Sensitive data protection system for data circulation and transactions on big data platforms
By establishing object delivery and ranking relationships at the edge nodes of the big data platform, and utilizing dynamic transfer reduction and bidirectional tracking technologies, the problem of leakage of comparison relationships for undelivered sensitive data in the ranking results is solved, thus achieving security protection and consistency of the ranking results.
Patent Information
- Application Number
- CN202611237079.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-14
- Publication Date
- 2026-11-17
AI Technical Summary
In existing technologies within big data platforms, the comparative relationships formed by sensitive data objects in the ranking results are easily leaked before delivery, leading to the leakage of unauthorized information. This is especially true when the purchasing party repeatedly adjusts the screening criteria or delivery quotas, allowing direct observation of the score relationships or score ranges of undelivered suppliers.
By solidifying object delivery permissions and priority relationship disclosure permissions at edge nodes, dynamic transfer reduction and bidirectional tracking are used to identify unauthorized comparison paths. Combined with alternating projection and output verification, the leakage of comparison relationships of undelivered sensitive data objects is suppressed, and the actual sorting order of authorized objects is maintained.
While maintaining the authenticity of the sorting results, it effectively prevents the leakage of comparison relationships of undelivered sensitive data objects, reduces the risk of association inferences formed by multiple sortings, minimizes disturbance to authorized order relationships, and improves the order consistency of protection results.
Smart Images

Figure CN123116318A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of data security and edge computing technology, and more specifically, to a sensitive data protection system for data circulation and transactions on big data platforms. Background Technology
[0002] In the process of data circulation and transaction, existing big data platforms mostly control the access of transaction participants to unauthorized data content through identity authentication, access permission verification, sensitive field masking, and result record filtering. For ranking-type transaction tasks, edge nodes generally first use the evaluation data of candidate objects to complete the ranking, and then extract the object records within the authorized scope from the ranking results for delivery. Taking supplier evaluation data transactions as an example, edge nodes rank candidate suppliers based on fulfillment rate, quality score and delivery stability. The purchaser only obtains the supplier list within the agreed quota. At the same time, it is required that the original scores be retained in the edge nodes, unauthorized supplier information not be output, and the delivery list still reflect the actual evaluation order. The existing processing method only verifies whether the final returned supplier record is within the authorized scope, and does not restrict the comparison relationship formed by non-returning suppliers participating in the ranking. When the purchaser adjusts the candidate scope, screening conditions or delivery quotas multiple times, the entry, exit and ranking changes of the same supplier in each result can be directly observed, thereby gradually determining the score relationship or score range between unauthorized suppliers and delivered suppliers. The technical problem this application aims to solve is: how to maintain the authenticity of the sorting results at edge nodes while preventing the leakage of comparison relationships formed by the participation of undelivered sensitive data objects in the sorting process. Summary of the Invention
[0003] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a sensitive data protection system for data circulation and transactions in a big data platform. By solidifying object delivery permissions and priority relationship disclosure permissions at edge nodes, a priority graph is constructed by combining the current and historical ranking relationships. Dynamic transfer reduction and bidirectional tracking are used to identify unauthorized comparison paths. Alternating projection and output verification are then performed on the position order of the removed objects. This maintains the actual ranking order of authorized objects while suppressing the leakage of comparison relationships of undelivered sensitive data objects, thus solving the problems mentioned in the background art.
[0004] To achieve the above objectives, the present invention provides the following technical solution: a sensitive data protection system for data circulation and transactions on a big data platform, comprising: The edge resource module includes edge computing nodes, controlled storage, transaction interfaces, and resource schedulers. It receives requests and reads sensitive data, generates object weights, relational weights, and transaction keys according to the recipient and sorting fields, and allocates tasks according to input pages, result pages, and idle time slots to form candidate sets, permission tables, and task tables. The sorting and reduction module is used to sort the candidate set according to the task table, write adjacent objects as order edges, take objects within the delivery number to form the deliverable result, merge the current and historical order edges with the same transaction key, and use a dynamic transfer reduction algorithm to delete order edges with alternative paths to form a reduced graph. The access determination module is used to trace bidirectionally from the object to be delivered along the simplified graph, write the object to be delivered with zero weight and the unauthorized endpoint of the first relation with zero weight priority edge into the removal table, delete the removed object and the connected edge, and form the retained graph; The positional projection module is used to delete objects removed from the candidate set, form positional vectors according to their order, generate topological order and positional boundaries from the retained graph, and use the Dykstra alternating projection algorithm to perform sequential projection, interval projection and residual write-back until the object block and the boundary fixed object are consistent in two adjacent rounds. Finally, the protection result is formed according to the positional value and topological order. The result output module includes a release controller, a security interface, and a record storage. It reconstructs the protection result sequence edge and reduces it again, performs bidirectional tracking, and sends the protection result and stores the reduced graph and transaction key when the out-of-table is empty and the number of protection results equals the number of delivery bits. When the out-of-table is not empty, it fills in the unauthorized objects in the unauthorized path and reprojects them. It stops sending when the number of protection results is lower than the number of delivery bits.
[0005] In a preferred embodiment, the edge resource module includes: After receiving a data transaction request at the transaction interface, the controlled storage is searched by object scope. The sensitive data records that are hit are written into the candidate set in ascending order of object identifier, and the receiver identifier, sorting field version and data version are concatenated to form a transaction key. Controlled storage refers to a storage device located on the data source side, addressed by storage page, and accessible only to edge computing nodes according to the task table. It is used to store sensitive data records, access control rules, candidate sets, and permission tables, and prohibits the transaction interface from directly reading or outputting the sensitive data records therein. Sensitive data records refer to data units that participate in data transactions and contain access-controlled content. Each sensitive data record must be bound to at least an object identifier, a sorting field value, and a data version, and the object identifier is used as the basis for locating records in the candidate set. An object identifier is a fixed-length code that uniquely corresponds to a sensitive data record within the same data version. It is used to retrieve sensitive data records, rank candidate sets, and form non-repeating object pairs. The sorting field version refers to the version code used to indicate the field, sorting direction, and same-value sorting rules used in this sorting. The data version refers to the version code used to indicate the batch or snapshot of sensitive data records read in this transaction. The sorting field version limits the sorting rules, and the data version limits the data content participating in the sorting. For each object in the candidate set, the receiver identifier and object identifier are concatenated to retrieve access control rules. If an allowed delivery item is matched, the object weight is written to 1; otherwise, the object weight is written to 0. Then, non-repeating object pairs are formed by sorting the object identifiers in ascending order. The receiver identifier, sorting field version, and object pair are concatenated to retrieve access control rules. If an allowed disclosure item is matched, the relationship weight is written to 1; otherwise, the relationship weight is written to 0, thus forming the permission table. The access control rules refer to the set of permission records established according to the recipient identifier. The object permission records use the recipient identifier and the object identifier as search terms, while the relationship permission records use the recipient identifier, the sorting field version, and the object pair as search terms. These rules are used to determine whether sensitive data records can be delivered and whether the order relationship of object pairs can be disclosed. Allowed deliverables refer to object permission records in access control rules that explicitly state that the recipient has the right to obtain the corresponding sensitive data records; allowed disclosures refer to object permission records in access control rules that explicitly state that the recipient has the right to obtain the relationship permission records of the corresponding object under the priority relationship of the specified sorting field version. Based on the candidate set input page and the permission table result page, a task is generated. The sum of the number of object logs and the number of records in the permission table is used as the task time slot number. From the edge computing nodes that can access the input page, have no less than the number of free pages and no less than the number of free time slots, they are sorted in ascending order according to the number of remaining time slots after the task is executed. If the order is the same, the first node is selected in ascending order of node number. The input page, result page and task time slot number are written into the task table, so that sensitive data records are kept in the controlled memory. The candidate set input page refers to the storage page in the controlled memory that stores the candidate set and is available for edge computing nodes to read. The permission table result page refers to the storage page in the controlled memory that is reserved for writing object permissions and relation permissions. Both are written to the task table with page numbers and the task access scope is limited by the page number. Free pages refer to storage pages that are not occupied by other tasks, are not reserved, and can be written to the permission table during the execution of the current task. The number of free pages is determined according to the number of unallocated storage pages of the edge computing node. The number of idle time slots refers to the number of execution time slots that have not yet been allocated to edge computing nodes in the current scheduling cycle. Each execution time slot is used to complete one object pair generation or one access control rule retrieval and corresponding permission writing. The resource scheduler obtains the number of idle time slots by subtracting the total number of node time slots from the total number of allocated time slots.
