability ≥ 90%; Lock out high-risk or higher paths with a total recovery probability of ≥70%; All effective recovery paths are locked down. This tiered blocking rule corresponds one-to-one with the risk level. The higher the blocking strength, the wider the range of risk covered. It can accurately match the blocking intensity according to the current risk level and avoid over-control.

[0212] The second step involves the coordinated blocking of high-risk recovery paths.

[0213] Based on the locked core association path list, combined with the global multi-dimensional index system of the restored association graph, the whole-link collaborative blocking is performed: for the core association edges within the path, temporary access permission interception, field call restriction, and flow channel circuit breaking are performed on the corresponding standard business nodes, service interfaces, and data transmission links to completely cut off the information splicing and restoration path across fields, nodes, and interfaces. After blocking, the total restoration probability of the path is synchronously verified to ensure that the total restoration probability of the path drops below the preset restoration threshold after blocking.

[0214] The third step is to classify and handle key units that have been exposed for real-name authentication.

[0215] Combining the risk level of restoration tendency and the optimal approval trigger conditions For the key exposed units with locked identities, four standardized procedures are implemented, with the procedures completely corresponding to those in the original invention:

[0216] Output restriction processing: Applicable to medium-risk recovery scenarios, it performs content truncation and desensitization granularity upgrade processing on key real-name exposed units that exceed the scope of business necessity, retaining only the smallest information fragments that meet the minimum requirements of the task, and prohibiting the output of complete field content;

[0217] Replacement output processing: Applicable to medium-to-high risk recovery scenarios, it performs irreversible anonymization replacement, pseudo-value filling, and generalization processing on key real-name exposed units, such as replacing the exact date of birth with the range of the year of birth, and replacing the detailed address with the administrative region to which it belongs. The output content can meet the business statistics requirements, but it has no identity recovery capability.

[0218] Circuit breaker output handling: Applicable to high-risk and extremely high-risk recovery scenarios, it completely blocks the output of key real-name exposed units, does not return any content related to the unit, and only returns a business compliance interception prompt, completely eliminating the unit's contribution to identity recovery;

[0219] Additional approval processing: When the cumulative exposure risk score of the access behavior reaches the optimal approval trigger condition. At any time, regardless of the restoration tendency level, the output process of the current access request will be immediately suspended, and the corresponding level of additional approval process will be automatically triggered. The access subject information, cumulative exposure status, restoration tendency risk, and high-risk restoration path list will be pushed to the approval subject. Only after the approval is approved can the corresponding output processing be executed according to the approval opinion. If the approval is rejected, the request will be terminated directly and the interception result will be returned.

[0220] The fourth step is to verify the risk of the disposal results.

[0221] For output content after collaborative blocking and tiered handling, a mandatory risk verification is performed: the cumulative subject locking probability after merging the post-handling content with historical output content must be reduced to below the maximum allowable recovery risk threshold, and all high-risk recovery paths must have been effectively blocked. If the verification fails, the control level will be immediately upgraded and circuit breaker output handling will be implemented.

[0222] S405: End-to-end operation audit and validity verification of execution results.

[0223] All execution operations and output results generated by S403 and S404 are subject to full-link auditing and multi-dimensional validity verification to form a closed-loop control.

[0224] The entire process of this access request is executed, generating standardized audit logs. The audit content includes: unique identifier of the access subject, identifier of the operation terminal, access timestamp, real-name information field requested, cumulative exposure status, restoration tendency judgment result, exposure suppression strategy execution details, de-identification rules, list of collaborative blocking paths, key unit handling methods, approval process records, final output content, and risk verification results. The audit logs are synchronously written to the access data audit database of S1, and a unique hash value is used for evidence storage, which is tamper-proof and deletable.

[0225] For the final output, perform three closed-loop validity checks:

[0226] Policy compliance verification: Verify that all executed operations fully comply with the rules of the exposure suppression policy, with no output content exceeding the scope or permissions, and that the desensitization granularity, field restriction range, path blocking strength, and approval triggering rules are all completely consistent with the optimal parameter combination;

[0227] Risk suppression effectiveness verification: Verify that the cumulative subject lock probability after merging the final output content with the historical output content does not exceed the preset maximum allowable restoration risk threshold, all high-risk restoration tendencies have been effectively controlled, and there is no compliance risk of identity restoration;

[0228] Business requirement compatibility verification: Verify the task satisfaction of the final output content, ensuring it is not lower than the minimum threshold corresponding to the access task requirements, and guarantee the normal operation of legal and compliant business while meeting security control requirements.

[0229] For results that fail verification, immediately intercept the output, upgrade the control policy and take action until all verifications pass.

[0230] S406: Generation of security control results and closed-loop update of full-link data.

