An archive sensitive fragment identification desensitization method

Abstract

The present application relates to the technical field of archive information opening processing, and particularly relates to a method for identifying and desensitizing archive sensitive fragments, which forms a fragment analysis object set by performing basic objectization processing on the to-be-processed archive data, constructs an archive fragment correlation graph, evaluates the isolation risk, correlation reconstruction risk and jigsaw path of the candidate sensitive fragment, determines the cooperative control level and target opening output content, and implements desensitization to generate an opening copy and an audit record. The present application constructs an archive fragment correlation graph by objectizing the to-be-processed archive data and combining archive metadata, evaluates the isolation risk, correlation reconstruction risk and jigsaw reconstruction path of the candidate sensitive fragment, determines the cooperative control level and target opening output content, realizes accurate desensitization, and generates a traceable audit record.

Description

Technical Field

[0001] This invention relates to the field of open processing technology for archival information, and more specifically, to a method for identifying and desensitizing sensitive archival fragments. Background Technology

[0002] Open access to archives is a crucial aspect of archival information services. During the generation of open copies of archives, it is typically necessary to review and process the archival text, tables, images, annotations, seals, attachments, and metadata containing information such as personal identity, work experience, organizational affiliation, kinship, awards and punishments, and approval opinions. This is to meet the needs of archival use while minimizing the risk of sensitive information leakage. Existing methods for desensitizing archives largely rely on manual review, keyword matching, sensitive field identification, rule validation, page area masking, and some natural language processing techniques. These methods are effective in identifying and processing explicit sensitive content such as ID numbers, contact information, names, addresses, and organization names, and are suitable for relatively standardized and clearly structured open access scenarios.

[0003] However, archival materials possess strong historical continuity, interrelationships, and complex formats. The same sensitive fact is often not concentrated in a single field or page, but rather scattered throughout the main text, table items, attachments, approval annotations, seals, and archival metadata. For example, a passage in the main text might only record a job adjustment, a table might record the person's name and family relationship, an annotation might record the approval opinion, and archival metadata might record the creation time, responsible person, and title. While the sensitivity of these contents may be limited when presented individually, by complementing each other with factual elements such as personnel, time, unit, matter, and relationship, a relatively complete appointment and dismissal process, kinship relationships, or organizational handling can be reconstructed. This type of jigsaw puzzle-like reconstruction risk, formed by multiple fragments, multiple formats, and multiple attributes, is difficult to fully identify based solely on the sensitivity hit results of a single word, field, or page.

[0004] Furthermore, existing processing procedures, when identifying desensitization targets, typically focus more on whether a single point of content hits a sensitive rule, while insufficiently considering the strength of association between different openable output units, the complementarity of factual elements, the closure of the evidence chain, and the amplifying effect of archival metadata on sensitive facts. In actual open processing, if only a single explicit sensitive fragment is masked, other unprocessed fragments may still reconstruct the target sensitive fact through multiple independent paths; if the masking scope is expanded to avoid risks, it may affect the structural integrity, semantic continuity, and utilization value of the open copy of the archive. Therefore, the processing of sensitive fragments in archives not only needs to identify the risks of the candidate sensitive fragments themselves, but also needs to determine whether there are paths that can form associations and reconstructions between the fragment and other open content, factual elements, and archival metadata.

[0005] Therefore, a more refined method for identifying and de-identifying sensitive archival fragments is still needed in the process of generating open copies of archives. This method should be able to form a set of fragment analysis objects based on basic object-oriented processing, establish the relationship between archival fragments by combining archival metadata, assess the isolation risk, association propagation risk, and combined reconstruction risk of candidate sensitive fragments, and further analyze the puzzle reconstruction path, path redundancy, and minimum cut-off node. By determining the level of collaborative control and the target open output content based on the risk assessment results and path cut-off results, the method can balance risk control, open access, and audit traceability requirements when implementing masking, replacement, generalization, or summarization processing. Summary of the Invention

[0006] In view of the shortcomings of the existing technology, the purpose of this invention is to provide a method for identifying and desensitizing sensitive segments in archives.

[0007] To achieve the above objectives, the present invention provides the following technical solution: A method for identifying and de-identifying sensitive segments in archives includes: Step 1: Obtain the archive data to be processed and the corresponding archive metadata. Perform basic objectification processing on the archive data to be processed, including text recognition, layout parsing, structure parsing and fragment segmentation, to obtain a fragment analysis object set containing open output units and factual elements, and determine candidate sensitive fragments from the open output units in the fragment analysis object set. Step 2: Based on the set of fragment analysis objects and archival metadata, construct an archival fragment association map; Step 3: Based on the archival fragment association map, perform isolated risk calculation, associated risk propagation, and combined reconstruction risk assessment on candidate sensitive fragments to obtain the risk assessment results corresponding to the candidate sensitive fragments; among which, the risk assessment results include the target sensitive fact, isolated risk value, associated risk closure, associated reconstruction risk value, and fragment association reconstruction gain value; Step 4: Based on the risk assessment results and the archival fragment association map, perform jigsaw puzzle reconstruction path analysis and path cutting analysis on the candidate sensitive fragments to obtain the path cutting results corresponding to the candidate sensitive fragments; among which, the path cutting results include the jigsaw puzzle reconstruction path set, jigsaw puzzle path redundancy, and jigsaw puzzle path minimum cutting node set; Step 5: Based on the risk assessment results and path cutting results, determine the collaborative control level and target open output content of the candidate sensitive fragments, and desensitize the target open output content according to the collaborative control level, generating an open copy of the desensitized file and desensitization audit records.

[0008] In one embodiment, the fragment analysis object set includes multiple fragment analysis objects, each fragment analysis object including an openable output unit and analysis attributes associated with the openable output unit. The analysis attributes include open output content, carrier type, page position, structural position, factual elements, open visibility weight, sensitive rule hit results, and candidate tags.

[0009] In one embodiment, the archival fragment association graph includes graph nodes, association edges, and association edge weights; the graph nodes include output unit nodes corresponding to openable output units, fact element nodes corresponding to fact elements, and archival attribute nodes corresponding to archival attributes in archival metadata; archival attributes are information items contained in archival metadata; association edges connect graph nodes, and the association edge weights represent the contribution of the association edge to the reconstruction of sensitive facts.

[0010] In one embodiment, the contribution of sensitive fact reconstruction is obtained by matching the co-occurrence, complementarity and mutual verification of the open output content, fact elements or archive attributes corresponding to the graph nodes at both ends of the associated edge to obtain the relationship strength and the criticality of the fact elements. This is then combined with the open visibility weight and weighted normalization to generate the associated edge weight.

