Sensitive information detection methods, devices, storage media and computer equipment
By constructing a bibliographic association graph and performing semantic processing, the problem of identifying and distinguishing the sources of multi-field combined leakage paths was solved, enabling accurate identification and tracing of combined leakage paths and ensuring the reliability and traceability of detection results.
Patent Information
- Application Number
- CN202610912566.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-06-24
- Publication Date
- 2026-11-10
AI Technical Summary
Existing technologies cannot identify combined leakage paths formed by the association of multiple fields or records, and cannot distinguish between public and restricted sources of the same semantic content, making it difficult to trace leakage paths.
By acquiring bibliographic metadata and performing semantic processing to generate document semantic units, a bibliographic association graph is constructed. Based on co-occurrence relationships and field attribution relationships, semantic indexes of different source types are matched, source status markers are determined, and leakage path detection is performed.
It enables accurate identification and tracing of combined leakage paths, reduces the confusion between publicly available and restricted source content in external detection, and ensures the traceability and accuracy of detection results.
Smart Images

Figure CN123069409A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of data security technology, and in particular to a method, apparatus, storage medium, and computer equipment for detecting sensitive information. Background Technology
[0002] In document service platforms, institutional knowledge bases, and scientific research management systems, users often export and send out bibliographic metadata data packages containing fields such as title, author, affiliation, funding, and abstract in batches due to business needs such as retrieval and citation, and results aggregation. Based on this, existing data leakage prevention detection methods mainly rely on sensitive word libraries, regular expressions, file fingerprint comparison, or generic named entity recognition technologies to perform independent detection of explicit sensitive objects such as ID numbers and classified keywords in single-field and single-text format.
[0003] However, existing item-by-item detection methods are insufficient to address the risk of combined leakage in bibliographic data dissemination scenarios. Specifically, while individual academic fields may not be sensitive, multiple fields or records combined through co-occurrence relationships such as authors, institutions, and projects may implicitly reveal sensitive clues such as research layout, unpublished results, or team collaborations. Furthermore, the same semantic content may coexist in both publicly available documents and internally restricted materials, and existing methods cannot distinguish the source, which can easily lead to misjudgments. In addition, the detection records lack a unified record of external dissemination behavior, source status, and field associations, making it difficult to trace the specific triggering evidence and leakage path afterward. Summary of the Invention
[0004] In view of this, this application provides a sensitive information detection method, apparatus, storage medium and computer equipment, the main purpose of which is to solve the technical problem that the prior art cannot identify the combined leakage path formed by the association of multiple fields or multiple records, and cannot distinguish between the public and restricted sources of the same semantic content, which makes it difficult to trace the leakage path.
[0005] According to a first aspect of the present invention, a method for detecting sensitive information is provided, comprising: Obtain the bibliographic metadata data package to be sent out and the corresponding sending context information, and perform semantic processing on the document fields in the bibliographic metadata data package to generate document semantic units; Based on the co-occurrence relationship of the document semantic units, the field affiliation relationship of the document semantic units, and the external context information, a bibliographic association graph is constructed; The document semantic unit is matched with semantic indexes of different source types, and based on the matching results and the outgoing context information, the source status marker of the document semantic unit at the corresponding node in the bibliographic association graph is determined, wherein the source status marker includes a restricted status marker; Leakage path detection is performed on the bibliographic association graph based on the source status marker. When there is a combined leakage path in the bibliographic association graph, an outward control result and detection record are generated based on the combined leakage path. The combined leakage path is a path in the bibliographic association graph that contains the restricted status marker and spans at least two of the document fields.
[0006] Optionally, the document fields include core content description information, process event description information, output attribution entity information, and field role identification information; the semantic processing of the document fields in the bibliographic metadata to generate document semantic units includes: semantic extraction of the core content description information and the process event description information to generate topic semantic information and event elements respectively; entity disambiguation of the output attribution entity information to generate entity identifiers; determining field roles based on the field role identification information, and generating document semantic units according to the topic semantic information, the event elements, the entity identifiers, and the field roles.
[0007] Optionally, the step of performing combined leakage path detection on the bibliographic association graph based on the source status marker includes: selecting a state node with the restricted status marker and a behavior node representing outward behavior in the bibliographic association graph as a search starting point based on the source status marker; traversing the connection edges in the bibliographic association graph starting from the search starting point to obtain candidate paths, and calculating the number of fields and the number of document records covered by the candidate paths; when there is a candidate path that covers at least two fields or document records, and there is a state node with the restricted status marker on the candidate path, the corresponding candidate path is determined as a combined leakage path.
[0008] Optionally, the method further includes: when the combined leakage path simultaneously contains nodes generated based on the author's affiliation field mapping, nodes generated based on the funding project field mapping, and nodes generated based on the title or keyword field mapping, the combined leakage path is labeled as a first sensitive category; when the combined leakage path simultaneously contains nodes generated based on the event element mapping in the abstract field, nodes generated based on the publication status field mapping, and status nodes based on the restricted status marker, the combined leakage path is labeled as a second sensitive category; when the combined leakage path simultaneously contains nodes generated based on the author field mapping and nodes generated based on the author's affiliation field mapping, and the combined leakage path covers at least two document records, the combined leakage path is labeled as a third sensitive category; wherein the first sensitive category, the second sensitive category, and the third sensitive category are used to trigger different outbound control results.
[0009] Optionally, the method further includes: when the combined leakage path does not satisfy any of the first sensitive category, the second sensitive category, and the third sensitive category, or when the combined leakage path simultaneously satisfies multiple of the first sensitive category, the second sensitive category, and the third sensitive category, the bibliographic association diagram, the source status marker, and the outgoing context information are input into a preset path detection model for fusion encoding, and a supplementary category of the combined leakage path is output.
[0010] Optionally, generating an outbound control result based on the combined leakage path includes: generating a first outbound control result when the combined leakage path is not detected and the outbound context information meets preset outbound conditions; generating a second outbound control result when the combined leakage path is detected, and the combined leakage path contains the restricted status marker, the outbound context information meets the preset outbound conditions, and the document field corresponding to the combined leakage path has a replacement field; generating a third outbound control result when the combined leakage path is detected, and the combined leakage path contains the restricted status marker, the outbound context information meets the preset outbound conditions, and the document field corresponding to the combined leakage path does not have a replacement field; generating a third or fourth outbound control result according to the approval status in the outbound context information when the combined leakage path is detected, the combined leakage path contains the restricted status marker, and the outbound context information does not meet the preset outbound conditions; wherein, the first outbound control result, the second outbound control result, and the third outbound control result are release results, and the fourth outbound control result is a blocking result.
[0011] Optionally, the method further includes: when multiple bibliographic metadata data packets are sent out within a preset time window, and the outgoing context information corresponding to the multiple bibliographic metadata data packets has at least one identical outgoing context attribute, merging the bibliographic association graphs corresponding to the multiple bibliographic metadata data packets into a session association graph, wherein the session association graph includes cross-time association edges established based on the outgoing context attribute; detecting cumulative combined leakage paths according to the session association graph, wherein the cumulative combined leakage path is a combined leakage path that spans at least two outgoing operations; and updating the outgoing control result and detection record of the corresponding outgoing operation according to the detected cumulative combined leakage path.
[0012] According to a second aspect of the present invention, a sensitive information detection device is provided, comprising: The semantic parsing module is used to obtain the bibliographic metadata data package to be sent out and the corresponding outgoing context information, and to perform semantic processing on the document fields in the bibliographic metadata data package to generate document semantic units. The association graph construction module is used to construct a bibliographic association graph based on the co-occurrence relationship of the document semantic units, the field affiliation relationship of the document semantic units, and the external context information; The index matching module is used to match the document semantic unit with semantic indexes of different source types, and based on the matching results and the outgoing context information, determine the source status mark of the document semantic unit at the corresponding node in the bibliographic association graph, wherein the source status mark includes a restricted status mark; The result output module is used to perform leakage path detection on the bibliographic association graph according to the source status marker. When there is a combined leakage path in the bibliographic association graph, it generates outward control results and detection records based on the combined leakage path. The combined leakage path is a path in the bibliographic association graph that contains the restricted status marker and spans at least two of the document fields.
[0013] According to a third aspect of the present invention, a storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the above-described sensitive information detection method.