[0006] In a preferred embodiment, the sorting reduction module includes: Based on the task table, the candidate set is read in the edge computing node. The object order is determined by the sorting field value. If the sorting field values are the same, the object order is determined by the object identifier in ascending order. Adjacent objects are written as current order edges carrying transaction keys and relation weights. The objects within the delivery bit range are taken to form the deliverable result. The sorting field value refers to a comparable value formed by reading a specified field from a sensitive data record according to the sorting field version and converting it according to the data format, ascending / descending order and the order of processing for the same value as specified by the sorting field version. It is used to determine the object order of objects in the candidate set. The delivery bit length refers to the positive integer recorded in the data transaction request, which is used to determine the number of objects entering the delivery result starting from the first object in the object sequence; Update the historical order edge that matches the transaction key with the current relationship weight in the permission table, then merge the current order edge with the updated historical order edge, check the order edge direction according to the object order, delete the order edge whose direction is opposite to the object order, and form the original relationship graph.
[0007] In a preferred embodiment, the sorting reduction module further includes: For each sequential edge in the original graph, the current sequential edge is temporarily deleted according to the dynamic transitive reduction algorithm, and an alternative path from the start point to the end point is searched. When the relation weight is one, the deletion result is retained only when the relation weights of all sequential edges in the alternative path are one. When the relation weight is zero, the deletion result is retained only when there is a sequential edge with a relation weight of zero in the alternative path. In other cases, the current sequential edge is restored until no sequential edge is deleted in one round of processing, forming a candidate reduced graph. The process of executing the dynamic transitive reduction algorithm is as follows: traverse the sequential edges in the original graph in ascending order of edge number, temporarily delete the current sequential edge, and then search for the alternative path from the start point to the end point of the current sequential edge. Based on the relational weight between the current sequential edge and the alternative path, decide whether to retain the deletion result or restore the current sequential edge. After completing one round of traversal, if a sequential edge has been deleted in this round, then traverse the graph again by edge number based on the deleted graph. If no sequential edge has been deleted in this round, then stop processing and output the candidate reduced graph. An alternative path is a directed path formed by connecting one or more edges other than the current edge after temporarily deleting the current edge, with the starting point and ending point being the same as the current edge. Based on object order, reverse bit set propagation is performed on the original relation graph and the candidate reduced graph respectively. All successor objects of each object are written into the full edge bit set, and successor objects reached through the first order edge with relation weight are written into the permission bit set. Then, the full edge bit set and permission bit set of the corresponding objects to be delivered in the two graphs are compared bit by bit. When there is a difference, the deleted order edge with the edge number first in the difference path is restored and propagated again until the full edge bit set and permission bit set of the two graphs are consistent bit by bit, forming the reduced graph. The process of performing reverse bit set propagation is as follows: assign bit numbers to each object according to the object order, set the full edge bit set and permitted bit set of each object in the original relation graph and the candidate reduction graph to zero, and then process the objects one by one from the last bit of the object order to the first bit; for each outgoing edge of the current object, merge the bit number of the outgoing edge endpoint and the full edge bit set of the endpoint into the full edge bit set of the current object; when the relation weight of the outgoing edge is one, merge the bit number of the outgoing edge endpoint and the permitted bit set of the endpoint into the permitted bit set of the current object, thereby recording the objects reached by the current object along any order edge and the objects reached along order edges with relation weights of one respectively; The objects to be delivered refer to those that are located in the order from the first position to the corresponding position in the delivery order and are written into the deliverable results.
[0008] In a preferred embodiment, the access determination module includes: Using the object to be delivered as the starting point for tracing, the object weight is read according to the object order. The objects to be delivered with a weight of zero are written to the removal table, and the objects to be delivered with a weight of one are written to the forward queue and the reverse queue respectively, forming a bidirectional tracing starting point set. The tracing path is expanded layer by layer from the starting point in the forward queue along the outgoing edges of the reduced graph, and from the starting point in the reverse queue along the incoming edges of the reduced graph. Edges in the same layer are read in ascending order of edge number. For objects that are visited repeatedly, only the tracing path with the edge number sequence first is retained. When an object with a weight of zero or an edge with a relation weight of zero is encountered, the current tracing path is stopped from being expanded, forming a set of overweighted paths. Outgoing edges are sequential edges in the simplified graph that start from the current object and connect to the next object in the direction from the previous object to the next object. They are used to trace the successor objects that the current object can reach along the object sequence direction. Incoming edges are sequential edges that end with the current object and connect from the previous object to the current object. They are used to trace the predecessor objects that can reach the current object in the reverse object sequence direction.
[0009] In a preferred embodiment, the access determination module further includes: For each path in the set of unauthorized paths, trace forward to find the endpoint of the first edge with a zero relation weight, and trace backward to find the starting point of the first edge with a zero relation weight. When the object weight is zero, directly take the corresponding object. Write the number of unauthorized paths that can be truncated for each object and the number of edges with a one relation weight lost after deletion into the object record. Select the first object in descending order of the number of truncated paths, ascending order of the number of lost edges, and descending order of object weight and write it into the removal table. Delete the truncated unauthorized paths until the set of unauthorized paths is empty. Remove the removed objects and their adjacent order edges from the simplified graph based on the removal table. Then, repeat the bidirectional tracing for the unremoved objects to be delivered. If no unauthorized path is formed, output the retained graph after deletion. If an unauthorized path is still formed, continue to write an object to the removal table according to the object record and trace again until no unauthorized path is formed or all objects to be delivered are written to the removal table.