[0231] Based on the execution and verification results of the entire process, standardized security control results are generated, and closed-loop updates of relevant data across the entire chain are completed to achieve dynamic iteration of the control system.

[0232] Based on the verified execution results, two types of standardized security control results are generated: Compliance de-identification results: For access behaviors that do not touch the preset restoration boundary and have no tendency to restore, the de-identification processing results generated by S403 and verified are output, along with a description of the de-identification rules and a compliance verification report.

[0233] Security control and handling results: For access behaviors that show a tendency to recover, output the corresponding hierarchical control and handling results, including details of collaborative blocking execution, handling methods of key units, approval process results, risk interception explanation, and if the approval is approved, output the corresponding compliance output content of the approval opinion at the same time. If the approval is rejected or the circuit breaker is handled, output the compliance interception notification at the same time.

[0234] Based on the execution result of this access behavior, the relevant data system is dynamically updated to achieve a closed-loop management system:

[0235] Historical access data update: The real-name exposure unit of the final output will be synchronously updated to the historical output exposure unit set of the corresponding access subject. Update access behavior audit logs to provide the latest data foundation for joint analysis of subsequent access requests;

[0236] Dynamic updates to the restored association graph: Based on the call records of this access behavior, update the historical access behavior correction coefficients of the corresponding call combination associations in the restored association graph. Recalculate the association strength value of the corresponding associated edges and the total restoration probability of the restoration path to complete the dynamic iteration of the graph and ensure that the graph always accurately represents the latest restoration path relationship.

[0237] Strategy optimization sample update: The scenario characteristics, cumulative exposure status, strategy execution effect, and risk verification results of this access request are stored in the strategy optimization sample library. This provides real business scenario samples for subsequent iterations of the particle swarm optimization algorithm's optimization model, continuously improving the scenario adaptability and risk control effectiveness of the exposure suppression strategy.

[0238] refer to Figure 3 , Figure 3 This is a flowchart for the execution of safety control measures.

[0239] This embodiment constructs a real-name information link set by collecting, associating, labeling, and temporally organizing the end-to-end data in the real-name business processing. Based on this, it performs exposure unit decomposition on the real-name information and establishes a restoration association graph. Combining current access requests and historical access results, it determines the cumulative exposure status. Then, it generates an exposure suppression strategy using a particle swarm optimization algorithm. Finally, it performs tiered desensitization, restoration tendency identification, collaborative blocking of associated paths, and approval processing on the output content. This solution can balance business processing needs and information security requirements, reduce the risk of real-name information being spliced ​​and restored across scenarios, and improve the dynamic desensitization capability, end-to-end auditing capability, and refined security control level in real-name business scenarios.

[0240] Example 2

[0241] This invention discloses a full-link de-identification and security management system for real-name information, including a link construction module, a restoration and association module, a policy generation module, and a management execution module:

[0242] The link construction module collects source data, flow data, and access data during the real-name business processing, performs subject association, node identification, and time sequence processing on various types of data, and constructs a real-name information link set.

[0243] The restoration association module is based on the real-name information link set. It performs splitting and identification processing on the real-name information involved in the link according to the identity pointing role, association supplementation role and combined restoration role to obtain the corresponding real-name exposure unit. Based on the association relationship of each real-name exposure unit in field supplementation, node connection and call combination, a restoration association map is established to represent the restoration path between different real-name exposure units.

[0244] The strategy generation module, based on real-name exposure units and restoration association graphs, performs joint analysis on the real-name exposure units involved in the current access request and those already output in historical access results. It identifies the degree of supplementation and connection between the current access behavior and existing exposure information, and determines the corresponding cumulative exposure status. Based on the cumulative exposure status, access task requirements, output availability range, and restoration risk constraints, a strategy parameter space is constructed. The particle swarm optimization algorithm is used to jointly optimize the desensitization granularity, field restriction range, path blocking strength, and approval triggering conditions to generate an exposure suppression strategy that matches the current access scenario.

[0245] Based on the exposure suppression strategy, the control and execution module performs link output control on the output content corresponding to the current access request, and combines the restoration association graph to implement collaborative blocking of association paths with restoration tendencies. For access behaviors that do not reach the preset restoration boundary, the module outputs the de-identification results corresponding to the access task. For access behaviors with restoration tendencies that form cross-field supplementation, cross-node connection or cross-interface splicing with historical access results, the module performs output restriction, output replacement, circuit breaker output or additional approval processing on key real-name exposure units in the association path, and generates the corresponding security control results.

[0246] refer to Figure 4 , Figure 4 This is a schematic diagram of the structure of a real-name information end-to-end de-identification and security management system.