[0014] According to a fourth aspect of the present invention, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the above-described sensitive information detection method.
[0015] This invention provides a sensitive information detection method, apparatus, storage medium, and computer equipment. By constructing a bibliographic association graph and detecting leakage paths, it achieves detection based on field semantic relationships and outward behavior relationships. The constructed bibliographic association graph enables academic information in a single document field to form a computable association structure within the same document record. This allows for the identification of combined leakage paths formed by multiple document fields before the bibliographic metadata is sent out, thereby locating sensitive clues to specific nodes and relationship edges, accurately identifying combined leakage risks. By matching document semantic units with semantic indexes of different source types and determining source status markers, it effectively distinguishes between public and restricted sources. The inclusion of restricted status markers in combined leakage path detection reduces the confusion between public and restricted source content in outward detection, preventing misjudgments due to the inability to distinguish sources. By generating outward control results and detection records, and by including path structure information in combined leakage paths, a traceable detection record mechanism is established. This facilitates accurate location of original document fields, source evidence, and leakage paths in subsequent processing, and determines different handling states according to the outward control results, achieving precise tracing of specific triggering evidence and leakage paths. The above method achieves comprehensive detection of semantic relationships and outward behavior relationships by constructing a bibliographic association graph, introducing restricted state markers, and defining combined leakage paths. It also identifies cross-field combined leakage and distinguishes between public and restricted sources, ensuring traceability of detection results.
[0016] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description
[0017] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings: Figure 1 A flowchart illustrating a sensitive information detection method provided by an embodiment of the present invention is shown; Figure 2 A flowchart illustrating another sensitive information detection method provided by an embodiment of the present invention is shown; Figure 3 This diagram illustrates the structure of a sensitive information detection device according to an embodiment of the present invention. Figure 4 This invention provides a schematic diagram of the structure of another sensitive information detection device according to an embodiment of the invention. Figure 5 A schematic diagram of the device structure of a computer device provided in an embodiment of the present invention is shown. Detailed Implementation
[0018] Exemplary embodiments of the present application will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present application are shown in the drawings, it should be understood that the present application may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this application will be thorough and complete, and will fully convey the scope of the present application to those skilled in the art.
[0019] This application provides a method for detecting sensitive information, such as... Figure 1 As shown, the method includes the following steps: 101. Obtain the bibliographic metadata data package to be sent out and the corresponding sending context information, and perform semantic processing on the document fields in the bibliographic metadata data package to generate document semantic units.
[0020] The bibliographic metadata data package is a collection of structured or semi-structured bibliographic records exported from a bibliographic service platform, institutional knowledge base, or scientific research management system. Each bibliographic record has a bibliographic record identifier and contains multiple bibliographic fields. The outgoing context information is a collection of behavioral information generated when the bibliographic metadata data package is exported from the system to an external processing object, including the export event identifier, export user, export entry point, outgoing destination, outgoing method, outgoing purpose, and export time. The bibliographic semantic unit is a data object formed by natural language processing of bibliographic fields, including bibliographic record identifier, field identifier, field role, entity type, entity value, normalized value, semantic vector, and source evidence identifier.
[0021] Specifically, the bibliographic metadata contains multiple bibliographic fields. The outgoing context information records the behavioral attributes of this outgoing operation. Based on this, the bibliographic fields are semantically processed and converted into bibliographic semantic units. Semantic processing can transform the human-readable but difficult-to-calculate text content in the original bibliographic fields into structured and standardized data objects. Each bibliographic semantic unit carries the core semantic information of the corresponding bibliographic field and serves as the basic node for subsequent construction of the bibliographic association graph.
[0022] In this embodiment, semantic processing solves the problem that the original document fields cannot be directly used in graph structure calculation. Specifically, by converting document fields into structured document semantic units, the originally discrete and unstructured academic information is transformed into standard data objects that can be processed by subsequent graph construction and path search algorithms. This provides the necessary input data for the subsequent construction of bibliographic association graphs based on co-occurrence and field attribution relationships, which is a technical prerequisite for realizing combined leakage path detection.
[0023] 102. Construct a bibliographic association graph based on the co-occurrence relationship of document semantic units, the field affiliation relationship of document semantic units, and external context information.
[0024] Co-occurrence relationships refer to the associations established between different semantic units of documents due to the sharing of the same entity identifier. They are used to characterize the implicit connections between different document records. The same entity identifier includes the same author identifier, the same institution identifier, the same project identifier, or the same subject semantics. Field attribution relationships refer to the subordinate associations between each field semantic unit within the same document record and its parent document record. They are used to characterize the internal structure of a single document record. The bibliographic association graph is a graph data structure with document semantic units and outward distribution behavior as nodes and field attribution, co-occurrence, source status, and derivation relationships as edges.
[0025] Specifically, when constructing the bibliographic association graph, each document semantic unit is mapped to a node in the graph. Connection edges are established between nodes based on co-occurrence relationships, and belonging edges between field nodes and document nodes are established based on field attribution relationships. Furthermore, outward behavior nodes and their derived edges with document nodes are introduced based on outward context information, thereby forming a graph data structure containing multiple types of nodes and multiple types of relationships, including documents, fields, entities, and behaviors.
[0026] In this embodiment, by constructing a bibliographic association graph, the originally discrete and isolated semantic units of documents are organized into a graph data model with an associated structure, solving the problem that traditional detection methods cannot characterize the association relationships between fields, between documents, and between fields and outgoing actions. Specifically, the introduction of co-occurrence relationships enables the same authors, institutions, projects, or topics scattered in different document records to be associated, forming cross-document association paths; field affiliation relationships preserve the structural integrity within a single document record; outgoing context information incorporates the outgoing action itself into the graph, enabling the detection process to associate the attributes of the outgoing operation; based on the above processing, the combined associations formed between multiple fields and multiple records can be represented and searched in the graph in the form of paths, providing a computable basic structure for subsequent detection of combined leakage paths.
[0027] 103. Match the semantic units of the document with semantic indexes of different source types, and based on the matching results and external context information, determine the source status marker of the corresponding node in the bibliographic association graph of the semantic unit of the document, wherein the source status marker includes the restricted status marker.
[0028] Specifically, the semantic indexes of different source types include public domain semantic indexes and restricted domain semantic indexes. Public domain semantic indexes are sets of searchable semantic data generated from publicly available source records, including published documents, published bibliographic records, published patents, and published project information. Restricted domain semantic indexes are sets of searchable semantic data generated from restricted source records, including unpublished bibliographic records, pre-published documents, internal novelty search materials, research project materials, and sets of restricted-access documents. Index terms are the smallest matching records in the public domain semantic index and the restricted domain semantic index, and both include source evidence identifiers, source library types, source time, access level, title vectors, abstract vectors, keyword vectors, institution vectors, and project funding vectors. Source status markers are source attributes generated after matching the semantic units of a document with the public domain semantic index and the restricted domain semantic index.
[0029] Specifically, the semantic units of a document are matched with semantic indexes of different source types to obtain matching results. The semantic indexes of different source types include at least public domain semantic indexes and restricted domain semantic indexes, corresponding to public source records and restricted source records, respectively. The matching process calculates the semantic similarity between the semantic units of the document and each index item, and outputs the source type of the match. Then, based on the matching results and external context information, a source status marker is determined for the node corresponding to each semantic unit of the document in the bibliographic association graph. The source status marker is used to characterize the source attribute of the semantic content carried by this node. The restricted status marker indicates that the semantic unit of the document has matched a restricted domain semantic index, that is, its semantic content exists in unpublished documents, pre-published documents, internal novelty search materials, research project materials, or restricted access document sets.
[0030] In this embodiment, a source attribute is labeled for each document semantic unit through dual-domain semantic index matching, solving the problem that existing technologies cannot distinguish between publicly available and restricted sources of the same semantic content. Specifically, detection methods that do not distinguish sources will mix background information in publicly available documents with similar semantic content in restricted materials. On the one hand, publicly available content is misjudged as sensitive information, causing false blocking; on the other hand, restricted content is missed because it is superficially similar to publicly available content. By introducing a source status marker and setting a restricted status marker, publicly available and restricted source content have different data states during the detection process. Publicly available content does not trigger leakage judgment, while nodes with restricted status markers enter the subsequent combined leakage path detection. This processing effectively reduces the mixing of publicly available and restricted content in external detection and improves the accuracy of detection.
