A method and system for dynamic classification of public security data
By constructing a variable subgraph and graph attention mechanism, combined with label graph and lineage propagation reasoning, the sensitivity of public security data is dynamically assessed, solving the problem of inaccurate identification of public security data in existing technologies, and achieving efficient sensitive data classification and security management.
Patent Information
- Application Number
- CN202611565045.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-10-10
- Publication Date
- 2026-11-17
AI Technical Summary
Existing technologies struggle to accurately identify and differentiate sensitive data in public security data platforms, especially when field naming is non-standard, annotations are missing, or abbreviations are confusing. Static field labeling methods lead to an underestimation of the level of derived data, making it impossible to accurately identify sensitive public security data.
By collecting metadata from the public security data platform, generating version snapshots, constructing change subgraphs, performing semantic encoding and graph attention mechanism updates, and combining tag graphs and kinship propagation reasoning, a data-sensitive evidence matrix is generated to achieve dynamic risk assessment and classification.
It enables accurate, reliable, and reasonable identification and classification of sensitive data in public security data, avoids the consumption of computing resources caused by full scanning, and ensures the accuracy and interpretability of data security level classification.
Smart Images

Figure CN123116341A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of data classification technology, and in particular relates to a method and system for dynamic classification of public security data. Background Technology
[0002] Public security data platforms typically aggregate heterogeneous data from multiple sources, including population, police reports, cases, trajectories, images, videos, vehicles, equipment, law enforcement supervision, data services, API interfaces, data assets, data lineage, and data security policies. In actual processing, a large number of derived fields, business tags, personnel profiles, and interface service data are generated around the individuals involved. These tags are not only important data assets for police report analysis but also key sensitive objects in public security data security governance.
[0003] Currently, existing data classification and grading schemes mostly rely on field names, field annotations, fixed regular expression rules, or manually maintained data dictionaries. While this approach can identify explicitly sensitive fields such as ID numbers, mobile phone numbers, and names, its accuracy is low when field naming is non-standard, annotations are missing, abbreviations are inconsistent, or sample values are incomplete. Furthermore, while a single field may not possess obvious sensitive attributes, the combination of multiple fields can reconstruct information such as person identity, activity trajectories, image features, case correlations, or social relationships. Sensitive data may also be inherited, enhanced, or attenuated during processes such as copying, format conversion, aggregation, and API release. Currently, if static field annotation methods are still used, the level of derived data is easily underestimated, making it impossible to accurately identify and distinguish sensitive public security data. Summary of the Invention
[0004] In view of this, embodiments of the present invention provide a method and system for dynamic classification of public security data, which is used to solve the current problem of difficulty in accurately distinguishing sensitive public security data.
[0005] In a first aspect of the present invention, a method for dynamic classification of public security data is provided, comprising: Collect metadata from the public security data platform and generate a version snapshot for the metadata; Based on the current version snapshot and historical version snapshots, determine the set of data changes, and perform a traversal search on the set of data changes in a heterogeneous graph composed of field lineage, interface reference relationship and field co-occurrence relationship to obtain the change subgraph; For each object in the change subgraph, a desensitization context is constructed. The desensitization context is semantically encoded to obtain a node semantic vector. A semantic evidence vector containing the node semantic vector, semantic candidate category probability, and semantic uncertainty is constructed. The changed subgraph is used as the updated node range, and the semantic evidence vector is used as the node initialization value. A field co-occurrence graph and edge features are constructed. Based on the edge features, the node relevance is updated through a graph attention mechanism to generate co-occurrence evidence composed of node semantic combination sensitivity scores and main contribution edges. Based on the candidate category probabilities in the semantic evidence vector, the node business label is determined. The node business label is used as a seed node. Risk transmission calculation is performed in the label graph composed of label nodes, node relationship types and node semantic combination sensitivity scores to obtain the label graph score. Based on the changed subgraph, the data lineage relationship edges are determined. Lineage propagation features are constructed for each lineage relationship edge. The node semantic combination sensitivity score and the label graph score are used as lineage propagation enhancement coefficients. The path risk of each lineage relationship edge is calculated. The path risks are weighted and aggregated to obtain the node lineage propagation risk. Based on the semantic evidence vector, the co-occurrence evidence, the tag graph score, and the node lineage transmission risk, a data-sensitive evidence matrix is constructed. Combined with prior weight boundary constraints, the comprehensive risk and risk level classification of public security data are determined.