[0010] In a preferred embodiment, the positional projection module includes: Based on the removal table, remove the objects to be removed from the candidate set. Subtract the number of objects to be removed from the previous position according to the order before deletion to form an initial position vector. Then perform a topological expansion on the retained graph. Each time, take the first object from the objects with zero in-degree in ascending order according to the order before deletion and the object identifier to form the topological order. Add one to the number of predecessor objects reached by each object along the retained graph to form the lower bound of the position vector. Subtract the number of successor objects from the number of retained objects to form the upper bound of the position vector. The topology expansion process is as follows: Objects in the retained graph are written to the unexpanded object set. The number of sequential edges pointing to each object from unexpanded objects is counted as the current in-degree. From the objects with a current in-degree of zero, the first object is selected in ascending order according to its pre-deletion sequence and object identifier and written to the topology sequence. Then, the current object and its outgoing edges are deleted from the unexpanded object set, and the current in-degree of each outgoing edge's endpoint is deducted. This process is repeated until the unexpanded object set is empty. When the unexpanded object set is not empty and there are no objects with a current in-degree of zero, the expansion stops and a sequence conflict record is generated. In-degree refers to the number of sequential edges in the current topology expansion round that terminate at an object and whose starting point is still within the set of unexpanded objects. It is used to determine whether the object is still subject to the sequential relationship of other unexpanded objects. A predecessor object is an object in the retained graph that has a directed path that starts from the corresponding object and reaches the current object along the order edge. The current object itself is not counted as a predecessor object. The rewriting cost is the sum of squared differences between the positional vector and the initial positional vector. The projection constraints are the increasing topological order and the positional value being between the lower and upper bounds of the positional order. After adding the order residual to the current positional vector, a single object block is established according to the topological order. When the mean of the left object block is higher than the mean of the right object block, the two blocks are merged and the mean is recalculated until the mean of each object block increases according to the topological order. Then, the positional difference before and after projection is written into the order residual to form the order projection vector. The positional value refers to the numerical position assigned to an object during the positional projection process. Objects with earlier values are arranged before objects with later values. The initial positional value is formed by the order of objects after they are deleted or moved out. Subsequent positional values are rewritten alternately by sequential projection and interval projection. The sequential residual refers to the object-by-object difference obtained by subtracting the sequential projection output position value from the input position value of the sequential projection in the same round. The sequential residual of the first round is set to zero. Before the sequential projection of the next round, the sequential residual of the previous round is added back to the current position vector bit by bit. This is used to retain the position rewriting amount that was canceled by the interval projection in the sequential constraint correction of the previous round.
[0011] In a preferred embodiment, the positional projection module further includes: For the sequential projection vector, the interval residual is added back bit by bit. The bit value below the lower bound of the bit order is written as the lower bound of the bit order, and the bit value above the upper bound of the bit order is written as the upper bound of the bit order. The bit difference before and after writing is updated as the interval residual to form the interval projection vector. Then, the interval projection vector is used as the bit order vector for the next round of sequential projection and interval projection are repeated so that the residual of the previous round participates in the bit order correction of the next round. In each round, the object block boundary, the objects written as the lower bound of the bit order, and the objects written as the upper bound of the bit order are sequentially encoded into round codes. When the round codes of two adjacent rounds are the same, the bit order value is solved based on the equality relationship of the average value of the object blocks and the fixed relationship of the boundary. The topology order and bit order boundary are re-verified. When the verification is passed, the objects are arranged in ascending order of bit order value. When the bit order values are the same, the topology order is used. The objects within the delivery bit number are taken to form the protection result. The process of solving the positional value is as follows: determine the object block and the boundary fixed object according to the round code; write the positional value specified by the boundary fixed object to the object block; calculate the arithmetic mean of the initial positional values of each object in the block for the object block without boundary fixed objects; then arrange the lower bound of the positional values of each object in the block in ascending order and take the last value as the lower bound of the block; arrange the upper bound of the positional values in ascending order and take the first value as the upper bound of the block; write the lower bound of the block when the arithmetic mean is lower than the lower bound of the block, and write the upper bound of the block when it is higher than the upper bound of the block; in other cases, retain the arithmetic mean and write the obtained block value to each object in the block; if the same object block has two different boundary fixed values, or if the block values of adjacent object blocks do not increase in topological order, continue to perform sequential projection and interval projection until the positional values of each object satisfy the object block mean equality relationship, boundary fixed relationship, topological order and positional boundary.
[0012] In a preferred embodiment, the result output module includes: Based on the object order in the protection results, the order edges are reconstructed, the reconstructed order edges are merged with the historical order edges that are consistent with the transaction key, and the order edges with alternative paths are deleted according to the dynamic transfer reduction algorithm to form a verification graph. Starting from the object in the protection result, perform bidirectional tracing along the outgoing and incoming edges of the review graph. Read the object's authority and relation authority according to the permission table. Write the object with zero authority and the unauthorized endpoint of the first relation with zero authority into the review remove table to form the review result.
[0013] In a preferred embodiment, the result output module further includes: When the review and removal table is empty and the number of objects in the protection result is equal to the number of delivery bits, the release controller writes the protection result, transaction key and candidate set version into the release record, and sends the protection result through the security interface when the release record is consistent with the transaction key and candidate set version in the record storage, and updates the historical order edge with the review graph at the same time. The release controller is a hardware control unit that connects the record storage and the security interface. It is used to receive the review results, check the number of objects, transaction keys and candidate set versions in the review remove-out table and protection results. It generates a transmission permission only when the review remove-out table is empty, the number of objects in the protection results is equal to the delivery bit, and the release record is consistent with the transaction key and candidate set version in the record storage. It blocks the security interface when any of the verification results are inconsistent. The record storage device is a non-volatile storage device addressed by transaction key, used to store historical order edges, candidate set versions, release records, and records of non-compliant results. When writing a release record, the storage location is located by the transaction key, the data range corresponding to the current protection result is verified by the candidate set version, and the historical order edge corresponding to the transaction key is replaced with a verification diagram after the security interface completes the transmission. The security interface is a data output interface controlled by the release controller. It is used to read the protection result in the release record after obtaining a transmission permission, send the protection result according to the receiving address specified in the transaction request, and return a transmission completion bit to the release controller. If a transmission permission is not obtained, the transmission content is inconsistent with the release record, or the transmission completion bit is not returned, the security interface stops outputting and does not change the historical sequence edge. When the verification table is not empty, the first unauthorized path is selected in ascending order of unauthorized path length, and then the first unauthorized object not written to the table is added to the table in ascending order of edge number. The bit order projection is then re-executed until the verification table is empty or the number of objects in the protection result is lower than the number of delivery bits. When it is lower than the number of delivery bits, the sending is stopped and a record of non-compliant result is written.