[0031] 104. Detect leakage paths in the bibliographic association graph based on the source status markers. When there are combined leakage paths in the bibliographic association graph, generate outward control results and detection records based on the combined leakage paths. The combined leakage path is a path in the bibliographic association graph that contains a restricted status marker and spans at least two document fields.
[0032] Among them, the combined leakage path is a sequence of nodes and edges in the bibliographic association graph that connects multiple document fields or multiple document records with outgoing context information and satisfies the sensitivity criteria; the outgoing control result refers to the disposal instructions executed on the bibliographic metadata package to be outgoing based on the detected combined leakage path and related conditions, which is used to determine whether the data package is released, released after desensitization, released after approval, or blocked; the detection record is a data record that records the export event, trigger field, source status, combined leakage path, and disposal action; multiple identifiers are generated and written into the detection record during the bibliographic export outgoing process, so that document fields, index hit results, combined leakage paths, and outgoing control results can be associated through the same export event identifier. The generated identifiers include the export event identifier, specifically the unique record identifier corresponding to a bibliographic metadata package outgoing process; the document record identifier, specifically the unique record identifier corresponding to a single document record in the bibliographic metadata package; the field identifier, specifically the unique record identifier corresponding to a single document field in the document record; and the source evidence identifier, specifically the unique record identifier corresponding to the source record of the index item.
[0033] Specifically, based on the identified source status markers, leakage path detection is performed on the bibliographic association graph. In particular, nodes with restricted status markers in the bibliographic association graph are the focus of attention. Paths that meet specific conditions are searched in the graph structure, namely, the path must contain at least one node with a restricted status marker, and the number of document fields corresponding to the nodes covered by the path is no less than two. The detection process is completed by a graph traversal algorithm, identifying all paths in the graph that meet the above conditions as combined leakage paths. Once a combined leakage path is detected, an outbound control result and a detection record are generated based on the path. The outbound control result is used to determine the processing method for the current outbound operation, such as allowing, allowing after desensitization, allowing after approval, or blocking, while the detection record is used to save detailed information of this detection.
[0034] In this embodiment, by defining and detecting combined leakage paths, the problem of existing technologies being unable to identify implicit leakage clues formed by combinations of multiple fields is solved. Considering that traditional detection methods only judge single fields independently, a single field may not be sensitive, but multiple fields may form sensitive information when combined. This application includes the relationship between multiple fields in the detection scope by searching for paths containing restricted status markers and spanning at least two document fields in the bibliographic association graph. As long as two or more document fields are associated through graph paths and there are restricted status markers on the paths, they are identified as combined leakage paths. This allows information such as author co-occurrence, institution co-occurrence, project co-occurrence, and topic co-occurrence, which were originally scattered in multiple fields, to be integrated and judged uniformly. This effectively detects the leakage risk of individual fields being insensitive, but multiple fields revealing sensitive clues such as research layout and unpublished results when combined. At the same time, the generated detection records will save the path information in association with the outward control results, providing a complete chain of evidence for subsequent tracing of the leakage source and auditing.
[0035] This invention provides a sensitive information detection method that constructs a bibliographic association graph and detects leakage paths to achieve detection based on field semantic relationships and outward behavior relationships. The constructed bibliographic association graph enables academic information in a single document field to form a computable association structure within the same document record. This allows for the identification of combined leakage paths formed by multiple document fields before the bibliographic metadata is sent out, thereby locating sensitive clues to specific nodes and relationship edges, and accurately identifying combined leakage risks. By matching document semantic units with semantic indexes of different source types and determining source status markers, the method effectively distinguishes between public and restricted sources. The inclusion of restricted status markers in combined leakage path detection reduces the confusion between public and restricted source content in outward detection, preventing misjudgments due to the inability to distinguish sources. By generating outward control results and detection records, and by including path structure information in the combined leakage paths, a traceable detection record mechanism is established. This facilitates subsequent processing to accurately locate the original document fields, source evidence, and leakage paths, and determines different handling states according to the outward control results, achieving precise tracing of specific triggering evidence and leakage paths. The above method achieves comprehensive detection of semantic relationships and outward behavior relationships by constructing a bibliographic association graph, introducing restricted state markers, and defining combined leakage paths. It also identifies cross-field combined leakage and distinguishes between public and restricted sources, ensuring traceability of detection results.
[0036] This application provides another method for detecting sensitive information, such as... Figure 2 As shown, it specifically includes: Step 201: Obtain the record metadata data to be sent out and the corresponding sending context information.
[0037] In this embodiment, the bibliographic metadata includes at least two bibliographic fields, which include at least two types of fields among title, author, author's affiliation, funding project, keywords, abstract, source, publication status, and references; the external transmission context information includes at least one of the following: exporting user, export entry point, export destination, export method, export purpose, and export time.
[0038] Step 202: Semantize the document fields in the bibliographic metadata data package to generate document semantic units.
[0039] Specifically, semantic extraction is performed on the core content description information and process event description information to generate topic semantic information and event elements, respectively; entity disambiguation is performed on the output attribution entity information to generate entity identifiers; field roles are determined based on field role identifier information, and document semantic units are generated based on topic semantic information, event elements, entity identifiers, and field roles.
[0040] In this implementation, the core content description information specifically refers to the core content portion of the title, keyword, and abstract fields. The process event description information specifically refers to the narrative text in the abstract field that describes the research methods, experimental data, results, and application scenarios. Based on this, word segmentation, phrase normalization, and entity recognition processing are performed on the core content description information to extract the research topics and research objects involved in the literature, generating topic semantic information to represent the core research direction and focus of the literature, in the form of structured topic tags or topic vectors. On the other hand, event element extraction processing is performed on the process event description information. Taking the abstract sentence as the processing object, through dependency parsing and role labeling, the predicate-object structure is identified, and the element types of research methods, experimental data, results, and application scenarios and their field positions in the abstract field are output to generate event elements, which are used to represent the specific research process and results described in the literature.
[0041] In addition, the output attribution entity information includes, but is not limited to, author field, author affiliation field, and funding project field. Entity disambiguation processing is performed on the output attribution entity information. Taking entity value, co-occurrence field, time information, and source evidence identifier as input, authors with the same name, abbreviations of institutions, abbreviations of funds, and thematic synonyms are mapped to normalized values to generate entity identifiers. Entity identifiers include at least one of author identifier, institution identifier, and project funding identifier, which are used to uniquely identify the same reference of an entity in different document records.
[0042] Subsequently, field roles are determined based on field role identification information, which includes at least one of field name, field content, and export template. Field roles are jointly determined through field name mapping, export template mapping, and field content classification. Field name mapping uses a correspondence table between platform field names, standard field names, and historical exported field names to normalize field names from different platforms or versions into standard field roles. Export template mapping records template identifiers, field order, field separators, and field meanings. Field roles are inferred based on the position and format of fields in the export template. Field content classification identifies names, institution names, project names, abstract sentences, keyword strings, and source names in field values. Content characteristics assist in determining field roles. Field roles are functional attributes of document fields in the bibliographic metadata data package, including at least one of title field, author field, author affiliation field, funding project field, keyword field, abstract field, source field, publication status field, and reference field. Among them, the publication status field records the document's status as open, pre-release, pending publication, internal circulation, and restricted access.
[0043] Finally, based on the generated topic semantic information, event elements, entity identifiers, and field roles, document semantic units are aggregated and generated. Each document semantic unit contains at least a document record identifier, field identifier, semantic content, entity identifier, field role, and source evidence identifier, which serve as the basic nodes for subsequent construction of the bibliographic association graph. When a field role is missing, the system infers the field role based on the field name, field content, and exported template.
[0044] Step 203: Construct a bibliographic association graph.
[0045] In this embodiment, the bibliographic association graph is a graph data structure, and its node types include document nodes generated by mapping each document record, field nodes generated by mapping each document field, author nodes generated by mapping the author field, institution nodes generated by mapping the author's affiliation field, project nodes generated by mapping the funding project field, subject nodes generated by mapping the subject semantic information, event nodes generated by mapping the event elements, source status nodes generated by mapping the source status marker, and outward behavior nodes generated by mapping the outward context information.