[0011] In one embodiment, the fragment association reconstruction gain value is determined as follows: the fragment association reconstruction gain value of the candidate sensitive fragment is determined by fusing the normalized increase of the association reconstruction risk value relative to the isolated risk value, the increase of the factual element completeness of the target sensitive fact, the increase of the evidence chain closure strength, the cross-carrier mutual evidence strength, and the amplification strength of the archive attribute visibility; wherein, the increase of factual element completeness is the increment of the association risk closure relative to the candidate sensitive fragment in the degree of factual element coverage of the target sensitive fact.

[0012] In one embodiment, the enhancement of the evidence chain closure strength is the increment of the associated risk closure relative to the candidate sensitive fragment in terms of the causal closure degree of the factual elements of the target sensitive fact; the cross-carrier mutual verification strength is the degree of mutual corroboration of the target sensitive fact by openable output units of different carrier types; and the archival attribute visibility amplification strength is the degree of supplementary restoration of the target sensitive fact by the archival attributes corresponding to the archival attribute nodes.

[0013] In one embodiment, obtaining the jigsaw puzzle path redundancy specifically includes: determining the path reconstruction contribution value based on each jigsaw puzzle reconstruction path formed by output unit nodes, fact element nodes, and associated edges in the jigsaw puzzle reconstruction path set, according to the number of fact elements it covers, the criticality of fact elements, the weight of associated edges, the weight of open visibility, and the difference in carrier type; and determining the jigsaw puzzle path redundancy of candidate sensitive segments based on the number of jigsaw puzzle reconstruction paths, the path reconstruction contribution value, and the residual reconstruction intensity of single-point occlusion.

[0014] In one embodiment, the residual reconstruction intensity of a single-point occlusion is the degree to which the remaining puzzle reconstruction path, after occluding the open output content of any output unit node, enables the factual elements of the target sensitive fact to reach a preset factual completeness threshold.

[0015] In one embodiment, obtaining the puzzle path redundancy further includes: determining the path independence based on the node overlap, associated edge overlap, fact element overlap, and carrier type overlap of any two puzzle reconstruction paths; determining puzzle reconstruction paths where the path independence between each pair is greater than a preset independence threshold and their respective path reconstruction contribution values ​​are greater than a preset contribution threshold as independent puzzle reconstruction paths; and correcting the puzzle path redundancy based on the number of independent puzzle reconstruction paths, the path reconstruction contribution values ​​of independent puzzle reconstruction paths, and the path independence between independent puzzle reconstruction paths.

[0016] In one embodiment, step five includes: determining the collaborative control level based on the fragment association reconstruction gain value and the puzzle path redundancy, and determining the open output content corresponding to the minimum cut-off node set of the puzzle path as the target open output content; determining the desensitization granularity and review path according to the collaborative control level, and performing masking, replacement, generalization or summarization processing on the target open output content to generate a desensitized open copy of the file and a desensitization audit record; wherein, the desensitization audit record includes the risk assessment result, path cut-off result, collaborative control level, desensitization method and review path.

[0017] Compared with the prior art, the present invention has the following beneficial effects: This invention collaboratively processes the archival data to be processed and the corresponding archival metadata, and performs text recognition, layout parsing, structural parsing and fragment segmentation on the archival data to be processed, to obtain a set of fragment analysis objects containing open output units and factual elements. It can include the archival text content, layout position, structural hierarchy and metadata information into the candidate sensitive fragment identification range, avoiding the one-sidedness of identification caused by judging sensitive content based on a single text field, and improving the completeness and stability of sensitive fragment discovery before the archives are opened. Based on the fragment analysis object set and archival metadata, an archival fragment association graph is constructed. Isolated risk calculation, associated risk propagation, and combined reconstruction risk assessment are carried out around candidate sensitive fragments. The risk assessment results include target sensitive facts, isolated risk values, associated risk closure, associated reconstruction risk values, and fragment association reconstruction gain values. It can identify the sensitivity risk of a single openable output unit itself, as well as the associated reconstruction risk formed by multiple openable output units, factual elements, and archival metadata complementing each other, thereby improving the ability to prevent and control sensitive fact scenarios that are pieced together. By further combining the risk assessment results and the archival fragment association map, the candidate sensitive fragments are analyzed for puzzle reconstruction path analysis and path cutting analysis. This yields the puzzle reconstruction path set, puzzle path redundancy, and minimum puzzle path cutting node set. Based on this, the level of collaborative control and the target open output content are determined. The target open output content is then subjected to masking, replacement, generalization, or summarization processing. This approach can reduce the probability of sensitive facts being reconstructed while minimizing unnecessary large-scale desensitization, thus better balancing the utilization value of open archival copies, processing accuracy, and the traceability of desensitization audit records. Attached Figure Description

[0018] Figure 1 This is an overall flowchart of a method for identifying and desensitizing sensitive archival fragments according to the present invention; Figure 2 This is a flowchart illustrating the basic objectification process and candidate sensitive fragment determination of the present invention. Figure 3 This is a flowchart illustrating the generation of associated edge weights in this invention; Figure 4 A flowchart generated based on the risk assessment results of this invention; Figure 5 This is a flowchart generated based on the path cutting results of the present invention. Detailed Implementation

[0019] Reference Figures 1 to 5 A method for identifying and de-identifying sensitive segments in archives, comprising: In this embodiment, all inputs involved in the calculation of weights, risks, and paths are processed using the same dimension. The data items involved in the calculation include the archive data to be processed, archive metadata, open copy parsing results, layout structure, field structure, directory attributes, manual review records, and rule base matching results. The normalization function is denoted as: In the formula, This is the value to be normalized. All coefficients in each weighted summation formula are non-negative, and the sum of coefficients within the same weighted summation formula is 1. Sensitive rule weights, relationship weights, carrier weights, and archive attribute weights participate in the normalization weighted calculation as rule base weights, and are not required to sum to 1 within the entire rule base. When the sum of rule base weights participating in a normalization weighted calculation is 0, the corresponding normalization calculation result is 0. Calculations using the sum of the keyness of factual elements as the denominator are performed; when the denominator is 0, the calculation result is 0. The threshold set is denoted as... The set of weight coefficients is denoted as Threshold set and weight coefficient set The rules for open access review of archives are determined by the archive category rules, historical open access review samples, or manual review configuration tables, and are written into the de-identification audit record. The archive open access review rule base records the correspondence between the target sensitive fact type, fact element set, fact element criticality, fact element relationship set, sensitive rule weight, relationship weight, carrier weight, archive attribute weight, carrier type, associated edge configuration rules, threshold, de-identification granularity, review path, and de-identification method.