[0014] The technical effects and advantages of this invention are as follows: 1. This solution uses object weight, relationship weight, and two-way tracking to jointly constrain the disclosure of ranking relationships, thereby maintaining the actual evaluation order while relatively suppressing the leakage of comparison relationships of undelivered sensitive objects; 2. After merging the current sequential edges with the historical sequential edges, perform dynamic transitive reduction to remove substitutable relationships and retain necessary reachable relationships, thereby relatively reducing the risk of association inference formed by multiple sortings; 3. Determine the objects to be removed from the unauthorized paths based on the number of truncations and the loss of permitted edges, so that the unauthorized relationship is eliminated in a targeted manner and the disturbance to the authorized priority relationship is relatively reduced; 4. Perform Dykstra alternating projection under topological order and positional boundary constraints to re-converge the positional order after object deletion, thereby improving the order consistency of the protection results. 5. Joint verification is implemented using the verification graph, candidate set version, and sending permission. Results are only output when permissions, quantity, and version are consistent, which relatively reduces the risk of abnormal results being sent. Attached Figure Description
[0015] Figure 1 This is a flowchart illustrating the sensitive data protection process of the present invention. Detailed Implementation
[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0017] Refer to the instruction manual appendix Figure 1 The sensitive data protection system for data circulation and transactions of the big data platform of the present invention includes: The edge resource module includes edge computing nodes, controlled storage, transaction interfaces, and resource schedulers. It receives requests and reads sensitive data, generates object weights, relational weights, and transaction keys according to the recipient and sorting fields, and allocates tasks according to input pages, result pages, and idle time slots to form candidate sets, permission tables, and task tables. After sensitive data transactions enter edge processing, it is necessary to first limit the data scope, permission status and computing resources of this transaction, and then provide a unified input to the sorting and reduction module. This implementation first locks the data version and access control rule version, then establishes a candidate set and permission table, and allocates computing tasks according to the storage pages and execution time slots of the edge computing nodes, so that sensitive data records are always kept in the controlled memory on the data source side. The controlled memory adopts a shared memory structure with page-addressable memory. Each memory page is configured with a page number and node access bits. The transaction interface is used for writing data transaction requests, and sensitive data records are read by the edge computing nodes specified by the task table. The implementation process includes the following steps: The data boundary is fixed based on the object scope and data version in the data transaction request; the edge computing node reads the controlled memory page by page, and only writes sensitive data records with the same data version and object identifiers located between the two ends of the object scope into the candidate set. The start and end object identifiers of the object scope are both included in the search scope. When duplicate object identifiers appear within the same data version, retain the sensitive data record with the earlier page number and the earlier position within the page, and write the duplicate record. When a sorting field is missing, the field format is inconsistent with the sorting field version, or the field value cannot be converted, the corresponding sensitive data record will not be written into the candidate set, and a field error record will be generated. After the candidate set is formed, it is sorted in ascending order according to the object identifier, and the receiver identifier, sorting field version and data version are concatenated in sequence according to the fixed field width to form the transaction key; if no qualified sensitive data record is obtained within the object range, no candidate record is written and the current processing is terminated. Permission status is solidified by writing object content permissions and object priority relationship permissions respectively; for each object in the candidate set, the object retrieval key is composed of receiver identifier, object identifier and locked access control rule version. When the allowed delivery item is matched in the access control rule, the object weight is written as one, and when it is not matched, the object weight is written as zero. Then, non-duplicate object pairs are generated in ascending order of object identifiers, and two direction bits are set for the same object pair. The relationship retrieval key is composed of the receiver identifier, sort field version, object pair, direction bits and access control rule version. When an allowed disclosure item is hit, the relationship weight bit is written as one, and when it is not hit, the relationship weight bit is written as zero. The object weight and the relation weights in both directions are written to the permission table according to the transaction key. When the access control rule is missing, the access control rule version changes, or the search key is duplicated, the corresponding weight is written to zero and a permission exception record is generated, which is used by the sorting and reduction module to read the corresponding relation weight according to the order edge direction. Task placement is based on the storage pages occupied by the candidate set and the reserved positions in the permission table; the resource scheduler uses the storage pages where the candidate set is located to form the input page, uses the reserved positions in the permission table to form the result page, and uses the number of object pairs plus the number of records in the permission table to form the number of task slots; The number of idle time slots for each edge computing node is calculated by subtracting the number of allocated time slots from the total number of node time slots. When the node access bit allows reading all input pages, the number of idle pages covers the number of result pages, and the number of idle time slots covers the number of task time slots, the corresponding edge computing node is written into the candidate node table. The candidate nodes are sorted in ascending order according to the number of remaining time slots after the task is written. If the order is the same, the first node is selected in ascending order of node number. Then, the transaction key, input page number, result page number, task time slot number and node number are written into the task table. When the candidate node table is empty, the resource scheduler splits the permission retrieval task according to the object pair order and recalculates the number of task time slots; if a single permission retrieval task still cannot be allocated, a resource shortage record is written and the transaction is stopped, and sensitive data records are kept in controlled storage. Taking supplier evaluation data transactions as a specific application scenario, the data transaction request is limited to reading supplier records with object identifiers from 100 to 120 under the same data version, and the version of the fulfillment rate sorting field is specified. Edge computing nodes exclude records with inconsistent data versions, duplicate object identifiers that are not the first record, and records with abnormal fulfillment rate fields, forming a candidate set arranged by object identifier. Then, the supplier object weight and the relationship weight in both directions are written into the candidate set input page and the permission table result page, and then the first edge computing node that meets the requirements of page access, storage page, and execution time slot is assigned to the first edge computing node. Through the above processing, the candidate set, permission table, and task table remain consistent under the fixed transaction key and access control rule version, providing definite input for order edge generation, relationship disclosure determination, and positional projection.
[0018] The sorting and reduction module is used to sort the candidate set according to the task table, write adjacent objects as order edges, take objects within the delivery number to form the deliverable result, merge the current and historical order edges with the same transaction key, and use a dynamic transfer reduction algorithm to delete order edges with alternative paths to form a reduced graph. The sorting reduction process is used to write the current sorting relationship and the historical sorting relationship already obtained by the receiver into the directed graph, and delete duplicate order edges without changing the object reachability relationship and the permitted reachability relationship, so as to retain the necessary relationship for subsequent bidirectional tracing. During execution, the transaction key, data version, sorting field version, and access control rule version are fixed; historical priority edges retain their historical disclosure status, while current priority edges use the current relation weight in the permission table to prevent permission changes from rewriting already disclosed relations; this implementation process includes the following steps: The candidate object sorting process first converts sensitive data records into unique object order; the edge computing node reads the candidate set according to the input page defined by the task table, and extracts the sorting field value from each sensitive data record according to the sorting field version. After completing the data format conversion, the data is arranged according to the sorting direction. When the values of the sorting field are the same, they are sorted in ascending order according to the object identifier. Sensitive data records with missing sorting fields, incorrect formats, or failed conversions are written into the sorting error record and do not participate in this sorting. After the object order is formed, connect adjacent objects starting from the first one, generate the current order edge according to the direction from the previous object to the next object, and form the edge key with the transaction key, the identifier of the previous object and the identifier of the next object; Edges with the same edge key are merged into one, and then edge numbers are assigned in ascending order according to the previous object's order, the next object's order, and the edge key. This time, the order edge reads the relation weight consistent with the edge direction from the permission table. If the number of deliverable bits exceeds the number of candidate objects, write a delivery quantity exception record and stop processing. Otherwise, take the first object in the order of objects up to the corresponding object in the delivery bit to form the delivery result. When historical relationships are merged into the original relationship graph, the historical disclosure status and the current permission status are saved separately; historical priority edges with the same transaction key are written to the current processing area, and the current permission field of the historical priority edge is updated with the current relationship weight in the permission table, while the historical disclosure field is retained. When reading historical order edges in graph operations, the historical disclosure field is used as the relation weight; when reading current order edges, the current relation weight is used as the relation weight. If a historical ranking edge endpoint does not enter the current candidate set, the endpoint remains in the original relationship graph for cross-request relationship tracking, but does not enter the pending delivery results; When both endpoints are within the candidate set, the edge direction is checked according to the object order. The reverse duplicate edges formed by the standard object's direction conversion are rewritten according to the direction position and deduplicated. Edges that are still opposite to the object order are written into the relationship conflict record and the processing is stopped. After completing the direction verification, merge the current sequential edges with the same edge key with the historical sequential edges, and retain the edge source and relation weight respectively to form the original relation graph without duplicate edges; After entering the dynamic transitive reduction operation, whether each sequential edge is retained is determined by the relational weight status of the alternative path; the edge computing node traverses the original relation graph in ascending order of fixed edge number and temporarily deletes the current sequential edge; When the relation weight of the current sequential edge is one, search only along the sequential edge with a relation weight of one from the start point to the end point of the current sequential edge. If an alternative path is found, retain the deletion result; if no alternative path is found, restore the current sequential edge. When the relation weight of the current sequential edge is zero, the search state is formed by the object identifier and whether it has passed through the sequential edge with the relation weight of zero. The search is carried out along the remaining sequential edges. When the destination is reached and the corresponding state has passed through the sequential edge with the relation weight of zero, the deletion result is retained. Otherwise, the current sequential edge is restored. After each deletion result is retained, the deleted edge and the alternative path used are written into the deletion record; after completing one round of traversal, as long as the sequential edge has been deleted in this round, traverse again from the first and second edges of the fixed edge number until no sequential edge has been deleted in one round of traversal, and output the candidate simplified graph. Each deletion result reduces one edge in order, and the total number of deletions is limited by the number of edges in the original graph, thus ensuring that the dynamic propagation reduction algorithm stops. The final confirmation of graph relationships is completed through bit set propagation and difference recovery; first, the original relationship graph is topologically arranged, and fixed-length bit numbers are assigned to all objects, and then the objects are processed in topological order: the endpoint bit number of each outgoing edge of the current object and the endpoint full edge bit set are written into the full edge bit set of the current object by bitwise OR. When the weight of the outgoing edge relationship is one, the endpoint bit number and the endpoint permission bit set are written into the current object permission bit set; the candidate reduction graph generates the corresponding bit set according to the same bit number and processing order. To verify the reverse reachability, the same propagation is performed on the reverse edges of the two graphs to form the predecessor full edge bit set and the predecessor permitted bit set; Compare bit by bit the successor full edge set, successor permitted bit set, predecessor full edge set, and predecessor permitted bit set of the object to be delivered in the two graphs; When a discrepancy occurs, the records to be deleted are retrieved based on the discrepancy object. The discrepancy path is determined in the original relation graph in ascending order of edge number, and the deleted edges with the first edge number in the discrepancy path are restored. Then, the process is re-propagated. When the four types of bit sets are consistent bit by bit, the candidate reduction graph retains all the arrival relationships and permitted arrival relationships related to the object to be delivered in the original relationship graph. Based on this, a reduction graph is written for the access determination module to read. Taking the supplier fulfillment rate sorting transaction as a specific scenario, the candidate set includes object identifiers A to H, with three delivery bits. This sorting results in A pointing to B, B pointing to C, and subsequent adjacent order edges; the record storage also stores the historical order edges of A pointing to D and D pointing to F under the same transaction key. Historical sequential edges retain their disclosed state and participate in the original relationship graph. The dynamic transfer reduction algorithm deletes sequential edges that can be replaced by similar permitted paths. Subsequently, the arrival relationships between A, B, C and historical objects are confirmed to be unchanged through forward and reverse bit sets. The resulting simplified graph retains the current ordering relationship and historical disclosure relationship with fewer order edges, and provides a definite graph structure for subsequent identification of whether the object to be delivered is associated with an unauthorized object.