[0046] Specifically, the following types of edges are established between nodes: field affiliation edges are established between document nodes and field nodes to represent the structural relationship of a field belonging to the document record it belongs to; affiliation edges are established between author nodes and institution nodes to represent the affiliation relationship between authors and institutions; funding edges are established between document nodes and project nodes to represent the funding relationship between projects and documents; topic edges are established between document nodes and topic nodes to represent the semantic association between documents and topics; event edges are established between document nodes and event nodes to represent the semantic association between documents and events; source status edges are established between document nodes and source status nodes to represent the source status attribute of documents; and derived edges are established between document nodes and outward behavior nodes to represent the outward behavior attribute of documents.
[0047] The above nodes and edges are constructed based on the following: field affiliation edges are generated based on document record identifiers and field identifiers; co-occurrence relationships are generated based on author identifiers, institution identifiers, project funding identifiers, topic semantic information, and event elements within the same export batch, used to establish affiliation edges between author nodes and institution nodes, funding edges between document nodes and project nodes, topic edges between document nodes and topic nodes, and event edges between document nodes and event nodes; source status edges are generated based on source status markers; and export edges are generated based on export event identifiers and outgoing context information.
[0048] Each node in the bibliographic association graph has a node identifier, node type, associated document record identifier, and source evidence identifier. Each relation edge has an edge identifier, starting node, ending node, edge type, and edge generation basis. The bibliographic association graph is updated synchronously with the addition, deletion, field replacement, and outward control results of the bibliographic metadata data package.
[0049] The resulting bibliographic association graph enables academic information in a single document field to form a computable association structure between the same document record, multiple document records, and outward behavior, providing a graph structure basis for subsequent detection of combined leakage paths.
[0050] Step 204: Match the semantic units of the document with the semantic indexes of different source types to determine the source status marker.
[0051] In this embodiment, the document semantic unit is matched with the public domain semantic index and the restricted domain semantic index respectively. When the document semantic unit only hits one of the semantic indexes, the corresponding source status mark, i.e., the public source mark or the restricted source mark, is directly generated. However, there are also cases where the same document semantic unit hits both the public domain semantic index and the restricted domain semantic index. The handling method for this case is explained below.
[0052] First, it is necessary to trigger the mixed source tag. When the same semantic unit of a document hits both the public domain semantic index and the restricted domain semantic index, the system does not directly determine the semantic unit of the document as either public or restricted source. Instead, it reads the field role, public source time, restricted source time, export entry, and external purpose corresponding to the semantic unit of the document, and generates a mixed source tag based on the field sensitivity attributes, the time sequence of the source, and the external release scenario constraints. The mixed source tag includes the restricted priority mixed source tag, the public priority mixed source tag, and the mixed source tag to be reviewed. These tags are used to indicate the differences in the public status of the same semantic content in different source domains and their external release control tendencies.
[0053] Second, define and assign a source time. The public source time is the first publication time of the source record corresponding to the index item in the public domain semantic index. When the same document semantic unit hits multiple public records in the public domain semantic index, the public source time is the earliest publication time that can be verified by the public domain index.
[0054] The restricted source time is the verifiable time among the following: the entry time of the source record corresponding to the index item in the restricted domain semantic index, the pre-release registration time, the formation time of novelty search materials, the formation time of scientific research project materials, and the formation time of restricted access records. When the same document semantic unit hits multiple restricted records in the restricted domain semantic index, the restricted source time is the time of the restricted record with the highest index hit confidence of this document semantic unit. If there are multiple highest matching records, the latest restricted record time is taken.
[0055] Both publicly available and restricted source times are recorded with timestamps in a unified time zone. When a source record only has a date, month, or year, it is aligned and compared according to the coarsest time granularity that the source record can provide. When the granularity of publicly available and restricted source times is inconsistent, they are aligned according to the coarser time granularity before comparison. If the chronological relationship still cannot be determined after alignment, it is treated as a missing source time.
[0056] Third, the source time difference is calculated. The system normalizes and compares the open source time and the restricted source time to obtain the source time difference. A positive value is taken when the restricted source is earlier than the open source, and a negative value is taken when the open source is earlier than the restricted source. The closer the source time difference is to 1, the earlier the corresponding semantic unit appears in the restricted domain than in the open domain. The calculation method is as follows:
[0057] in, Let be the source time difference of the semantic unit of the j-th document; The restricted source time corresponding to the hit of the restricted domain semantic index for the j-th document semantic unit; The public source time corresponding to when the j-th document semantic unit hits the public domain semantic index; j is the document semantic unit number. The source time normalization scale is greater than zero, and its value is not less than the minimum time resolution interval set by the platform. (·) is the hyperbolic tangent function, used to compress the difference between publicly available source time and restricted source time into a finite interval.
[0058] The source time normalization scale can be determined based on the historical average publication time interval between publicly available bibliographies and internal bibliographies on the platform, or it can be determined based on the publication lag period set by the institution. The value is limited to between -1 and 1. The closer to 1, the earlier the restricted source appeared relative to the public source, and the more likely the semantic unit is to reflect internal information that has not yet been stably disclosed.
[0059] Fourth, calculate the mixed priority. The system further combines field roles and outbound context to generate a mixed priority:
[0060] in, The mixed priority of the j-th document semantic unit; Assign role weights to the fields; The field role to which the j-th document semantic unit belongs Corresponding role sensitivity coefficient; The field role for the j-th document semantic unit; Source time weighting; This refers to the aforementioned time difference in the source; Weights for outgoing context; For export entry External use Commonly corresponding context constraint coefficients; This is the export entry point corresponding to the j-th document semantic unit; This refers to the external dissemination purpose corresponding to the j-th document semantic unit.
[0061] The field role weight, source time weight, and outgoing context weight all range from 0 to 1, and the sum of the three is 1. The role sensitivity coefficient can be pre-calibrated according to the field role. The role sensitivity coefficients of the abstract field, the funding project field, and the publication status field are higher than those of the title field and the source field. The context constraint coefficient can be calibrated according to the export entry and outgoing purpose. The batch export entry, cross-institutional outgoing purpose, and non-public retrieval and citation purpose correspond to higher values, while the public retrieval and citation purpose corresponds to lower values. The role sensitivity coefficient and the context constraint coefficient both range from 0 to 1. The higher the value, the higher the contribution of the corresponding field role or outgoing scenario to the restricted outgoing determination.
[0062] Hybrid priority is only used for priority comparison when field role, source time, and outbound context have all been validly evaluated. When the field role belongs to the restricted priority field role set and the source time difference is positive, a public priority hybrid source tag will not be generated due to a low context constraint coefficient. When the export entry or outbound purpose is missing, the context constraint coefficient will be handled with a conservative value corresponding to the non-public retrieval and citation purpose.
[0063] Fifth, the decision threshold is dynamically adjusted. The decision threshold used to generate mixed-source markers can be adjusted to a limited extent based on recent detection records. The adjustment method is as follows:
[0064] in, The mixed priority threshold used in the h-th outbound decision period; This represents the upper limit of the mixed priority threshold. This is the lower limit of the mixed priority threshold; The baseline threshold for mixed priority; The threshold adjustment step size is greater than 0 and not greater than 1, which is used to limit the adjustment range of the mixed priority threshold within a single outbound decision period; The sensitivity ratio for manual review and confirmation corresponds to the h-th outbound judgment cycle; The threshold is the baseline value for manually verifying the sensitivity ratio; h is the outbound judgment cycle number.
[0065] Preferably, both the sensitivity ratio for manual review and the baseline value for the sensitivity ratio for manual review are between 0 and 1. When the sensitivity ratio for manual review is higher than the baseline value, the mixed priority threshold decreases, making it easier to trigger the restricted priority mixed source flag. When the sensitivity ratio for manual review is lower than the baseline value, the mixed priority threshold increases, increasing the number of public priority or pending review results. The threshold adjustment uses consecutive outbound records under the same exporting user, the same outbound destination, or the same outbound purpose as the statistical window. It is updated when there are valid review records in the window. Valid review records are those that have been manually confirmed and whose corresponding detection records have not been withdrawn, duplicated, merged, or changed in field version. When the number of valid review records in the statistical window is lower than the minimum number of reviews set by the platform, or when any version of the public domain semantic index, restricted domain semantic index, role sensitivity coefficient, or context constraint coefficient is switched, the mixed priority threshold is not updated in this outbound judgment cycle, and the threshold of the previous outbound judgment cycle is used. If there is no threshold in the previous outbound judgment cycle, the mixed priority baseline threshold is used.