[0020] Step 1: Obtain the archive data to be processed and the corresponding archive metadata. Perform basic objectification processing on the archive data to be processed, including text recognition, layout parsing, structure parsing and fragment segmentation, to obtain a fragment analysis object set containing open output units and factual elements. Then, identify candidate sensitive fragments from the open output units in the fragment analysis object set. Transform the original archive data to be processed into a fragment analysis object set with uniform analysis granularity. Text recognition extracts textual content from archives; layout analysis preserves the positional relationships within archive pages; structural analysis extracts the internal hierarchy and field relationships; and fragment segmentation delineates content boundaries suitable for open review. Openly accessible output units correspond to the specific content that may be presented in open copies of the archive, while factual elements correspond to sensitive factual information such as personnel, time, location, events, and relationships within the archive content. Candidate sensitive fragments are determined from openly accessible output units, avoiding reliance solely on isolated word hits for judgment, and enabling the archive content to enter an analysis state involving fragments, factual elements, and metadata.

[0021] Acquire the archive data to be processed and the corresponding archive metadata. The archive data to be processed includes paper scans, electronic documents, image attachments and tabular records. The corresponding archive metadata includes information such as file number, title, creation time, responsible person, retention period, security classification, open status, page range and volume order. When performing text recognition on the data to be processed, the system extracts text from the main body, tables, headers and footers, seals, and annotations. When performing layout analysis on the data to be processed, the system records the position of text blocks, table cells, image areas, and signature areas on the page. When performing structural analysis on the data to be processed, the system identifies the structural positions of titles, paragraphs, table entries, table fields, attachment descriptions, signatures, and approval opinions. When segmenting the data to be processed, the system divides the content into multiple openable output units according to semantic integrity, page boundaries, and structural hierarchy, and extracts factual elements such as personnel names, ID numbers, positions, units, times, locations, events, kinship, and rewards and punishments from each openable output unit. For example, for a cadre appointment and removal approval file, the name column, date of birth column, work experience section, family member table, and approval opinion section can be divided into different open output units, and the personnel identity, appointment time, employing unit and family relationship appearing in them can be recorded as factual elements respectively. A set of fragment analysis objects is generated based on the openable output unit and its factual elements. The fragment analysis object set includes multiple fragment analysis objects. Each fragment analysis object includes an openable output unit and analysis attributes associated with the openable output unit. The open output content in the analysis attributes records the text, table, or image content that the openable output unit intends to present in the open copy of the archive. The carrier type records that the openable output unit comes from the main text, table, picture, seal, annotation, or attachment. The page position records the coordinates of the openable output unit on the page, the page number, and the adjacent content. The structural position records the title level, field name, or volume position of the openable output unit. The factual elements record the sensitive factual information contained in the openable output unit. The open visibility weight is determined based on the presentation completeness, readability, positional prominence, and proximity relationship of the openable output unit to other content in the open copy. The sensitive rule hit result records whether the openable output unit hits sensitive rules such as identity information, resume information, kinship, organizational matters, and disciplinary matters. The candidate tag records whether the openable output unit is identified as a candidate sensitive fragment. Let any openable output unit be denoted as ,and The set of related factual elements is denoted as Open visibility weight is denoted as The completeness of the presentation is recorded as Readability is denoted as Positional salience is denoted as The strength of the proximity relationship is denoted as Presenting completeness according to The proportion of the area of ​​characters, fields, or image regions retained in the open copy relative to the original corresponding content is determined; readability. Determined based on text recognition confidence, structural parsing confidence, or manual input verification results; positional salience. Determined by normalizing the salience level of titles, headers, signatures, annotations, closings, and body paragraphs within the page and structure; proximity strength. according to The minimum page distance and structural distance between the unit and other open output units on the same page, in the same paragraph, in the same table row / column, in the same attachment, or within the same volume are normalized and determined. Open visibility weight. Determine by the following formula: In the formula, All are non-negative weighting coefficients, and .

[0022] In this embodiment, the sensitive rule hit result includes the hit sensitive rule, the matching confidence score, and the hit factual element. The output unit can be made open. The set of sensitive rules that are hit is denoted as Sensitive rules The rule weight is denoted as , sensitive rules The matching confidence is denoted as The intensity of the hit of the sensitive rule is recorded as The strength of the hit rate of sensitive rules. Determine by the following formula: In the formula, For the archive open access review rule base and sensitive rules The corresponding rule weights, Obtained by normalization of results from keyword matching, entity recognition, field matching, page layout matching, or manual review. Fact elements. The degree of importance is recorded as The data is generated by configuring and normalizing the factual elements according to their sensitive fact type, field role, identity orientation, and matter sensitivity, based on the archive open review rule base. Open output units are available. The degree of importance of factual elements is denoted as ,according to The degree of importance of each factual element The average value is determined; When empty, Set to 0. The strength of the file attribute association is denoted as . According to the weight of matching file attributes Matching strength The weighted average is determined; when there are no matching file attributes, Take 0.

[0023] In this embodiment, the candidate determination value is denoted as Determine using the following formula: In the formula, All are non-negative weighting coefficients, and The candidate decision threshold is denoted as... The determination is made by the archive open review rule base or the manual review configuration table; when Not less than At that time, Marked as candidate sensitive fragments, below At that time, no candidate sensitive fragments are marked.

[0024] For example, the name, work unit, and relationship fields in the family member table are only partial information when viewed individually, but when combined with the cadre's name, work unit, and formation time, the family relationship of a specific person can be restored. Therefore, the open visibility weight of this open output unit and the result of the sensitive rule are jointly written into the corresponding fragment analysis object, and enter the candidate sensitive fragment range according to the candidate tag.

[0025] Step two involves constructing an archival fragment association graph based on the set of fragment analysis objects and archival metadata. The content, structural, and attribute relationships between the fragment analysis object set and the archival metadata are incorporated into the archival fragment association graph. This graph establishes connections between open output units, factual elements, and relevant information in the archival metadata, demonstrating co-occurrence, complementarity, and mutual corroboration relationships between different fragments. For cases where a single fragment may seem insensitive, but the combination of multiple fragments is sufficient to reconstruct sensitive facts, the archival fragment association graph can provide a more complete basis for association analysis.