[0019] The access determination module is used to trace bidirectionally from the object to be delivered along the simplified graph, write the object to be delivered with zero weight and the unauthorized endpoint of the first relation with zero weight priority edge into the removal table, delete the removed object and the connected edge, and form the retained graph; After the simplified diagram is formed, the access determination needs to check the delivery permissions of the object to be delivered as well as the disclosure scope formed by the historical priority relationship and the current priority relationship of the object to be delivered. This implementation method controls access to object content using object authority and controls access to the hierarchical relationship between objects using relation authority. During bidirectional tracking, it identifies objects that cause unauthorized disclosures and removes them. Then, through round-by-round deletion and review, a retention map is formed. This implementation process includes the following steps: The tracking preparation phase is used to fix the tracking starting point and exclude objects to be delivered without delivery permissions; according to the object priority, the objects to be delivered and the object permissions in the permission table are read in sequence. When the object permission is zero, the corresponding object to be delivered is written to the removal table and an insufficient object permission flag is written. When the object weight is one, the object identifier to be delivered, the object sequence number, and the starting number are written into the forward queue and the reverse queue respectively to form a bidirectional tracking starting set. When the permission table is missing the object permission corresponding to the object to be delivered, it is treated as having a zero object permission and written into the permission missing record; when the same object to be delivered appears repeatedly, only the record with the first object priority is retained, and the rest of the records are written into the record of the repeated starting point. The graph expansion uses a layer-by-layer approach to identify unauthorized boundaries; the forward queue expands from the tracking start point along the outgoing edge direction, and the reverse queue expands from the tracking start point along the incoming edge direction. Each tracking state is written with the tracking start point, current object, tracking direction, and edge number sequence. States at the same layer are processed in ascending order of edge number sequence. When the same tracking starting point, tracking direction and current object form a repeated state, only the tracking path with the edge number sequence first is retained, and different tracking starting points or different tracking directions are saved separately. When a historical object not registered in the permission table is reached, the object's permissions are read according to the current fixed access control rule version. If no object permission record is found, the object's permissions are written to zero. When an object with a weight of zero is encountered, the path to it is written into the overweight path set and expansion stops. When a sequential edge with a relation weight of zero is encountered, the path containing the corresponding sequential edge is written into the overweight path set and expansion along the corresponding sequential edge stops. The remaining sequential edges continue to the next level until both the forward queue and the reverse queue are empty. The overstepping of authority path reduction is used to retain the priority edge with a relationship weight of one while truncating the disclosure relationship; the positive overstepping of authority path takes the end point of the first priority edge with a relationship weight of zero as the candidate object to be removed, and the reverse overstepping of authority path takes the starting point of the first priority edge with a relationship weight of zero as the candidate object to be removed. When the object weight is zero, the corresponding object is directly removed. When a candidate object to be removed does not belong to the candidate set, read the first object that belongs to the candidate set and has not yet been written to the removal table along the unauthorized path towards the starting point of the tracking. If the starting point of the tracking is in the candidate set, take the starting point of the tracking; For each candidate object to be removed, the number of paths in the current unauthorized path set that contain the corresponding object is counted as the number of truncated paths, and the number of relations with weights that are deleted along with the corresponding object after it is removed from the simplified graph is counted as the number of permitted edge losses. The first object is then written into the removal table in descending order of the number of truncated paths, ascending order of the number of permitted edge losses, descending order of the object's weight, and ascending order of the object's identifier. After writing, delete the unauthorized paths containing the corresponding objects, and recalculate the object records for the remaining unauthorized paths until the set of unauthorized paths is empty; The retained graph generation ensures that the removal results cover all unauthorized relationships through deletion verification; based on the removal table, the removed objects and the sequential edges with the removed objects as the starting or ending points are deleted from the simplified graph to form a graph to be verified; then, the forward queue and reverse queue are re-established with the unremoved objects to be delivered, and a new set of unauthorized paths is generated according to the same bidirectional tracking rules. When the new set of unauthorized paths is empty, the graph to be kerneled is written as the reserved graph; when the new set of unauthorized paths is not empty, the object record is re-formed and an object is written to the removal table, and then deleted and tracked again. Each round of review adds at least one object to be removed from the candidate set, and the number of rounds is limited by the number of objects in the candidate set. When all objects to be delivered are written into the removal table, an empty retention graph is output and a record of no compliant objects to be delivered is written, so that the positional projection module can stop forming the protection result. In the supplier fulfillment sequencing task, the object weights of objects A, B and C to be delivered are all one. However, in the simplified diagram, B reaches object X along the historical order edge, and the relation weight of B pointing to X is zero. The forward tracing writes the path from B to X into the overweight path set. If X is not in the current candidate set, it backtracks along the path and writes B as a candidate to be removed. When another path leads from C through object D to object Y with zero weight, objects D and C participate in the calculation of the number of truncated paths and the number of permissible edge losses, and the objects to be removed are determined in a predetermined order. After deleting the selected object, perform bidirectional tracing again. If no unauthorized path is formed, output the reserved graph so that subsequent positional projections only process candidate objects whose prohibited order relationships are not disclosed.