[0066] Sixth, handle abnormal situations. When the source time is missing, or when the lower of the index hit confidence values returned by the public domain semantic index and the restricted domain semantic index is lower than the platform's preset retrieval confidence lower limit, do not generate restricted priority mixed source tags or public priority mixed source tags based on the mixed priority threshold, and transfer the corresponding document semantic unit to the mixed source tag to be reviewed.
[0067] Among them, index hit refers to the matching result between the semantic unit of the document and the index item; index hit confidence is the degree of semantic matching between the semantic unit of the document and the index item, and its value ranges from 0 to 1; the lower limit of retrieval confidence is the minimum matching degree of the index hit result participating in the source status labeling determination, which is determined by the verification set of public source records and restricted source records; source time missing refers to the state where the index item does not record a verifiable source time, or the public source time and the restricted source time cannot be compared sequentially according to a unified time granularity.
[0068] Seventh, generate mixed source tags. The method for generating mixed source tags is as follows: like ,but , like ,but , In other cases, ; in, The hybrid source tag generated for the j-th document semantic unit; The restricted priority hybrid source tag for the j-th document semantic unit; For the j-th document semantic unit, use the public priority mixed source tag; The j-th document semantic unit is a mixed source marker to be verified; For the roles mentioned above; For a set of roles with restricted priority fields; This refers to the aforementioned time difference in the source; The aforementioned mixed priority; The aforementioned mixed priority threshold; For public priority field role set; For the aforementioned external purposes; The set of publicly accessible citation and indexing uses is a set of external publication codes, including codes corresponding to publicly accessible citations, summaries of publicly available results, statistics of publicly available papers, and verification of publicly available sources. When an external publication purpose does not belong to the set of publicly accessible citation and indexing uses, it is treated as a non-publicly accessible citation and indexing purpose. When an external publication purpose is missing, the system reads the purpose code based on the export entry point and the external publication destination. When neither the export entry point nor the external publication destination can determine the purpose code, the external publication purpose is treated as a non-publicly accessible citation and indexing purpose.
[0069] The restricted priority field role set and the public priority field role set are non-overlapping field role sets; when field role configurations conflict, the restricted priority field role set shall prevail, which shall include at least the abstract field, the funding project field, and the publication status field; the public priority field role set shall include at least the title field and the source field.
[0070] When the abstract field, funding project field, or publication status field hits the restricted domain semantic index, and the restricted source time is earlier than the public source time, the system generates a restricted priority mixed source tag; when the semantic content corresponding to the title field or source field is in the public priority field role set, and the external publication purpose belongs to the public retrieval and citation purpose set, the system generates a public priority mixed source tag; if other fields in the same document record have already formed a restricted priority mixed source tag, the title field or source field will still participate in the bibliographic association diagram record as public side evidence, but will not separately cover the restricted priority mixed source tag.
[0071] Eighth, record the source status marker. During the above processing, the mixed source marker is written to both the source status node and the detection record. The source status node records the field role, public source time, restricted source time, export entry, external purpose, and mixed priority. The detection record records the index hit result that triggers the mixed source marker, the direction of time difference, and the corresponding external control basis. The detection record also saves the judgment version identifier. The judgment version identifier is used to record the public domain semantic index version, restricted domain semantic index version, field role configuration version, threshold configuration version, weight configuration version, and bibliographic association graph construction version used in this detection. It is only used for result backtracking, review alignment, and version rollback.
[0072] When the source time or restricted source time is missing, the system does not perform a source time difference comparison and generates a mixed source tag to be reviewed, so as to avoid restricted semantic content being mistakenly sent out as public semantic content due to incomplete source time.
[0073] Through the above methods, the system can make fine-grained source status determinations for semantic content that exists in both public and restricted domains based on field roles, differences in source time, and outgoing scenarios. This ensures that background information in public documents and similar semantic content in restricted materials have different data states during the detection process, thereby reducing the confusion between public and restricted source content in outgoing detection.
[0074] Step 205: Perform combined leakage path detection on the bibliographic association graph based on the source status markers.
[0075] Specifically, based on the source status markers, state nodes with restricted status markers and behavior nodes representing outward behavior are selected in the bibliographic association graph as the search starting point. Starting from the search starting point, the connecting edges in the bibliographic association graph are traversed to obtain candidate paths, and the number of fields and document records covered by the candidate paths are calculated. When there are candidate paths that cover no less than two fields or document records, and there are state nodes with restricted status markers on the candidate paths, the corresponding candidate paths are determined as combined leakage paths.
[0076] Specifically, when a combined leakage path contains nodes generated based on the author's affiliation field, nodes generated based on the funding project field, and nodes generated based on the title or keyword field, the combined leakage path is labeled as the first sensitive category. When a combined leakage path contains nodes generated based on the event element in the abstract field, nodes generated based on the publication status field, and status nodes based on the restricted status marker, the combined leakage path is labeled as the second sensitive category. When a combined leakage path contains nodes generated based on the author field and nodes generated based on the author's affiliation field, and the combined leakage path covers at least two document records, the combined leakage path is labeled as the third sensitive category. The first, second, and third sensitive categories are used to trigger different outbound control results.
[0077] In this implementation, when a combined leakage path is marked as a first sensitive category, it indicates that the path involves a combination of institutional, project, and topic associations, typically corresponding to sensitive information at the research layout level. For the first sensitive category, if the outgoing context information meets the preset outgoing conditions and a replacement field exists, the system generates a de-identified release result; if the preset outgoing conditions are met but no replacement field exists, the system generates an approved release result. When a combined leakage path is marked as a second sensitive category, it indicates that the path involves a combination of abstract events, publication status, and restricted sources, typically corresponding to sensitive information at the level of unpublished results. For the second sensitive category, the system generates an approved release result or a blocked result based on the satisfaction status of the outgoing destination and outgoing purpose. When a combined leakage path is marked as a third sensitive category, it indicates that the path involves a cross-record combination of author and institution associations, typically corresponding to sensitive information at the level of team collaboration. For the third sensitive category, if the outgoing context information meets the preset outgoing conditions, the system generates a release result or a de-identified release result; if the preset outgoing conditions are not met, the system generates an approved release result or a blocked result based on the approval status.
[0078] Specifically, the preset outbound conditions consist of data items that the system can read, including the exporting user role, export entry type, outbound destination type, outbound method, outbound purpose code, approval status, access level, and organization scope. The satisfaction status of the preset outbound conditions is determined by the above data items. When the access level allows the corresponding outbound purpose, and the outbound destination type, outbound method, and organization scope all match the access level, the outbound context information satisfies the preset outbound conditions.
[0079] In addition, the different types of outbound control results mentioned above have the following specific meanings: release result, which corresponds to the case where no combined leakage path is detected and the outbound context information meets the preset outbound conditions; release result after desensitization, which corresponds to the case where there is a replacement field and the replacement field meets the outbound purpose; release result after approval, which corresponds to the case where the mark, approval status or organization scope verification needs to be reviewed; and blocking result, which corresponds to the case where there is a conflict between the sensitive category and the outbound destination type, outbound method or outbound purpose code.
[0080] Furthermore, when the combined leakage path does not satisfy any of the first, second, and third sensitive categories, or when the combined leakage path simultaneously satisfies multiple categories of the first, second, and third sensitive categories, the bibliographic association diagram, source status marker, and outgoing context information are input into the preset path detection model for fusion encoding, and a supplementary category of the combined leakage path is output.
[0081] In this embodiment, when the sensitive category of the combined leakage path cannot be clearly determined by rules, the path detection model is triggered to perform supplementary identification. Specifically, the path detection model is triggered in the following two situations: First, the combined leakage path does not satisfy any of the first, second, and third sensitive categories. In this case, the path cannot be classified into any known sensitive category by the preset node combination rules, and the model needs to perform supplementary identification. Second, the combined leakage path simultaneously satisfies multiple categories of the first, second, and third sensitive categories. In this case, the path simultaneously meets the labeling conditions of multiple sensitive categories, resulting in category classification conflicts, and the model needs to make a comprehensive judgment on it.