[0247] Figure 4 The collaborative relationship between the link construction module, the restoration association module, the policy generation module, and the control execution module is shown. The link construction module is used to form a set of real-name information links, the restoration association module is used to form a restoration association graph, the policy generation module is used to determine the cumulative exposure status and generate an exposure suppression policy by combining the current access request and the historical access results, and the control execution module is used to complete the output control, collaborative blocking, and security control result generation based on the exposure suppression policy.

[0248] It should be noted that, Figure 4 The division of the modules in this invention is only for illustrating the functional composition relationship of the invention and does not constitute a limitation on the physical deployment of the modules.

[0249] The specific functions of each module described above are implemented with reference to the relevant content in the full-link desensitization and security management method for real-name information described in Example 1, and will not be repeated here.

[0250] The above description is merely a preferred embodiment of the present invention. 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 those skilled in the art, 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 method for end-to-end de-identification and security control of real-name information, characterized in that, Includes the following steps: Collect source data, flow data and access data during the real-name business processing, perform entity association, node identification and time sequence processing of various data, and construct a real-name information link set; Based on the real-name information link set, the real-name information is split into real-name exposure units, and a restored association graph is established based on the relationship between each real-name exposure unit. Based on real-name exposure units and restored correlation graphs, joint analysis is performed on current access requests and historical access results to determine the cumulative exposure status. Particle swarm optimization algorithm is used for iterative optimization to generate corresponding exposure suppression strategies. Based on the exposure suppression strategy, link output control is performed on the current access request, and collaborative blocking is implemented on associated paths that have a tendency to recover, generating security control results.

2. The method for end-to-end de-identification and security management of real-name information as described in claim 1, characterized in that, The real-name exposure unit splitting process includes: For all real-name information fields in the real-name information link set, using all real-name subjects as a unified calculation benchmark, calculate the matching range reduction ratio and subject locking probability of each field in individual and combined states. Based on the quantitative evaluation results of the narrowing of the matching range and the probability of locking the subject, the real name information field is labeled as the corresponding main function attribute label in the identity pointing function, association supplementation function or combined restoration function. For fields that have multiple function attributes at the same time, the main function attribute label is determined according to the preset priority. The real-name information field is split according to the standardized splitting rules corresponding to each main function attribute. The basic exposure risk score of each splitting unit is calculated based on the main function attribute label and the node type to which it belongs, and a real-name exposure unit library is generated.

3. The method for end-to-end de-identification and security management of real-name information as described in claim 2, characterized in that, The establishment of the reconstructed correlation map includes: For all real-name exposure units in the real-name exposure unit library, relationship mining is performed in three scenarios: field supplementation, node connection, and call combination. Based on the quantitative formulas corresponding to each scenario, the relationship strength values ​​of field supplementation, node connection, and call combination are calculated. Invalid relationships that do not reach the effective relationship threshold are removed, and a full relationship library of real-name exposure units is generated. Using real-name exposed units in the real-name exposed unit library as graph nodes and valid associations in the full association database as directed weighted edges, the direction of the edges and the corresponding association strength value are assigned according to the logical direction of the restoration path to construct a directed weighted restoration association graph. In the directed weighted reconstruction association graph, directed links that meet the reconstruction probability threshold are defined as valid reconstruction paths. All valid reconstruction paths are mined and their attributes are solidified. A global multidimensional index system is established to obtain a fully structured reconstruction association graph.

4. The method for end-to-end de-identification and security management of real-name information as described in claim 1, characterized in that, Iterative optimization using the particle swarm optimization algorithm includes: Using cumulative exposure status, access task requirements, output availability range, and recovery risk constraints as boundary conditions, a four-dimensional strategy parameter space is constructed, which includes desensitization granularity, field restriction range, path blocking strength, and approval triggering conditions. The parameter combinations in the strategy parameter space are mapped to particle position vectors, and a multi-objective fitness function is constructed with task satisfaction score, risk inhibition rate score and approval efficiency score as objective terms. The weights of each objective in the multi-objective fitness function are dynamically adjusted based on the cumulative exposure level. Particle velocity and position are iteratively updated according to the standard particle swarm rules. The globally optimal parameter combination is output to generate an exposure suppression strategy.

5. The method for end-to-end de-identification and security management of real-name information as described in claim 1, characterized in that, Perform joint analysis on current access requests and historical access results, including: Lock all real-name exposed units involved in the current access request and generate a set of exposed units for the current request; collect the valid real-name exposed units that have been output in the past within the preset backtracking window of the access subject and generate a set of historically output exposed units; Based on the restored association graph, a full association analysis is performed on the current request exposure unit set and the historical output exposure unit set to calculate the degree of supplementation and connection between the current access behavior and the historical exposure information. Based on the degree of supplementation, connection relationship and basic exposure risk score of each real-name exposure unit, the cumulative exposure risk score is calculated, the cumulative exposure status level is determined according to the preset level classification rules, and a standardized cumulative exposure status data packet is generated.