[0026] When constructing an archival fragment association graph based on the fragment analysis object set and archival metadata, a corresponding output unit node is established for each openable output unit in the fragment analysis object set, a corresponding fact element node is established for each fact element associated with the openable output unit, and a corresponding archival attribute node is established for archival attributes contained in the archival metadata, such as file number, title, creation time, responsible person, retention period, security classification, open status, page range, and volume order. Association edges are configured between output unit nodes, fact element nodes, and archival attribute nodes. The configuration of association edges is based on the direct inclusion relationship between openable output content and fact elements, the semantic inheritance relationship between different openable output units, the limiting relationship between fact elements and archival attributes, and the correspondence between time, personnel, location, unit, and behavior within the same archival matter. For example, in a cadre appointment and removal approval file, a work experience segment corresponds to an output unit node, which includes the employing unit, the time of employment, and the position, which correspond to fact element nodes respectively. The formation time and the responsible person in the file metadata correspond to file attribute nodes respectively. When the employing unit and the responsible person are the same, and the time of employment and the time of formation are temporally adjacent, an association edge is set between the corresponding graph nodes, so that the file fragment association graph can record the multi-source association between fragment content, fact elements, and file attributes. Co-occurrence, complementarity, and mutual verification matching are performed on the open output content, factual elements, or archival attributes corresponding to the nodes at both ends of the related edge. The contribution of the related edge to the reconstruction of sensitive facts is calculated, and the contribution of the sensitive fact reconstruction is used as the weight of the related edge. Among them, co-occurrence matching determines the relationship strength based on the common appearance of people, units, times, and places in the same page, the same paragraph, the same table row or column, or the same volume. Complementarity matching determines the keyness of factual elements based on the degree of supplementation of identity provided by one end of the graph node and time, place, or event content provided by the other end of the graph node. Mutual verification matching determines the enhancement coefficient based on the consistency of the same target sensitive fact in different carrier types such as text, tables, seals, annotations, and attachments. The weight of the related edge is generated by weighted normalization of relationship strength, keyness of factual elements, open visibility weight, and enhancement coefficient. The archival fragment association map is denoted as... ,in For a set of graph nodes, Let be the set of associated edges. The two graph nodes connected by any associated edge are denoted as . and The weight of the associated edge is denoted as .node The set of factual elements carried or connected is denoted as ,node The set of factual elements carried or connected is denoted as The co-occurrence intensity is denoted as The degree of complementarity is denoted as The mutual verification strength is denoted as The strength of the relationship is denoted as Relationship strength Determine by the following formula: In the formula, All are non-negative weighting coefficients, and Co-occurrence intensity Determined by normalization of the co-occurrence relationship of the two nodes in the same page, paragraph, table row / column, attachment, or volume; degree of complementarity. The strength of mutual verification is determined by the proportion of newly added factual elements covering the same sensitive fact at both ends of the node. The consistency verification relationship of the same person, time, unit, matter or relationship is determined by different carrier types through normalization.

[0027] The criticality of the factual elements at both ends of the associated edge is denoted as The keyness is determined by the highest value among the factual elements carried or connected by the two endpoints; when neither endpoint carries or connects any factual elements... Set to 0. Node and nodes The visible mean at both ends is denoted as The visibility value is determined by the average of the visibility values ​​of the two endpoints. The visibility value of the output unit node is taken as the open visibility weight of the corresponding openable output unit. The visibility value of the fact element node is taken as the average of the open visibility weights of the openable output units containing the fact element. The visibility value of the file attribute node is taken as the normalized value of the degree to which the file attribute can be read in open copies or open directories.

[0028] The enhancement coefficient is denoted as .node The carrier type is denoted as ,node The carrier type is denoted as Conflicts in factual elements are marked as follows: When node With nodes When the values ​​of the same type of factual elements contradict each other, ,otherwise Enhancement coefficient Determine by the following formula: Related edge weights Determine by the following formula: In the formula, All are non-negative weighting coefficients, and .

[0029] For example, if the open output content contains a person's name and employer, the file attributes contain the corresponding responsible person and creation time, and the appendix annotations record the approval opinions, the three elements, after co-occurrence, complementarity, and mutual verification matching, will result in a high contribution to the reconstruction of sensitive facts. This contribution is written into the associated edge weight to facilitate a detailed assessment of the associated reconstruction risk of candidate sensitive fragments.

[0030] Step 3: Based on the archival fragment association graph, perform isolated risk calculation, associated risk propagation, and combined reconstruction risk assessment on candidate sensitive fragments to obtain the corresponding risk assessment results. These results include the target sensitive fact, isolated risk value, associated risk closure, associated reconstruction risk value, and fragment associated reconstruction gain value. Isolated risk calculation examines the degree to which the candidate sensitive fragment itself exposes the target sensitive fact. Associated risk propagation examines the risk diffusion range of the candidate sensitive fragment after it is connected to other information through the archival fragment association graph. Combined reconstruction risk assessment examines the ability to reconstruct the target sensitive fact when multiple openable output units, factual elements, and archival metadata work together. By using target sensitive facts, isolated risk values, associated risk closures, associated reconstruction risk values, and fragment associated reconstruction gain values, we can distinguish between single-point exposure risks and associated reconstruction risks, and avoid underestimating the sensitive risks generated by fragment combinations.

[0031] Based on the archival fragment association graph, isolation risk calculation and association risk propagation are carried out on candidate sensitive fragments. The open output content, factual elements, open visibility weight, sensitive rule hit results, and graph nodes and associated edge weights connected to the candidate sensitive fragments are read. According to the completeness of the identity information, resume information, organizational matters, kinship, disciplinary matters, etc. contained in the candidate sensitive fragments themselves, the isolation risk value of the candidate sensitive fragments exposing the target sensitive facts alone is calculated. During the process of related risk propagation, starting from the output unit node corresponding to the candidate sensitive fragment, the process extends along the related edge to the fact element node and the archive attribute node. Graph nodes whose related edge weight exceeds the preset propagation threshold, whose fact element criticality reaches the preset criticality threshold, and whose open visibility weight reaches the preset visibility threshold are included in the related risk closure. The related reconstruction risk value is determined based on the coverage of personnel, time, location, unit, matter, and relationship by the graph nodes within the related risk closure. The target sensitive facts are recorded as The target involves sensitive facts The corresponding set of factual elements is denoted as The target involves sensitive facts. The sensitive rules matched by the candidate sensitive fragments, and the set of factual elements contained in the candidate sensitive fragments. And the determination of fact type matching relationships in the archive open access review rule base; when the same candidate sensitive fragment matches multiple target sensitive facts, the risk value of each target sensitive fact is calculated separately and the corresponding risk assessment results are retained. Unless otherwise specified, candidate sensitive fragments The associated reconstruction risk value, fragment associated reconstruction gain value, puzzle path redundancy, minimum cut-off node set of the puzzle path, and collaborative control value are all within the scope of the current target sensitive facts. The following calculation is performed. The same candidate sensitive fragment... When dealing with multiple sensitive facts related to different objectives, each sensitive fact generates a risk assessment result, a pathway disruption result, and a collaborative control level. For any given object... The set of factual elements it covers is denoted as object Sensitive facts related to the target The completeness of factual elements is denoted as Determine using the following formula: Candidate Sensitive Fragments The isolated risk value is denoted as Determine using the following formula: In the formula, for Open visibility weight, for The strength of the hit of sensitive rules, for Sensitive facts related to the target The completeness of factual elements.