[0082] When any of the above situations occur, the system inputs the citation association graph, source status markers, and outbound context information into a preset path detection model. The model fuses and encodes the node features, edge features, source status features, and outbound behavior features in the graph, and outputs supplementary categories for the combined leakage paths. These supplementary categories are used to generate corresponding outbound control results. When multiple combined leakage paths correspond to the same derived event, and some paths have their sensitive categories determined by rules while others have supplementary categories output by the model, the system determines the final outbound control result for that derived event according to a preset priority order, and writes the sensitive category or supplementary category of each path along with the corresponding outbound control result into the detection record.
[0083] In one specific implementation, the method for detecting combined leakage paths from the bibliographic association graph is as follows: First, the search starting point is selected. The system uses source status nodes and outbound behavior nodes as anchoring starting points. Source status nodes include those with restricted source tags, restricted priority mixed source tags, or mixed source tags awaiting verification. Outbound behavior nodes are formed by at least one of the following: exporting user, export entry point, export destination, export method, export purpose, and export time. When a source status node with a mixed source tag enters the path search, the public priority mixed source tag is only written into the passed node set as public-side source evidence. The restricted priority mixed source tag and the mixed source tag awaiting verification serve as anchoring criteria for candidate sensitive paths. Specifically, the anchoring starting point is selected as follows:
[0084] in, For the set of anchoring starting points; This is the b-th node in the bibliographic association graph; For node attribute identification; For source status node attribute identifiers with restricted source tags; For source status node attribute identifiers with restricted priority mixed source tags or mixed source tags pending review; For external distribution, it serves as a node attribute identifier.
[0085] Secondly, path search and candidate path generation are performed. The system is triggered from the nodes in the anchoring starting point set, and performs a limited depth traversal according to the field belonging edge, topic edge, event edge, funding edge, membership edge, and derived edge to obtain the candidate path set:
[0086] in, A set of candidate sensitive paths; This represents the a-th traversal path; The traversal path is obtained by taking the anchored starting point set as the starting point, the limited edge set as the passable edges, and the limited traversal depth as the maximum number of steps; For sensitive coverage; Set a threshold for candidate path coverage; before writing the path into the candidate path set, perform loop pruning and duplicate edge deduplication. When there are multiple paths from the same anchoring start point to the same ending node, prioritize retaining the path with higher sensitive coverage. When the sensitive coverage is the same, retain the path with fewer edges.
[0087] Then, the sensitivity coverage is calculated. The sensitivity coverage indicates whether a traversal path has expanded from a single sensitive node to a path with the potential for combined leakage. The calculation method is as follows:
[0088] in, , , , These are respectively: field coverage weight, document record coverage weight, source status coverage weight, and outward behavior coverage weight; , , , These are the corresponding coverage indicators; the field coverage indicator is 1 when the traversal path spans at least two document fields; the document record coverage indicator is 1 when the traversal path spans at least two document records; the source status coverage indicator is 1 when the traversal path contains at least one of the following: restricted source marker, restricted priority mixed source marker, or mixed source marker to be reviewed; and the outward behavior coverage indicator is 1 when the traversal path contains an outward behavior node.
[0089] Finally, the combined leakage path is determined. When the number of fields covered by the candidate path is not less than two, or the number of document records covered is not less than two, and the candidate path has at least one of the following: restricted source marker, restricted priority mixed source marker, or mixed source marker to be verified, the candidate path is determined as a combined leakage path.
[0090] The path detection model is trained using a supervised training method. The training samples take candidate sensitive paths as input and are labeled with sensitive types, trigger nodes, and trigger edges obtained through manual review. The model fuses and encodes node representations, edge representations, source state representations, and outward behavior representations to output supplementary sensitive types for candidate sensitive paths. Supplementary sensitive types include at least one of the following: leakage of scientific research layout, leakage of unpublished results, leakage of project funding relationships, leakage of team cooperation relationships, and leakage of restricted bibliographic sets. The model's inference output also includes trigger nodes, trigger edges, and path confidence categories. Path confidence categories include one of the following: high confidence category, medium confidence category, and low confidence category. When the path confidence category is low confidence category, the corresponding candidate sensitive path is written into the detection record and enters the review process. When there is a difference between the manual review result and the inference output, the review result is written into the training sample set.
[0091] When a combined leakage path contains nodes generated based on the author's affiliation field, nodes generated based on the funding project field, and nodes generated based on the title or keyword field, the path is labeled as the first sensitive category, such as a research layout leakage type. When a combined leakage path contains nodes generated based on event elements in the abstract field, nodes generated based on the publication status field, and status nodes based on the restricted status marker, the path is labeled as the second sensitive category, such as an unpublished result leakage type. When a combined leakage path contains nodes generated based on both the author field and the author's affiliation field, and the path covers at least two document records, the path is labeled as the third sensitive category, such as a team collaboration leakage type. These sensitive categories serve as the basis for confirming combined leakage paths and generating outbound control results, without changing the original definitions of nodes and edges in the bibliographic association graph. The priority order for type classification is: unpublished results take precedence over research layout, and research layout takes precedence over candidate types to be verified.
[0092] 206. Generate external control results based on combined leakage paths.
[0093] Specifically, when no combined leakage path is detected and the outbound context information meets the preset outbound conditions, a first outbound control result is generated; when a combined leakage path is detected, and the combined leakage path contains a restricted status marker, the outbound context information meets the preset outbound conditions, and the document field corresponding to the combined leakage path has a replacement field, a second outbound control result is generated; when a combined leakage path is detected, and the combined leakage path contains a restricted status marker, the outbound context information meets the preset outbound conditions, and the document field corresponding to the combined leakage path does not have a replacement field, a third outbound control result is generated; when a combined leakage path is detected, and the combined leakage path contains a restricted status marker, and the outbound context information does not meet the preset outbound conditions, a third or fourth outbound control result is generated according to the approval status in the outbound context information; the first, second, and third outbound control results are release results, i.e., outbound is allowed, and the fourth outbound control result is a blocking result, i.e., outbound is prohibited.
[0094] Specifically, the existence of a replacement field means that the field role already has corresponding replacement content in the replacement field mapping table, and the corresponding document field can be written back according to the field identifier. The replacement field is determined by the field role and the sensitivity type. For example, the author field is replaced with the first author, corresponding author, participating author or other author roles; the author affiliation field is replaced with the college level, institution level, regional level or industry level institution level; the funding project field is replaced with the national level project, provincial and ministerial level project, horizontal project or internal project category; the abstract field is replaced with the abstract event element text formed by summarizing the research object, research method and application scenario; and the keyword field is replaced with the research topic category.
[0095] The replacement field mapping table records the field role, sensitivity type, replacement field type, and replacement action. The replacement action includes at least one of generalization, deletion, and fixed text replacement. After the replacement field is generated, it is written back to the position corresponding to the original field identifier. The detection record saves the field identifier before replacement, the replacement field, the replacement action, and the corresponding combined leakage path. When the corresponding document field does not have a replacement field, the combined leakage path does not generate a desensitized release result.
[0096] 207. When multiple citation metadata data packets with the same outgoing context attributes are sent out within a preset time window, their citation association graphs are merged into a session association graph containing cross-association edges, in order to detect cumulative combined leakage paths that span at least two outgoing operations.
[0097] Specifically, when multiple bibliographic metadata data packets are sent out within a preset time window, and the outgoing context information corresponding to these bibliographic metadata data packets has at least one common outgoing context attribute, such as the same exporting user, the same outgoing destination, the same outgoing purpose, or the same exporting entry point, the system merges the bibliographic association graphs corresponding to the multiple bibliographic metadata data packets into a session association graph. The session association graph is an extended graph structure that retains cross-level association edges between multiple outgoing operations on the basis of the original bibliographic association graph. Specifically, the session association graph retains the node identifier, edge identifier, source status mark, and outgoing behavior node in each bibliographic association graph, and establishes cross-level association edges according to the same author identifier, the same institution identifier, the same project funding identifier, the same research topic, or the same outgoing destination. In this way, the bibliographic association graphs that were originally scattered in different outgoing operations are integrated into a unified session association graph, so that the association relationship across outgoing operations can be identified and detected.