[0020] The positional projection module is used to delete objects removed from the candidate set, form positional vectors according to their order, generate topological order and positional boundaries from the retained graph, and use the Dykstra alternating projection algorithm to perform sequential projection, interval projection and residual write-back until the object block and the boundary fixed object are consistent in two adjacent rounds. Finally, the protection result is formed according to the positional value and topological order. After the access determination module outputs the shift table and the retention map, the positional projection processing needs to retain the permission order relationship while excluding unauthorized objects, and restrict the positional order of objects from falling into the position range determined by the retention map; This implementation method first determines the objects and topological order involved in the projection, then alternately corrects the order relationship and positional boundaries using the Dykstra alternating projection algorithm, and finally obtains a unique positional order based on the stable constraint state to form the protection result; the implementation process includes the following steps: First, the range of objects participating in the bit order operation is limited by the removal table; edge computing nodes remove the removed objects from the candidate set and write the remaining objects into the projection object set; when historical objects in the graph do not enter the projection object set, they are retained only for their tracking role in the access determination and are not included in the bit order vector. For each object in the projected object set, subtract the number of objects removed from the object before it from the object's order before deletion to obtain the initial order value, and write it as the initial order vector according to the object's order before deletion. When the removal table contains objects outside the candidate set, the corresponding deletion item is ignored and an abnormal removal range record is written. When the projection object set is empty, the positional projection is stopped and an unprotected result record is generated. Topological relationships and positional boundaries are formed based on the sequential edges of the retained graph within the projected object set; edge computing nodes only retain sequential edges whose two ends both belong to the projected object set, write the projected object into the unexpanded object set, and count the number of sequential edges from the unexpanded object to itself as the current in-degree; In each round, the first object is selected from the objects with a current in-degree of zero in ascending order of their previous order. If the previous order of objects is the same, the first object is selected in ascending order of their object identifiers. The selected object is written into the topological order. Then, the selected object and its outgoing edges are deleted from the unexpanded object set, and the current in-degree of the endpoint of each outgoing edge is decremented by one. Topology expansion is completed when the unexpanded object set is cleared; if the unexpanded object set still contains objects and there are no objects with a current in-degree of zero, a sequence conflict record is generated and processing stops. After completing the topological expansion, count the number of different successor objects that each object can reach along the sequential edge direction, and the number of different predecessor objects that can reach the current object in the opposite direction. Add one to the number of predecessor objects to form the lower bound of the positional order, and subtract the number of successor objects from the number of projected objects to form the upper bound of the positional order. When entering sequential projection, the positional rewriting comes at the cost of the sum of squared differences of the deviation from the initial positional vector, and the positional value of the preceding object in the topological order is not higher than the positional value of the following object as the order constraint. The first round of sequential residuals is set to zero one by one. After adding the sequential residuals to the current position vector, a single object block is built for each object according to the topological order. The block average is formed by dividing the sum of the position values in the block by the number of objects in the block. When the mean value of the left object block is higher than the mean value of the right object block, the two blocks are merged and the block mean value is recalculated until the mean value of each object block remains unchanged along the topological order. The block average is stored as a fraction. The order of the two block averages is determined by cross-multiplying the numerator with the denominator of the other block to avoid rounding off decimals and changing the merged result. Write the block mean of the block containing each object as the sequential projection value, then subtract the sequential projection output position value from the sequential projection input position value of this round to form a new sequential residual, and output the sequential projection vector; In the interval correction phase, the sequential projection vector is received and positional boundary constraints are applied. The interval residuals of the first round are set to zero one by one. The interval residuals of the previous round are added to the sequential projection vector. When the resulting positional value is lower than the lower bound of the positional value, it is written as the lower bound of the positional value. When it is higher than the upper bound of the positional value, it is written as the upper bound of the positional value. When it is in between, the original value is retained. Then, the positional value of the interval projection input is subtracted from the positional value of the interval projection output to form a new interval residual. The interval projection vector is output and used as the current positional vector for the next round. Each round encodes the start and end positions of the object block, the lower bound fixed object bit set, and the upper bound fixed object bit set according to a fixed width to form the round code; When two adjacent round codes are the same, the object block is determined according to the round code. For object blocks without fixed boundary objects, the arithmetic mean of the initial position values within the block is calculated. For object blocks with fixed boundary objects, the fixed value is read. The lower bound of the block is formed by the last value of the lower bound of the position within the block, and the upper bound of the block is formed by the first value of the upper bound of the position within the block. The block value is restricted to between the lower bound and the upper bound of the block. When an object block has two different fixed values, the object block is split between the two fixed objects and the projection continues alternately; when the lower bound of the block is higher than the upper bound of the block, a bit order conflict record is generated and processing stops. When the obtained bit sequence values satisfy the object block mean relationship, bit sequence boundary and topology order non-decreasing relationship, the objects are arranged in ascending order of bit sequence values. When the bit sequence values are consistent, the topology order is used, and the objects within the delivery bit number are taken to form the protection result. When the number of remaining objects is less than the delivery bit number, a protection result insufficient record is written, so that the result output module stops sending. Taking supplier fulfillment rate as the implementation scenario, after deleting two removed objects from the candidate set, the original third, fifth and sixth ranked suppliers are rewritten in their initial order according to the number of removed objects in front. The retention diagram restricts that the preceding supplier cannot be ranked after the following supplier, and at the same time, the allowable order range is determined based on the predecessor and successor objects of each supplier. Sequential projection merges object blocks that violate the order relationship, while interval projection writes objects that have exceeded the order range back to the boundary. The two residuals are compensated for in subsequent rounds by the correction amount canceled out by another projection. After the round code is stabilized, the fractional order is obtained according to the object block and boundary relationship, and a protection result that meets the requirements of access permission, historical order relationship and delivery bit length is formed.