[0032] Candidate Sensitive Fragments The corresponding output unit node is denoted as , No. The set of closure nodes obtained by round expansion is denoted as The maximum expansion depth is denoted as The associated risk closure is denoted as The initial set of closure nodes is .when At that time, the first The set of nodes in the round closure is determined by the following formula: In the formula, To propagate the edge weight threshold, The threshold for the criticality of node factual elements. The node visibility threshold, For nodes The visible value of the node. Node The importance of factual elements By node Determine the highest value among the criticality of the factual elements carried or connected; Node When no factual elements are carried and not connected, Set to 0. The completion time is no more than [number missing]. The associated risk closure is obtained after round expansion. .

[0033] Related risk closure Sensitive facts related to the target The completeness of factual elements is denoted as ,according to Using the same calculation method, derived from the associated risk closure The set of factual elements covered is determined. The strength of the chain of evidence closure is denoted as... , by associated risk closure The set of factual elements covered and the target sensitive facts The matching ratio between the factual element relationship sets is determined, and then sorted by relationship weight. The weighted normalization yields the result. The cross-carrier mutual verification strength is denoted as... , by associated risk closure The set of carrier types covered and the target sensitive facts The matching ratio between the associated carrier type sets is determined, and then weighted according to the carrier. Obtained by weighted normalization. The magnification intensity of the visible file attributes is denoted as... , by associated risk closure The set of covered archival attributes and the target sensitive facts The matching ratio between the associated sets of archival attributes is determined, and then weighted according to the archival attributes. The result is obtained through weighted normalization. When the target involves sensitive facts... When the set of factual element relationships, the set of associated carrier types, or the set of associated archive attributes are not configured, or when the sum of the weights in the corresponding weighted normalization calculation is 0, or The corresponding strength is set to 0. The average edge weight of the closure is denoted as... The weight is determined by the average weight of the associated edges within the closure; if there are no associated edges within the closure, Set to 0. The risk value for associated reconstruction is denoted as . Determine using the following formula: In the formula, All are non-negative weighting coefficients, and .

[0034] The fragment association reconstruction gain value is denoted as The positive normalization improvement of the associated reconstruction risk value relative to the isolated risk value is denoted as... The value is determined by dividing the positive part of the reconstructed risk value minus the isolated risk value by 1 and then dividing it by the difference between the isolated risk value and 1; when the difference between 1 and the isolated risk value is 0, Set to 0. The positive increase in the completeness of the closure fact element relative to the completeness of the candidate sensitive fragment fact element is denoted as . The value is determined by the positive part of the difference between the two. The positive increase in the closure strength of the closure evidence chain relative to the closure strength of the candidate sensitive fragment alone covering the evidence chain is denoted as... The strength of the evidence chain closure for a candidate sensitive fragment alone is determined by the positive part of the difference between the two. Using the same calculation method, based on candidate sensitive fragments Individually covered factual element relationship set and target sensitive facts Determining the matching ratio between sets of factual element relationships; target sensitive facts When no fact element relationship set is configured or the sum of corresponding relationship weights is 0, this strength is 0. Fragment association reconstruction gain value. Determine by the following formula: In the formula, All are non-negative weighting coefficients, and .

[0035] For example, an open output unit may only record that a person once held a certain position, and the isolated risk value is low. However, after the open output unit is connected to the formation time, responsible person, approval opinion and family member table through the file fragment association graph, the background of the person's appointment and dismissal and the relationship of related persons can be restored. This range is written into the associated risk closure. A combined reconstruction risk assessment is conducted on the openable output units, fact element nodes, and archive attribute nodes within the associated risk closure to determine the target sensitive facts, the associated reconstruction risk value, and the fragment associated reconstruction gain value. The fragment association reconstruction gain value is obtained by fusing the normalized increase of the association reconstruction risk value relative to the isolated risk value, the increase of the completeness of the factual elements of the target sensitive fact, the increase of the evidence chain closure strength, the cross-carrier mutual verification strength, and the visible amplification strength of the archive attributes. Among them, the increase of the completeness of the factual elements is determined based on the increment of the association risk closure relative to the candidate sensitive fragment in the coverage of the factual elements of the target sensitive fact; the increase of the evidence chain closure strength is determined based on the increment of the association risk closure relative to the candidate sensitive fragment in the causal closure of the factual elements of the target sensitive fact; the cross-carrier mutual verification strength is determined based on the degree of mutual corroboration of the target sensitive fact by open output units of different carrier types such as text, tables, pictures, seals, annotations or attachments; and the visible amplification strength of the archive attributes is determined based on the degree of supplementation and restoration of the target sensitive fact by the archive attributes corresponding to the archive attribute nodes. For example, the main text paragraphs provide the job title, the table fields provide the personnel identity, the seal area provides the responsible unit, and the archive metadata provides the formation time. When the four types of information together close the personnel, unit, time, and behavioral elements of the target sensitive fact, the fragment association reconstruction gain value increases accordingly. Together with the target sensitive fact, isolated risk value, associated risk closure, and associated reconstruction risk value, they form the risk assessment result corresponding to the candidate sensitive fragment.

[0036] Step four involves performing jigsaw puzzle reconstruction path analysis and path segmentation analysis on candidate sensitive fragments based on the risk assessment results and the archival fragment association map. This yields path segmentation results for each candidate sensitive fragment. The path segmentation results include a jigsaw puzzle reconstruction path set, jigsaw puzzle path redundancy, and a set of minimum segmentation nodes. The jigsaw puzzle reconstruction path analysis identifies the reconstruction chains formed between candidate sensitive fragments and other fragments and factual elements, resulting in a jigsaw puzzle reconstruction path set. The jigsaw puzzle path redundancy reflects the degree to which other paths still exist to reconstruct the target sensitive fact after partially obscuring content. The path segmentation analysis identifies the set of minimum segmentation nodes that have a critical impact on blocking reconstruction paths. This approach not only focuses on whether a candidate sensitive fragment itself is sensitive but also on whether the remaining information in the open copy can still be pieced together to reconstruct the target sensitive fact.