[0098] In addition, the preset time window is an implementation parameter for determining whether multiple outbound operations are included in continuous outbound operations. It is configured by the system based on the queryable time range of the export logs and the outbound audit cycle. When multiple outbound operations fall into the same preset time window and meet at least one of the following conditions: same export user, same outbound destination, same outbound purpose, or same export entry, the corresponding record association graph is included in the same session association graph. When the preset time window configuration changes, the changed configuration version is written into the detection record.
[0099] Furthermore, based on the session association graph, the system detects cumulative combined leakage paths. A cumulative combined leakage path refers to a combined leakage path that spans at least two outgoing operations. The data items specifically include session identifier, outgoing operation identifier set, cross-node set, cross-edge set, cumulative sensitivity type, and cumulative outgoing control result. When a cumulative combined leakage path is detected, the system updates the outgoing control result and detection record of the corresponding outgoing operation according to the path.
[0100] The detection record is a complete data record of the detection results of each outbound operation, including the export event identifier, export user, export time, outbound destination, outbound method, outbound purpose, document record identifier, field identifier, field role, source evidence identifier, source status marker, combined leakage path, sensitivity type, outbound control result, replacement field, approval status, and model version. The detection record is associated with and saved with the original field identifiers of the bibliographic metadata data package, so that the trigger field, source evidence, path node, path relationship, and handling action can be queried according to the outbound operation identifier. For details, please refer to step 204.
[0101] Meanwhile, the version identifier is a combination of the public domain semantic index version, restricted domain semantic index version, field role configuration version, threshold configuration version, weight configuration version, path detection model version, and bibliographic association graph construction version used in a single detection. The review status is the manual confirmation status of the external control result corresponding to the detection record, including at least one of unreviewed, confirmed, and returned. When there is a manual review result for the detection record, the manual review result is associated and saved with the external operation identifier, combined leakage path, and version identifier.
[0102] Through the above methods, the system can identify cumulative combined leakage paths across outbound operations, preventing users from circumventing single detections by sending data in batches, thereby improving the comprehensiveness and accuracy of data leakage prevention detection.
[0103] Furthermore, as Figure 1 To specifically implement the method, this application provides a sensitive information detection device, such as... Figure 3 As shown, the device includes: a semantic parsing module 301, an association graph construction module 302, an index matching module 303, and a result output module 304.
[0104] The semantic parsing module 301 is used to obtain the bibliographic metadata data package to be sent out and the corresponding outgoing context information, and to perform semantic processing on the document fields in the bibliographic metadata data package to generate document semantic units. The association graph construction module 302 is used to construct a bibliographic association graph based on the co-occurrence relationship of the semantic units of the document, the field affiliation relationship of the semantic units of the document, and the external context information; The index matching module 303 is used to match the semantic units of the document with semantic indexes of different source types, and determine the source status marker of the corresponding node of the semantic unit of the document in the bibliographic association graph based on the matching results and the external context information. The source status marker includes the restricted status marker. The result output module 304 is used to detect leakage paths in the bibliographic association graph based on the source status marker. When there is a combined leakage path in the bibliographic association graph, it generates outward control results and detection records based on the combined leakage path. The combined leakage path is a path in the bibliographic association graph that contains a restricted status marker and spans at least two document fields.
[0105] In specific application scenarios, document fields include core content description information, process event description information, output attribution entity information, and field role identification information. The semantic parsing module 301 is specifically used to extract semantics from the core content description information and process event description information to generate topic semantic information and event elements, respectively; to perform entity disambiguation on the output attribution entity information to generate entity identifiers; to determine field roles based on field role identification information; and to generate document semantic units based on topic semantic information, event elements, entity identifiers, and field roles.
[0106] In specific application scenarios, the result output module 304 is specifically used to select state nodes with restricted state tags and behavior nodes that represent outward behavior in the bibliographic association graph based on the source state tags, as the search starting point; starting from the search starting point, it traverses the connection edges in the bibliographic association graph to obtain candidate paths, and calculates the number of fields and document records covered by the candidate paths; when there are candidate paths that cover no less than two fields or document records, and there are state nodes with restricted state tags on the candidate paths, the corresponding candidate paths are determined as combined leakage paths.
[0107] In specific application scenarios, the result output module 304 is also used to mark the combined leakage path as a first sensitive category when nodes generated based on the author's affiliation field mapping, nodes generated based on the funding project field mapping, and nodes generated based on the title or keyword field mapping all exist on the combined leakage path; when nodes generated based on the event element mapping in the abstract field mapping, nodes generated based on the publication status field mapping, and status nodes based on the restricted status marker all exist on the combined leakage path, it is marked as a second sensitive category; when nodes generated based on the author field mapping and nodes generated based on the author's affiliation field mapping all exist on the combined leakage path, and the combined leakage path covers at least two document records, it is marked as a third sensitive category; wherein, the first sensitive category, the second sensitive category, and the third sensitive category are used to trigger different outbound control results.
[0108] In specific application scenarios, the result output module 304 is also used to input the bibliographic association diagram, source status marker and outgoing context information into the preset path detection model for fusion encoding when the combined leakage path does not meet any of the first sensitive category, the second sensitive category and the third sensitive category, or when the combined leakage path meets multiple categories of the first sensitive category, the second sensitive category and the third sensitive category at the same time, and output the supplementary category of the combined leakage path.
[0109] In specific application scenarios, the result output module 304 is specifically used to generate a first outbound control result when no combined leakage path is detected and the outbound context information meets the preset outbound conditions; to generate a second outbound control result when a combined leakage path is detected, and the combined leakage path contains a restricted status marker, the outbound context information meets the preset outbound conditions, and the document field corresponding to the combined leakage path has a replacement field; to generate a third outbound control result when a combined leakage path is detected, and the combined leakage path contains a restricted status marker, the outbound context information meets the preset outbound conditions, and the document field corresponding to the combined leakage path does not have a replacement field; and to generate a third or fourth outbound control result according to the approval status in the outbound context information when a combined leakage path is detected, the combined leakage path contains a restricted status marker, and the outbound context information does not meet the preset outbound conditions. Among these, the first, second, and third outbound control results are release results, and the fourth outbound control result is a blocking result.
[0110] In specific application scenarios, such as Figure 4 As shown, the above-mentioned device also includes a multi-batch detection module 305, which is specifically used to merge the citation association graphs corresponding to the multiple citation meta-data packets into a session association graph when multiple citation meta-data packets are sent out within a preset time window and the outgoing context information corresponding to the multiple citation meta-data packets has at least one outgoing context attribute. The session association graph includes cross-time association edges established based on the same outgoing context attribute. The module detects cumulative combined leakage paths according to the session association graph, where the cumulative combined leakage path is a combined leakage path that spans at least two outgoing operations. Based on the detected cumulative combined leakage paths, the module updates the outgoing control results and detection records of the corresponding outgoing operations.
[0111] It should be noted that other corresponding descriptions of the functional units involved in the sensitive information detection device provided in this embodiment can be found in [reference]. Figure 1 and Figure 2 The corresponding descriptions in [the document] will not be repeated here.
[0112] Based on the above, Figure 1 and Figure 2 Accordingly, this embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the above-described sensitive information detection method.
[0113] Based on this understanding, the technical solution of this application can be embodied in the form of a software product. The software product to be identified can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, or portable hard drive), including several instructions to cause a computer device (such as a personal computer, server, or network device) to execute the sensitive information detection methods of various implementation scenarios of this application.
[0114] Based on the above, Figure 1 and Figure 2 The method shown, and Figure 3 and Figure 4 The illustrated embodiment of the sensitive information detection device, in order to achieve the above objectives, such as... Figure 5 As shown, this embodiment also provides a physical device for sensitive information detection. This device includes a communication bus, a processor, a memory, and a communication interface. It may also include input / output interfaces and a display device. The various functional units can communicate with each other via the bus. The memory stores a computer program, and the processor executes the program stored in the memory to perform the sensitive information detection method described in the above embodiment.
[0115] Optionally, the physical device may also include a user interface, a network interface, a camera, radio frequency (RF) circuitry, sensors, audio circuitry, a Wi-Fi module, etc. The user interface may include a display screen, input units such as a keyboard, etc., and optional user interfaces may also include USB interfaces, card reader interfaces, etc. The network interface may optionally include standard wired interfaces, wireless interfaces (such as Wi-Fi interfaces), etc.