[0021] The result output module includes a release controller, a security interface, and a record memory. It reconstructs the protection result order edge and reduces it again, performs bidirectional tracking, and sends the protection result and stores the reduced graph and transaction key when the shift table is empty and the number of protection results equals the number of delivery bits. When the shift table is not empty, it fills in the unauthorized objects in the unauthorized path and reprojects them. It stops sending when the number of protection results is lower than the number of delivery bits. Before the protection results enter the sending stage, it is necessary to verify again whether the rearranged object order introduces new unauthorized relationships, and ensure that the data version, permission status and historical order edge on which the review is based are consistent with the bit order projection stage. This implementation uses a verification diagram to receive the protection results, a verification removal table to record unauthorized objects that have not yet been eliminated, and a release controller to control the security interface to send the data when the version, quantity, and permissions are consistent. The implementation process includes the following steps: The verification graph is built starting from the object order in the protection results; the edge computing nodes write adjacent objects as reconstruction order edges pointing from the previous object to the next object according to the arrangement order of the protection results, and use transaction keys and current relationship weights for the reconstruction order edges; Then, all historical order edges with the same transaction key are read from the record storage, the historical disclosure status is retained and merged with the reconstruction order edges according to the edge key. When the edge keys are the same, only one order edge is retained. At the same time, the current source and the historical source are recorded. The merged graph is processed by a dynamic transitive reduction algorithm to remove sequential edges with alternative paths, and the fully reachable and permitted reachable relationships are verified. When a reverse sequential edge, missing edge endpoints, or incomplete reading of historical sequential edges occurs, write an exception record to the verification graph and stop sending. The relationship graph is written as the verification graph after passing the verification. The permission review adopts the same bidirectional tracing rule as the access judgment module; it takes the object in the protection result as the starting point of the tracing, and expands layer by layer along the outgoing and incoming edges of the review graph, and reads the object's rights and the relationship rights consistent with the edge direction according to the permission table; When the object weight is zero, the corresponding object is written to the review and removal table. When the relation weight is zero, the blocking object corresponding to the first relation weight zero priority edge is written to the review and removal table. When the blocked object does not belong to the current candidate set, read the first object that belongs to the current candidate set and has not yet been written to the removal table along the overriding path to the tracing starting point. If there are no other candidate objects in the path, take the tracing starting point. After completing bidirectional tracking, the verification chart, verification removal table, number of objects in the protection result, transaction key, and candidate set version are written into the verification result; each time the protection result is regenerated, the old verification removal table is cleared first, and then the verification chart and verification result are regenerated. The release permission is determined by the record to be released in the record storage and the result of this review; the candidate set version is generated according to the transaction key, the candidate set object identifier order, the move-out table object identifier order and the bit order projection round code, and remains unchanged during the review; After the bit sequence projection is completed, the transaction key and candidate set version are first written into the record storage to form a record to be released. The release controller then checks and verifies that the shift table is empty, the number of objects in the protection result is equal to the number of bits delivered, the transaction key is consistent, the candidate set version is consistent, the data version has not changed, and the access control rule version has not changed. If all verification results are consistent, the controller generates a send permission and writes the protection result, transaction key, candidate set version, and send sequence number into the release record; if any field is inconsistent, the send permission is revoked and the security interface is blocked. After obtaining permission to send, the security interface generates data frames according to the object order in the protection result. Each data frame carries a sending sequence number, object order number, and allowed delivery field. After all data frames have been sent, a completion frame is sent. After the receiving end returns a transmission completion bit that matches the transmission sequence number, the record memory rewrites the record to be released as a released record and updates the historical sequence edges with a verification graph that includes all historical sequence edges and the current reconstructed sequence edges. When the removal table is not empty, the first path is taken in ascending order of the length of the overriding path, and then the first blocking object belonging to the current candidate set and not yet written to the removal table is taken in ascending order of the edge number and added to the removal table. Only one object is added in each round and the positional projection is re-executed. After a new protection result is generated, the review and removal table is cleared and the protection is reviewed again until the review and removal table is empty or the number of objects in the protection result is less than the number of bits delivered. If the number of protection results is insufficient, the transmission completion bit is not returned, or the transmission content is inconsistent with the release record, the transmission is stopped, the transmission sequence number is invalidated, and a record of non-compliant results is written. The historical sequence edge remains in its original state. Taking the sending of supplier ranking results as an example, the protection result contains three supplier objects, and the verification graph is formed by the reconstruction ranking edges of the three objects and the historical ranking edges under the same transaction key. When bidirectional tracking finds that one of the objects can still reach the object with a relational weight of zero along the historical order edge, the supplier objects in the current candidate set that can truncate the corresponding over-weight path are added to the removal table, and the positional projection is performed again to reconstruct the verification graph. After verifying that the removed table is cleared and the protection result still contains three objects, the release controller checks the transaction key, candidate set version, data version, and access control rule version. It then completes the frame transmission via the security interface and receives the transmission completion bit before updating the historical sequence edge, thereby ensuring that the transmission result is consistent with the final access permissions and historical disclosure relationship.
[0022] exist Figure 1 It should be noted that: the data transaction request passes through the edge resource module, sorting and reduction module, access determination module, positional projection module and result output module in sequence, respectively completing the generation of candidate set and permission table, order relationship reduction, unauthorized path identification, protection positional reconstruction and sending for review; the solid arrows in the figure uniformly indicate the data transmission direction and processing order relationship between the modules. When the review is passed, the protection result is sent. When the review fails, it returns to the positional projection module for reprocessing. When the number of protection results is insufficient, the sending is stopped and the processing result is recorded.
[0023] Overall Execution and Implementation Description: This solution first involves the edge resource module reading sensitive data from the data source side, fixing the recipient, data version, sorting field version, and access permissions to form a candidate set, permission table, and task table. Then, within the edge computing nodes, object sorting is completed, merging the current and historical order relationships, and using a dynamic reduction algorithm to remove duplicate relationships, forming a reduced graph that retains necessary comparison relationships. The access determination module then traces the reduced graph forward and backward from the objects to be delivered, identifying paths where object content or order relationships exceed the authorized scope, and writing the corresponding objects into the removal table. The positional projection module deletes the removed objects and rearranges the objects using the Dykstra alternating projection algorithm while preserving the authorized order and positional boundaries, forming the protection result. Finally, the result output module reconstructs and verifies the order relationships. Results are only sent when the number of protection results, transaction version, and permission status are consistent; otherwise, objects causing unauthorized relationships are removed, and sending stops when the delivery quantity cannot be met. Taking a supplier performance evaluation transaction as an example, when a buyer requests the top three suppliers by performance rate, the platform first reads the supplier records of the same data version from the data source side and determines whether the buyer can obtain the records of each supplier and whether it can know the order of the suppliers. After the edge computing node completes the performance rate ranking, it merges the current ranking relationship with the ranking relationship previously obtained by the buyer. If it finds that the order of a certain supplier to be delivered can indirectly expose the performance position of another supplier that is not authorized to disclose, the supplier that can truncate the relationship is removed, and the order of the remaining suppliers is recalculated. After the re-ranking result is tracked and verified again, the buyer only receives the three suppliers within its authority and their controlled order. The performance value and relative position of the undelivered suppliers will not be disclosed along with the ranking result.
[0024] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A system for protecting sensitive data in data flow and transaction of a big data platform, characterized in that, include: The edge resource module includes edge computing nodes, controlled storage, transaction interfaces, and resource schedulers. It receives requests and reads sensitive data, generates object weights, relational weights, and transaction keys according to the recipient and sorting fields, and allocates tasks according to input pages, result pages, and idle time slots to form candidate sets, permission tables, and task tables. The sorting and reduction module is used to sort the candidate set according to the task table, write adjacent objects as order edges, take objects within the delivery number to form the deliverable result, merge the current and historical order edges with the same transaction key, and use a dynamic transfer reduction algorithm to delete order edges with alternative paths to form a reduced graph. The access determination module is used to trace bidirectionally from the object to be delivered along the simplified graph, write the object to be delivered with zero weight and the unauthorized endpoint of the first relation with zero weight priority edge into the removal table, delete the removed object and the connected edge, and form the retained graph; The positional projection module is used to delete objects removed from the candidate set, form positional vectors according to their order, generate topological order and positional boundaries from the retained graph, and use the Dykstra alternating projection algorithm to perform sequential projection, interval projection and residual write-back until the object block and the boundary fixed object are consistent in two adjacent rounds. Finally, the protection result is formed according to the positional value and topological order. The result output module includes a release controller, a security interface, and a record storage. It reconstructs the protection result sequence edge and reduces it again, performs bidirectional tracking, and sends the protection result and stores the reduced graph and transaction key when the out-of-table is empty and the number of protection results equals the number of delivery bits. When the out-of-table is not empty, it fills in the unauthorized objects in the unauthorized path and reprojects them. It stops sending when the number of protection results is lower than the number of delivery bits.
2. The sensitive data protection system for data circulation and transactions on a big data platform according to claim 1, characterized in that: The edge resource module includes: After receiving a data transaction request at the transaction interface, the controlled storage is searched by object scope. The sensitive data records that are hit are written into the candidate set in ascending order of object identifier, and the receiver identifier, sorting field version and data version are concatenated to form a transaction key. For each object in the candidate set, the receiver identifier and object identifier are concatenated to retrieve access control rules. If an allowed delivery item is matched, the object weight is written to 1; otherwise, the object weight is written to 0. Then, non-repeating object pairs are formed by sorting the object identifiers in ascending order. The receiver identifier, sorting field version, and object pair are concatenated to retrieve access control rules. If an allowed disclosure item is matched, the relationship weight is written to 1; otherwise, the relationship weight is written to 0, thus forming the permission table. A task is generated based on the candidate set input page and the permission table result page. The sum of the number of object pairs and the number of records in the permission table is used as the task time slot number. From the edge computing nodes that can access the input page, have no less than the number of free pages and no less than the number of free time slots, they are sorted in ascending order according to the number of remaining time slots after the task is executed. If the order is the same, the first node is selected in ascending order of node number. The input page, result page and task time slot number are written into the task table, so that sensitive data records are kept in the controlled memory.