[0037] Based on the risk assessment results and the archival fragment association graph, a jigsaw puzzle reconstruction path analysis is performed on the candidate sensitive fragments. The target sensitive facts, associated risk closures, associated reconstruction risk values, and fragment association reconstruction gain values ​​in the risk assessment results are read. Taking the output unit node corresponding to the candidate sensitive fragment as the starting point of the path, the fact element nodes that can supplement the target sensitive facts, the output unit nodes with high open visibility weights, and the associated edges whose associated edge weights reach the preset edge weight thresholds are included in the path search range, forming a jigsaw puzzle reconstruction path set composed of output unit nodes, fact element nodes, and associated edges. For each puzzle reconstruction path, the path reconstruction contribution value is calculated according to the number of factual elements it covers, the criticality of factual elements, the weight of related edges, the weight of open visibility, and the difference in carrier type. The number of factual elements reflects the amount of personnel, time, place, unit, matter, relationship, etc. in the sensitive facts of the target covered by the path. The criticality of factual elements reflects the sensitivity of the core identity, core behavior or core relationship in the path. The difference in carrier type reflects the degree of mutual support between different carriers such as text, tables, seals, annotations, and attachments. Candidate Sensitive Fragments The corresponding set of puzzle reconstruction paths is denoted as Any puzzle reconstruction path is denoted as... ,in , The path length is, and .path The set of included nodes is denoted as The set of associated edges is denoted as The set of output unit nodes is denoted as The set of factual elements is denoted as And record the carrier types contained in the path. The path edge weight threshold is denoted as... Puzzle reconstruction path set Starting from The node is located in the associated risk closure. There are related edges between adjacent nodes, and the weight of related edges within the path is not less than 1. The path consists of a set of factual elements of the path and a set of sensitive factual elements of the target that intersect.

[0038] path Sensitive facts related to the target The contribution of factual elements coverage is recorded as The key level is determined by the ratio of the sum of the keyness of the factual elements covered by the path to the sum of the keyness of the target factual elements; when the sum of the keyness of the target factual elements is 0, Set to 0. The average edge weight of the path is denoted as... The average open visibility weight of a path is determined by the average weight of the associated edges within the path; it is 0 when there are no associated edges in the path. The average open visibility weight of a path is denoted as... The value is determined by the average of the open visibility weights corresponding to the output unit nodes within the path; it is set to 0 when there are no output unit nodes. The path carrier type difference is denoted as... The contribution value is determined by the ratio between the number of carrier types included in the path and the number of carrier types involving sensitive facts related to the target in the closure; a value of 0 is taken when the denominator is 0. The path reconstruction contribution value is denoted as... Determine using the following formula: In the formula, All are non-negative weighting coefficients, and .

[0039] Any two puzzle reconstruction paths are denoted as and .path With path The overlap of nodes, the overlap of associated edges, the overlap of factual elements, and the overlap of carrier types are respectively denoted as... The degree of overlap is determined by the ratio of the size of the intersection to the size of the union of the corresponding sets, with 0 taken when the union is empty. The path independence is denoted as... Determine using the following formula: In the formula, All are non-negative weighting coefficients, and The set of independent puzzle reconstruction paths is denoted as... Path reconstruction contribution value Scan from high to low; when the path to be included The path reconstruction contribution value is not lower than the contribution threshold. And with Each path already included The path independence is not lower than the independence threshold. At that time, Included .

[0040] The set of high-contribution puzzle reconstruction paths is denoted as The contribution value of path reconstruction is no less than Furthermore, the contribution of factual element coverage is not less than the factual completeness threshold. The path composition. The residual reconstruction intensity of single-point occlusion is denoted as... After the output unit nodes covered by the path reconstructed by the high-contribution puzzle are occluded, the remaining path still reaches the factual completeness threshold. The factual element completeness is determined by mean normalization; when there are no maskable output unit nodes. Take 0.

[0041] The normalization benchmark for the number of paths is denoted as ,and It is a positive number. The normalized value of the number of independent paths is denoted as... The number of paths reconstructed based on independent puzzle pieces and The proportion is determined by normalization. The average contribution value of the independent puzzle reconstruction path is denoted as... The independence is determined by the average contribution value of each independent puzzle reconstruction path; it is set to 0 when there is no independent puzzle reconstruction path. The average independence of each independent puzzle reconstruction path is denoted as... The redundancy is determined by the average path independence among the independent puzzle reconstruction paths; a value of 0 is used when the number of independent puzzle reconstruction paths is less than or equal to 1. Puzzle path redundancy is denoted as... Determine using the following formula: In the formula, All are non-negative weighting coefficients, and ; The classification rules or manual review configuration table shall determine and be written into the anonymization audit record.

[0042] For example, a candidate sensitive fragment only shows the job title, another table output cell node shows the person's name and employing unit, the file attribute node shows the creation time, and the annotation area shows the approval opinions. After the above nodes are connected by related edges, a puzzle reconstruction path is formed. This path can fill in the personnel, unit, time and job title in the target sensitive fact. The path reconstruction contribution value is rated as high. Based on the puzzle reconstruction path set, path cutting analysis is carried out and the path cutting results are determined. When calculating the puzzle path redundancy, the number of puzzle reconstruction paths, the path reconstruction contribution value of each puzzle reconstruction path, and the residual reconstruction intensity of single-point occlusion are counted. The residual reconstruction intensity of single-point occlusion is determined according to the degree to which the remaining puzzle reconstruction path still enables the factual elements of the target sensitive fact to reach the preset factual completeness threshold after occluding the open output content of any output unit node. If the family member table, the responsible person's file attributes, and the approval opinion annotations can still restore the target sensitive fact after occluding the job position paragraph, then the candidate sensitive segment has a high puzzle path redundancy. Based on this, the node overlap, associated edge overlap, fact element overlap, and carrier type overlap of any two puzzle reconstruction paths are compared to calculate the path independence. Puzzle reconstruction paths with a path independence greater than a preset independence threshold and a path reconstruction contribution value greater than a preset contribution threshold are identified as independent puzzle reconstruction paths. The puzzle path redundancy is corrected based on the number of independent puzzle reconstruction paths, the path reconstruction contribution value of independent puzzle reconstruction paths, and the path independence between independent puzzle reconstruction paths. Path cutoff analysis in high-contribution jigsaw puzzle reconstruction path sets The process is performed above. The set of candidate cutoff nodes is denoted as... , Reconstructing the path set from high-contribution puzzle pieces It consists of output unit nodes and fact element nodes that have open output content mapping relationships. Any candidate cut-off node is denoted as... ,node The loss from open use is denoted as The cost of desensitization processing is denoted as The audit load is recorded as Cut-off cost is denoted as Losses due to open access By node The impact of the processed open output content of the mapping on field integrity, semantic continuity, page layout integrity, and the necessity of archival use is normalized and determined; the cost of desensitization processing. Determined by normalizing the processing complexity based on masking, replacement, generalization, or summarization; audit workload. The cost is determined by normalization based on manual review requirements, the degree of dispute, and the complexity of the review process. Determine by the following formula: In the formula, All are non-negative weighting coefficients, and .