6. The method for end-to-end de-identification and security management of real-name information as described in claim 1, characterized in that, Implement coordinated blocking of association pathways showing a tendency to relapse, including: For the current access behavior, calculate the incremental restoration risk value, cross-scenario association matching degree and splicing risk coefficient, compare the three core quantitative indicators with the restoration boundary judgment threshold, determine whether the current access behavior has a tendency to supplement restoration across fields, a tendency to connect restoration across nodes or a tendency to splice restoration across interfaces, and mark the corresponding restoration tendency risk level. Based on the path blocking strength, the range of high-risk restoration paths that need to be blocked is identified in the restoration association graph. Temporary access permission interception, field call restriction and flow channel circuit breaking are performed on the standard business nodes, service interfaces and data transmission links corresponding to the core association edges within the path. After the blocking is completed, the total path recovery probability is verified. Once it is confirmed that the probability has dropped below the preset recovery threshold, the full-link collaborative blocking is completed and the blocking record is fixed.

7. The method for end-to-end de-identification and security management of real-name information as described in claim 1, characterized in that, Perform link output control on the current access request, including: For access behaviors that do not reach the preset restoration boundary, based on the desensitization granularity in the exposure suppression strategy, the corresponding processing of no desensitization, mild desensitization, moderate desensitization, severe desensitization or complete blocking is performed on the real-name exposed unit to be output, and a preliminary desensitization output result is generated; For access behaviors that show a tendency to revert to previous behavior, based on the risk level of the tendency to revert to previous behavior and the approval triggering conditions, restrict output, replace output, circuit breaker output or add approval processing for key real-name exposed units in the associated path. Perform policy compliance verification, restoration risk verification, and business requirement adaptability verification on all output results. If the verification fails, the control level is upgraded and the handling is re-executed until all verifications pass and a security control result is generated.

8. The method for end-to-end de-identification and security management of real-name information as described in claim 1, characterized in that, The construction of the real-name information link set includes: By combining bypass traffic mirroring collection, system log embedding collection, full audit collection of interface calls, and storage node snapshot collection, we perform full real-time data capture of source data, flow data, and access data, and perform unique hash value storage for each captured data. Data cleaning, format normalization, and metadata standardization extraction are performed on all data. The unique root identifier of the real-name subject is used as the binding benchmark. The association processing of the real-name subject is performed on all data subjects through two methods: direct matching and chain association matching. The node attribution of all data is determined according to the standard business node classification system. The time series of all data is organized based on the business life cycle time series baseline, and a real-name information link set with the real-name information life cycle flow trajectory as the core content is constructed.

9. The method for end-to-end de-identification and security management of real-name information as described in claim 1, characterized in that, After generating the security control results, it also includes: A standardized operation audit log is generated for the entire process. The operation audit log includes the unique identifier of the access subject, the cumulative exposure status, the result of the recovery tendency judgment, the details of the exposure suppression strategy execution, and the final output content. A unique hash value is used for evidence storage to ensure that it cannot be tampered with. The real-name exposure unit of the final output is synchronously updated to the historical output exposure unit set of the corresponding access subject. Based on the current call record, the historical access behavior correction coefficient of the corresponding relationship in the restoration association graph is updated. The association strength value and the total restoration probability of the effective restoration path are recalculated to complete the dynamic iteration of the graph. The characteristics of this access scenario, the effect of policy execution, and the results of risk verification are stored in the policy optimization sample library for continuous iteration of the particle swarm optimization algorithm's optimization model.

10. A real-name information end-to-end de-identification and security management system, used to implement the real-name information end-to-end de-identification and security management method as described in claims 1-9, characterized in that, It includes a link construction module, a restoration and association module, a policy generation module, and a control and execution module: The link construction module collects source data, flow data, and access data during the real-name business processing, performs entity association, node identification, and time sequence processing on various types of data, and constructs a real-name information link set. The restoration association module, based on the real-name information link set, performs real-name exposure unit splitting on the real-name information in it, and establishes a restoration association graph based on the association relationship between each real-name exposure unit. The strategy generation module, based on real-name exposure units and restored correlation graphs, performs joint analysis on current access requests and historical access results to determine the cumulative exposure status and uses particle swarm optimization algorithm for iterative optimization to generate corresponding exposure suppression strategies. The control and execution module, based on the exposure and suppression strategy, performs link output control on the current access request, and at the same time implements coordinated blocking on related paths that have a tendency to recover, generating security control results.