3. The sensitive data protection system for data circulation and transactions on a big data platform according to claim 2, characterized in that: The sorting and reduction module includes: Based on the task table, the candidate set is read in the edge computing node. The object order is determined by the sorting field value. If the sorting field values are the same, the object order is determined by the object identifier in ascending order. Adjacent objects are written as current order edges carrying transaction keys and relation weights. The objects within the delivery bit range are taken to form the deliverable result. Update the historical order edge that matches the transaction key with the current relationship weight in the permission table, then merge the current order edge with the updated historical order edge, check the order edge direction according to the object order, delete the order edge whose direction is opposite to the object order, and form the original relationship graph.
4. The sensitive data protection system for data circulation and transactions on a big data platform according to claim 3, characterized in that: The sorting and reduction module also includes: For each sequential edge in the original graph, the current sequential edge is temporarily deleted according to the dynamic transitive reduction algorithm, and an alternative path from the start point to the end point is searched. When the relation weight is one, the deletion result is retained only when the relation weights of all sequential edges in the alternative path are one. When the relation weight is zero, the deletion result is retained only when there is a sequential edge with a relation weight of zero in the alternative path. In other cases, the current sequential edge is restored until no sequential edge is deleted in one round of processing, forming a candidate reduced graph. Based on object order, reverse bit set propagation is performed on the original relation graph and the candidate reduced graph respectively. All successor objects of each object are written into the full edge bit set, and successor objects reached through the first-order edge with relation weight are written into the permitted bit set. Then, the full edge bit set and permitted bit set of the corresponding objects to be delivered in the two graphs are compared bit by bit. When a difference occurs, the deleted first-order edge in the difference path is restored and propagated again until the full edge bit set and permitted bit set of the two graphs are consistent bit by bit, forming the reduced graph.
5. The sensitive data protection system for data circulation and transactions on a big data platform according to claim 4, characterized in that: The access determination module includes: Using the object to be delivered as the starting point for tracing, the object weight is read according to the object order. The objects to be delivered with a weight of zero are written to the removal table, and the objects to be delivered with a weight of one are written to the forward queue and the reverse queue respectively, forming a bidirectional tracing starting point set. The tracing path is expanded layer by layer from the starting point in the forward queue along the outgoing edges of the reduced graph, and from the starting point in the reverse queue along the incoming edges of the reduced graph. Edges in the same layer are read in ascending order of edge number. For objects that are visited repeatedly, only the tracing path with the edge number sequence first is retained. When an object with a weight of zero or a sequential edge with a relation weight of zero is encountered, the current tracing path is stopped from being expanded, forming a set of overweighted paths.
6. The sensitive data protection system for data circulation and transactions in a big data platform according to claim 5, characterized in that: The access determination module also includes: For each path in the set of unauthorized paths, trace forward to find the endpoint of the first edge with a zero relation weight, and trace backward to find the starting point of the first edge with a zero relation weight. When the object weight is zero, directly take the corresponding object. Write the number of unauthorized paths that can be truncated for each object and the number of edges with a one relation weight lost after deletion into the object record. Select the first object in descending order of the number of truncated paths, ascending order of the number of lost edges, and descending order of object weight and write it into the removal table. Delete the truncated unauthorized paths until the set of unauthorized paths is empty. Remove the removed objects and their adjacent order edges from the simplified graph based on the removal table. Then, repeat the bidirectional tracing for the unremoved objects to be delivered. If no unauthorized path is formed, output the retained graph after deletion. If an unauthorized path is still formed, continue to write an object to the removal table according to the object record and trace again until no unauthorized path is formed or all objects to be delivered are written to the removal table.
7. The sensitive data protection system for data circulation and transactions on a big data platform according to claim 6, characterized in that: The positional projection module includes: Based on the removal table, remove the objects to be removed from the candidate set. Subtract the number of objects to be removed from the previous position according to the order before deletion to form an initial position vector. Then perform a topological expansion on the retained graph. Each time, take the first object from the objects with zero in-degree in ascending order according to the order before deletion and the object identifier to form the topological order. Add one to the number of predecessor objects reached by each object along the retained graph to form the lower bound of the position vector. Subtract the number of successor objects from the number of retained objects to form the upper bound of the position vector. The rewriting cost is the sum of squared differences between the positional vector and the initial positional vector. The projection constraints are topological order increase and positional values between the lower and upper bounds of the positional vector. After adding the order residual to the current positional vector, a single object block is established according to the topological order. When the mean of the left object block is higher than the mean of the right object block, the two blocks are merged and the mean is recalculated until the mean of each object block increases according to the topological order. Then, the positional difference before and after projection is written into the order residual to form the order projection vector.
8. The sensitive data protection system for data circulation and transactions on a big data platform according to claim 7, characterized in that: The positional projection module further includes: For the sequential projection vector, the interval residual is added back bit by bit. The bit value below the lower bound of the bit order is written as the lower bound of the bit order, and the bit value above the upper bound of the bit order is written as the upper bound of the bit order. The bit difference before and after writing is updated as the interval residual to form the interval projection vector. Then, the interval projection vector is used as the bit order vector for the next round of sequential projection and interval projection are repeated so that the residual of the previous round participates in the bit order correction of the next round. In each round, the object block boundary, the objects written as the lower bound of the bit order, and the objects written as the upper bound of the bit order are sequentially encoded into round codes. When the round codes of two adjacent rounds are consistent, the bit order value is solved based on the equality relationship of the object block average and the fixed relationship of the boundary. The topology order and bit order boundary are re-verified. When the verification is successful, the objects are arranged in ascending order of bit order value. When the bit order values are the same, the topology order is used. The objects within the delivery bit number are taken to form the protection result.
9. The sensitive data protection system for data circulation and transactions in a big data platform according to claim 8, characterized in that: The result output module includes: Based on the object order in the protection results, the order edges are reconstructed, the reconstructed order edges are merged with the historical order edges that are consistent with the transaction key, and the order edges with alternative paths are deleted according to the dynamic transfer reduction algorithm to form a verification graph. Starting from the object in the protection result, perform bidirectional tracing along the outgoing and incoming edges of the review graph. Read the object's authority and relation authority according to the permission table. Write the object with zero authority and the unauthorized endpoint of the first relation with zero authority into the review remove table to form the review result.
10. The sensitive data protection system for data circulation and transactions in a big data platform according to claim 9, characterized in that: The result output module also includes: When the review and removal table is empty and the number of objects in the protection result is equal to the number of delivery bits, the release controller writes the protection result, transaction key and candidate set version into the release record, and sends the protection result through the security interface when the release record is consistent with the transaction key and candidate set version in the record storage, and updates the historical order edge with the review graph at the same time. When the verification table is not empty, the first unauthorized path is selected in ascending order of unauthorized path length, and then the first unauthorized object not written to the table is added to the table in ascending order of edge number. The bit order projection is then re-executed until the verification table is empty or the number of objects in the protection result is lower than the number of delivery bits. When it is lower than the number of delivery bits, the sending is stopped and a record of non-compliant result is written.