[0043] When the set of high-contribution puzzle reconstruction paths is empty, or the set of candidate cut nodes is empty, the minimum cut node set of the puzzle path is recorded as an empty set, the target open output content set is recorded as an empty set, and the corresponding empty set status is recorded in the path cutting results; candidate sensitive fragments The risk value of associated reconstruction, the gain value of fragment associated reconstruction, and the redundancy of the jigsaw puzzle path are still used to determine the collaborative control level. This process does not change the risk assessment results of candidate sensitive fragments.

[0044] Candidate cut-off nodes The variable for whether a node is selected into the minimum cut node set of the puzzle path is denoted as . , It is a binary variable. The minimum cut node set of the jigsaw puzzle path is determined according to... The objective function for solving the coverage optimization problem is: Reconstruct the path for any high-contribution puzzle piece. Set the following constraints: After solving, The candidate cut nodes form the minimum cut node set of the puzzle path. When multiple solutions have the same objective function value, the solution with the smaller total open utilization loss is selected; if the total open utilization loss is the same, the solution with fewer nodes is selected. Output unit nodes are mapped to their corresponding open output content; fact element nodes are mapped to the open output content that contains that fact element and participates in the high-contribution puzzle reconstruction path. The open output content obtained from mapping each node in the target open output content set is composed of the open output content obtained from mapping each node in the target open output content set. The resulting set of puzzle reconstruction paths Puzzle path redundancy Minimum cut node set of the puzzle path and target open output content set Jointly written into candidate sensitive fragments The corresponding path cutting results.

[0045] Step 5: Based on the risk assessment results and path cutting-off results, determine the collaborative control level and target open output content of the candidate sensitive segments, and desensitize the target open output content according to the collaborative control level, generating an open copy of the desensitized file and desensitization audit records. This implements the risk assessment results and path cutting-off results into specific desensitization processes. The level of coordinated control is determined based on the risk intensity, reconstructive capability, and puzzle reconstruction path of candidate sensitive fragments, reflecting the appropriate processing intensity for different candidate sensitive fragments. The target open output content corresponds to the specific open output content requiring de-identification, avoiding indiscriminate processing of all archive content. After de-identification is implemented according to the coordinated control level, a de-identified open copy of the archive is generated; simultaneously, a de-identification audit record is created, documenting the risk assessment results, path cutting results, coordinated control level, and corresponding de-identification processing details, facilitating verification of the de-identification basis and process by reviewers.

[0046] Based on the risk assessment results and path cutting results, the collaborative control level of candidate sensitive fragments is determined, and the open output content corresponding to the minimum cutting node set of the jigsaw puzzle path is determined as the target open output content. When determining the collaborative control level, the target sensitive facts, isolated risk values, associated risk closures, associated reconstruction risk values, and fragment associated reconstruction gain values ​​in the risk assessment results are read. At the same time, the jigsaw puzzle reconstruction path set, jigsaw puzzle path redundancy, and jigsaw puzzle path minimum cutting node set in the path cutting results are read. The fragment associated reconstruction gain value and jigsaw puzzle path redundancy are jointly judged. Candidate sensitive fragments with higher fragment associated reconstruction gain values ​​and higher jigsaw puzzle path redundancy are classified into higher collaborative control levels, and candidate sensitive fragments with lower fragment associated reconstruction gain values ​​and lower jigsaw puzzle path redundancy are classified into lower collaborative control levels. Candidate Sensitive Fragments The coordinated regulation value is denoted as The fragment association reconstruction gain value is denoted as The redundancy of the puzzle path is denoted as Coordinated regulation value Determine by the following formula: In the formula, and All are non-negative weighting coefficients, and .

[0047] The threshold for coordinated regulation level is denoted as ,and .when Below At that time, candidate sensitive fragments Into ;when Not less than And lower than At that time, candidate sensitive fragments Into ;when Not less than And lower than At that time, candidate sensitive fragments Into ;when Not less than At that time, candidate sensitive fragments Into .in, These are the collaborative control levels, from low to high. The desensitization rule base configures the corresponding desensitization granularity and review path for each collaborative control level; the desensitization granularity includes field or phrase level, fact element level, multi-fact element combination level, and summary level; the review path includes rule verification, sampling review, manual review, and manual review and confirmation.

[0048] Desensitized audit records document sensitive facts related to the target. Isolated risk value Related risk closure Risk value of related reconstruction Fragment association reconstruction gain value Puzzle reconstruction path set Puzzle path redundancy Minimum cut node set of the puzzle path Target open output content set Coordinated regulation levels and threshold sets and weight coefficient set The desensitization audit record also records the desensitization method, processing time, review path, and manual confirmation result for the target open output content. When the same target open output content is targeted by multiple candidate sensitive fragments, the desensitization method is determined according to the desensitization rule corresponding to the higher level of collaborative control, and the corresponding candidate sensitive fragment and path cutting result are recorded in the desensitization audit record.

[0049] For example, in an open copy of a file, the main text paragraphs only show a person's work experience, the table area shows their family member relationships, the file attributes include the creation time and the responsible person, and the annotation area contains approval opinions. If the minimum cut node set of the puzzle path points to the personnel relationship field in the table area and the approval opinions in the annotation area, then the open output content corresponding to the above personnel relationship field and approval opinions will be determined as the target open output content. The anonymization granularity and review path are determined according to the level of collaborative control. The target open output content is subjected to masking, replacement, generalization, or summarization processing, generating an open copy of the anonymized archive and an anonymization audit record. At higher levels of collaborative control, the anonymization granularity is refined to the level of factual elements such as name, ID number, unit name, specific position, precise date, kinship terminology, and approval conclusion. The review path is configured to combine automatic result review with manual verification. Content that directly closes the target's sensitive facts is masked; unit, time, or position content that still needs to retain archival value is replaced or generalized; and content where multiple factual elements support each other in a narrative is summarized. At lower levels of collaborative control, the anonymization granularity remains at the field or phrase level, and the review path primarily uses rule verification and sampling review. For example, specific names are masked with unidentifiable characters, precise birth dates are generalized to a range of years, specific employing units are replaced with units of the same level, and approval opinions containing complete appointment and dismissal processes are rewritten as a summary of matters without identity references. After anonymization, the open copy of the document retains the document structure, page order, and necessary open information. The anonymization audit record simultaneously records the risk assessment results, path cutting results, collaborative control levels, anonymization methods, and audit paths, facilitating traceability of the processing basis for each target open output content.