[0116] Those skilled in the art will understand that the structure of the sensitive information detection physical device provided in this embodiment does not constitute a limitation on the physical device, and may include more or fewer components, or combine certain components, or have different component arrangements.
[0117] The storage medium may also include an operating system and a network communication module. The operating system is a program that manages the hardware and software resources of the aforementioned physical device, supporting the operation of information processing programs and other software and / or programs to be identified. The network communication module is used to enable communication between the various components within the storage medium, as well as communication with other hardware and software in the information processing physical device.
[0118] Through the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware platform, or it can be implemented by hardware. By applying the technical solution of this application, and by constructing a bibliographic association graph and detecting combined leakage paths, detection of semantic relationships and outward behavior relationships oriented towards fields is achieved. The constructed bibliographic association graph enables academic information in a single document field to form a computable association structure within the same document record. This allows for the identification of combined leakage paths formed by multiple document fields before the bibliographic metadata is sent out, thereby locating sensitive clues to specific nodes and relationship edges, and accurately identifying combined leakage risks. By matching document semantic units with semantic indexes of different source types and determining source status markers, the public and restricted sources are effectively distinguished. The inclusion of restricted status markers in combined leakage path detection reduces the confusion between public and restricted source content in outward detection, preventing misjudgments due to the inability to distinguish sources. By generating outward control results and detection records, and by including path structure information in the combined leakage paths, a traceable detection record mechanism is established. This facilitates accurate location of original document fields, source evidence, and leakage paths in subsequent processing, and determines different handling statuses according to the outward control results, achieving precise tracing of specific triggering evidence and leakage paths. The above method achieves comprehensive detection of semantic relationships and outward behavior relationships by constructing a bibliographic association graph, introducing restricted state markers, and defining combined leakage paths. It also identifies cross-field combined leakage and distinguishes between public and restricted sources, ensuring traceability of detection results.
[0119] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of a preferred embodiment, and the modules or processes shown in the drawings are not necessarily essential for implementing this application. Those skilled in the art will understand that the modules in the apparatus of the embodiment can be distributed within the apparatus of the embodiment as described, or can be modified to be located in one or more apparatuses different from this embodiment. The modules of the above-described embodiment can be combined into one module, or further divided into multiple sub-modules.
[0120] The serial numbers in this application are for descriptive purposes only and do not represent the superiority or inferiority of any particular implementation scenario. The above disclosures are merely a few specific implementation scenarios of this application; however, this application is not limited thereto, and any variations conceived by those skilled in the art should fall within the protection scope of this application.
Claims
1. A method for detecting sensitive information, characterized in that, include: Obtain the bibliographic metadata data package to be sent out and the corresponding sending context information, and perform semantic processing on the document fields in the bibliographic metadata data package to generate document semantic units; Based on the co-occurrence relationship of the document semantic units, the field affiliation relationship of the document semantic units, and the external context information, a bibliographic association graph is constructed; The document semantic unit is matched with semantic indexes of different source types, and based on the matching results and the outgoing context information, the source status marker of the document semantic unit at the corresponding node in the bibliographic association graph is determined, wherein the source status marker includes a restricted status marker; Leakage path detection is performed on the bibliographic association graph based on the source status marker. When there is a combined leakage path in the bibliographic association graph, an outward control result and detection record are generated based on the combined leakage path. The combined leakage path is a path in the bibliographic association graph that contains the restricted status marker and spans at least two of the document fields.
2. The method according to claim 1, characterized in that, The document fields include core content description information, process event description information, output attribution entity information, and field role identification information; the semantic processing of the document fields in the bibliographic metadata data package to generate document semantic units includes: Semantic extraction is performed on the core content description information and the process event description information to generate topic semantic information and event elements, respectively. Entity disambiguation is performed on the output attribution entity information to generate entity identifiers; Field roles are determined based on the field role identification information, and document semantic units are generated according to the topic semantic information, the event elements, the entity identifiers, and the field roles.
3. The method according to claim 1, characterized in that, The step of detecting leakage paths in the bibliographic association graph based on the source status marker includes: Based on the source status marker, select the status node with the restricted status marker and the behavior node used to characterize the outward behavior in the bibliographic association graph as the search starting point; Starting from the search origin, traverse the connecting edges in the bibliographic association graph to obtain candidate paths, and calculate the number of fields and document records covered by the candidate paths; When there are at least two candidate paths that cover a number of fields or document records, and there are state nodes with the restricted state marker on the candidate path, the corresponding candidate path is determined as a combined leakage path.
4. The method according to claim 1 or 3, characterized in that, The method further includes: When the combined leakage path contains nodes generated based on the author's affiliation field mapping, nodes generated based on the funding project field mapping, and nodes generated based on the title or keyword field mapping, the combined leakage path is marked as the first sensitive category. When the combined leakage path simultaneously contains nodes generated based on the mapping of event elements in the summary field, nodes generated based on the mapping of the publication status field, and status nodes based on the restricted status marker, the combined leakage path is marked as the second sensitive category. When the combined leakage path contains both nodes generated based on the author field mapping and nodes generated based on the author affiliation field mapping, and the combined leakage path covers at least two document records, the combined leakage path is marked as the third sensitive category. The first sensitive category, the second sensitive category, and the third sensitive category are used to trigger different outgoing control results.
5. The method according to claim 4, characterized in that, The method further includes: When the combined leakage path does not satisfy any of the first sensitive category, the second sensitive category, and the third sensitive category, or when the combined leakage path simultaneously satisfies multiple of the first sensitive category, the second sensitive category, and the third sensitive category, the bibliographic association diagram, the source status marker, and the outgoing context information are input into a preset path detection model for fusion encoding, and a supplementary category of the combined leakage path is output.
6. The method according to claim 1, characterized in that, The generation of outward control results based on the combined leakage path includes: When no combined leakage path is detected and the outgoing context information meets the preset outgoing conditions, a first outgoing control result is generated; When the combined leakage path is detected, and the combined leakage path contains the restricted status marker, the outbound context information meets the preset outbound conditions, and the literature field corresponding to the combined leakage path has a replacement field, a second outbound control result is generated. When the combined leakage path is detected, and the combined leakage path contains the restricted status marker, the outbound context information meets the preset outbound conditions, and the literature field corresponding to the combined leakage path does not have a replacement field, a third outbound control result is generated. When the combined leakage path is detected, and the combined leakage path contains the restricted status marker, and the outgoing context information does not meet the preset outgoing conditions, the third outgoing control result or the fourth outgoing control result is generated according to the approval status in the outgoing context information. Among them, the first outgoing control result, the second outgoing control result, and the third outgoing control result are release results, and the fourth outgoing control result is a blocking result.
7. The method according to claim 1, characterized in that, The method further includes: When multiple record metadata data packets are sent out within a preset time window, and the outgoing context information corresponding to the multiple record metadata data packets has at least one common outgoing context attribute, the record association graphs corresponding to the multiple record metadata data packets are merged into a session association graph, wherein the session association graph includes cross-time association edges established based on the outgoing context attribute. The cumulative combined leakage path is detected based on the session association graph, wherein the cumulative combined leakage path is a combined leakage path that spans at least two outgoing operations; Based on the detected cumulative combined leakage paths, update the outbound control results and detection records of the corresponding outbound operations.
8. A sensitive information detection device, characterized in that, include: The semantic parsing module is used to obtain the bibliographic metadata data package to be sent out and the corresponding outgoing context information, and to perform semantic processing on the document fields in the bibliographic metadata data package to generate document semantic units. The association graph construction module is used to construct a bibliographic association graph based on the co-occurrence relationship of the document semantic units, the field affiliation relationship of the document semantic units, and the external context information; The index matching module is used to match the document semantic unit with semantic indexes of different source types, and based on the matching results and the outgoing context information, determine the source status mark of the document semantic unit at the corresponding node in the bibliographic association graph, wherein the source status mark includes a restricted status mark; The result output module is used to perform leakage path detection on the bibliographic association graph according to the source status marker. When there is a combined leakage path in the bibliographic association graph, it generates outward control results and detection records based on the combined leakage path. The combined leakage path is a path in the bibliographic association graph that contains the restricted status marker and spans at least two of the document fields.
9. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.
10. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.