[0050] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for identifying and desensitizing sensitive segments in archives, characterized in that, include: Step 1: Obtain the archive data to be processed and the corresponding archive metadata. Perform basic objectification processing on the archive data to be processed, including text recognition, layout parsing, structure parsing and fragment segmentation, to obtain a fragment analysis object set containing open output units and factual elements, and determine candidate sensitive fragments from the open output units in the fragment analysis object set. Step 2: Based on the set of fragment analysis objects and archival metadata, construct an archival fragment association map; Step 3: Based on the archival fragment association map, perform isolated risk calculation, associated risk propagation, and combined reconstruction risk assessment on candidate sensitive fragments to obtain the risk assessment results corresponding to the candidate sensitive fragments; among which, the risk assessment results include the target sensitive fact, isolated risk value, associated risk closure, associated reconstruction risk value, and fragment association reconstruction gain value; Step 4: Based on the risk assessment results and the archival fragment association map, perform jigsaw puzzle reconstruction path analysis and path cutting analysis on the candidate sensitive fragments to obtain the path cutting results corresponding to the candidate sensitive fragments; among which, the path cutting results include the jigsaw puzzle reconstruction path set, jigsaw puzzle path redundancy, and jigsaw puzzle path minimum cutting node set; Step 5: Based on the risk assessment results and path cutting results, determine the collaborative control level and target open output content of the candidate sensitive fragments, and desensitize the target open output content according to the collaborative control level, generating an open copy of the desensitized file and desensitization audit records.

2. The method for identifying and desensitizing sensitive archival fragments according to claim 1, characterized in that, The fragment analysis object set includes multiple fragment analysis objects. Each fragment analysis object includes an openable output unit and analysis attributes associated with the openable output unit. The analysis attributes include open output content, carrier type, page position, structural position, factual elements, open visibility weight, sensitive rule hit results, and candidate tags.

3. The method for identifying and desensitizing sensitive archival fragments according to claim 2, characterized in that, The archival fragment association graph includes graph nodes, association edges, and association edge weights. Graph nodes include output unit nodes corresponding to open output units, fact element nodes corresponding to fact elements, and archival attribute nodes corresponding to archival attributes in archival metadata. Archival attributes are information items contained in archival metadata. Association edges connect graph nodes, and association edge weights represent the contribution of association edges to the reconstruction of sensitive facts.

4. The method for identifying and desensitizing sensitive archival fragments according to claim 3, characterized in that, The contribution of sensitive fact reconstruction is obtained by matching the co-occurrence, complementarity and mutual verification of the open output content, fact elements or archive attributes corresponding to the graph nodes at both ends of the associated edge to obtain the relationship strength and the criticality of the fact elements. This is then combined with the open visibility weight and weighted normalization to generate the associated edge weight.

5. The method for identifying and desensitizing sensitive archival fragments according to claim 4, characterized in that, The fragment association reconstruction gain value is determined as follows: based on the normalized increase of the association reconstruction risk value relative to the isolated risk value, the increase of the factual element completeness of the target sensitive fact, the increase of the evidence chain closure strength, the cross-carrier mutual evidence strength, and the amplification strength of the archive attribute visibility, the fragment association reconstruction gain value of the candidate sensitive fragment is determined by fusion; among which, the increase of factual element completeness is the increment of the association risk closure relative to the candidate sensitive fragment in the degree of factual element coverage of the target sensitive fact.

6. The method for identifying and desensitizing sensitive archival fragments according to claim 5, characterized in that, The increase in the strength of the evidence chain closure is the increment of the degree of causal closure of the factual elements of the target sensitive fact relative to the candidate sensitive fragment; the cross-carrier mutual verification strength is the degree of mutual corroboration of the target sensitive fact by open output units of different carrier types; the archival attribute visibility amplification strength is the degree of supplementation and restoration of the target sensitive fact by the archival attributes corresponding to the archival attribute nodes.

7. The method for identifying and desensitizing sensitive archival fragments according to claim 1, characterized in that, The acquisition of puzzle path redundancy specifically includes: based on each puzzle reconstruction path formed by output unit nodes, fact element nodes and associated edges in the puzzle reconstruction path set, determining the path reconstruction contribution value according to the number of fact elements covered, the criticality of fact elements, the weight of associated edges, the weight of open visibility and the difference in carrier type; and determining the puzzle path redundancy of candidate sensitive segments based on the number of puzzle reconstruction paths, the path reconstruction contribution value and the residual reconstruction intensity of single-point occlusion.

8. The method for identifying and desensitizing sensitive archival fragments according to claim 7, characterized in that, The residual reconstruction intensity of a single-point occlusion is the degree to which the remaining puzzle reconstruction path, after occluding the open output content of any output unit node, enables the factual elements of the target sensitive fact to reach a preset factual completeness threshold.

9. The method for identifying and desensitizing sensitive archival fragments according to claim 8, characterized in that, The acquisition of puzzle path redundancy also includes: determining path independence based on the node overlap, associated edge overlap, fact element overlap, and carrier type overlap of any two puzzle reconstruction paths; identifying puzzle reconstruction paths where the path independence between each pair is greater than a preset independence threshold and their respective path reconstruction contribution values ​​are greater than a preset contribution threshold as independent puzzle reconstruction paths; and correcting puzzle path redundancy based on the number of independent puzzle reconstruction paths, the path reconstruction contribution values ​​of independent puzzle reconstruction paths, and the path independence between independent puzzle reconstruction paths.

10. A method for identifying and desensitizing sensitive archival fragments according to any one of claims 1-9, characterized in that, Step five includes: determining the collaborative control level based on the fragment association reconstruction gain value and the puzzle path redundancy, and identifying the open output content corresponding to the minimum cut-off node set of the puzzle path as the target open output content; determining the desensitization granularity and review path according to the collaborative control level, and performing masking, replacement, generalization, or summarization processing on the target open output content to generate a desensitized open copy of the archive and a desensitization audit record; wherein, the desensitization audit record includes the risk assessment results, path cut-off results, collaborative control level, desensitization method, and review path.

Images