[0006] In a second aspect of the present invention, a dynamic classification system for public security data is provided, comprising: The data acquisition module is used to collect metadata from the public security data platform and generate version snapshots for the metadata. The change subgraph generation module is used to determine the set of data changes based on the current version snapshot and historical version snapshots. It then performs a traversal search on the set of data changes in a heterogeneous graph composed of field lineage, interface reference, and field co-occurrence relationships to obtain the change subgraph. The semantic evidence construction module is used to construct a desensitized context for each object in the change subgraph, obtain a node semantic vector by semantic encoding of the desensitized context, and construct a semantic evidence vector containing the node semantic vector, semantic candidate category probability and semantic uncertainty. The co-occurrence evidence construction module is used to use the changed subgraph as the update node range, the semantic evidence vector as the node initialization value, construct the field co-occurrence graph and edge features, update the node relevance based on the edge features through a graph attention mechanism, and generate co-occurrence evidence composed of node semantic combination sensitivity score and main contribution edge. The tag graph reasoning module is used to determine the node business tag based on the candidate category probability in the semantic evidence vector, and use the node business tag as a seed node to perform risk transmission calculation in the tag graph composed of tag nodes, node relationship types and node semantic combination sensitivity scores to obtain the tag graph score. The bloodline propagation reasoning module is used to determine the data bloodline relationship edges based on the changed subgraph, construct bloodline propagation features for each bloodline relationship edge, and use the node semantic combination sensitivity score and the label graph score as bloodline propagation enhancement coefficients to calculate the path risk of each bloodline relationship edge. The path risks are then weighted and aggregated to obtain the node bloodline propagation risk. The risk fusion output module is used to construct a data-sensitive evidence matrix based on the semantic evidence vector, the co-occurrence evidence, the tag graph score, and the node lineage transmission risk, and to determine the comprehensive risk and risk level classification of public security data by combining prior weight boundary constraints.
[0007] In a third aspect of the present invention, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable by the processor, wherein the processor executes the computer program to implement the steps of the method as described in the first aspect of the present invention.
[0008] In a fourth aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the method provided in the first aspect of the present invention.
[0009] In this embodiment of the invention, affected objects are identified through incremental differential metadata analysis, semantically encoded, combined sensitive patterns are identified through field co-occurrence graphs, business label risks are determined through label graph inference, derived data risks are calculated through lineage propagation inference, and the final risk level is obtained by fusing multi-source evidence. This allows for accurate identification and differentiation of sensitive public security data. Risk inference based on dimensions such as field co-occurrence, label graphs, and lineage relationships ensures the accuracy, rationality, and interpretability of data level classification. Furthermore, processing can be performed only on changed subgraph objects, avoiding the computational resource consumption caused by a full scan of public security data. Attached Figure Description
[0010] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0011] Figure 1 A flowchart illustrating a dynamic classification method for public security data according to an embodiment of the present invention; Figure 2 This is another flowchart illustrating a dynamic classification method for public security data provided in one embodiment of the present invention; Figure 3This is a schematic diagram of the structure of a dynamic classification system for public security data provided in one embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of an electronic device provided in one embodiment of the present invention. Detailed Implementation
[0012] To make the objectives, features, and advantages of this invention more apparent and understandable, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described below are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.
[0013] It should be understood that the terms "comprising" and other similar expressions in the specification, claims, and accompanying drawings of this invention are intended to cover a non-exclusive inclusion, such as a process, method, system, or apparatus that includes a series of steps or units and is not limited to the listed steps or units. Furthermore, "first" and "second" are used to distinguish different objects and are not intended to describe a specific order.
[0014] Please see Figure 1 The present invention provides a flowchart illustrating a dynamic classification method for public security data, comprising: S101. Collect metadata from the public security data platform and generate a version snapshot for the metadata; The metadata of the public security data platform can include databases, tables, fields, interfaces, SQL statements, sample value patterns, business tags, data lineage, historical review records, and access call records, etc., and a version snapshot, i.e., a data state copy, is generated for each piece of metadata. In this embodiment, a data version snapshot is generated every time the metadata of the public security data platform is collected.
[0015] Specifically, for each metadata object, a global object identifier (oid), version number (v), collection timestamp (ts), and source credibility (r) are set, and a versioned metadata snapshot (M_v) is output.
[0016] The sample values of global objects are format-preserving and desensitized or type-placed before entering subsequent processing. During processing, object association is based on object identifiers rather than directly using real identity identifiers (such as ID cards).
[0017] S102. Based on the current version snapshot and historical version snapshots, determine the data change set, and perform a traversal search on the data change set in the heterogeneous graph composed of field lineage, interface reference relationship and field co-occurrence relationship to obtain the change subgraph; The current version snapshot M_v is compared with the historical version snapshot M_(v-1), and the data change set ΔM_v = {m∈M_v | Hash(m,v)≠Hash(m,v-1)} is calculated. Objects in the data change set ΔM_v are used as seeds (e.g., unique object identifiers). A depth-limited traversal is performed on the heterogeneous graph composed of lineage relationships, interface reference relationships, and existing co-occurrence relationships. Affected objects are expanded along co-occurrence relationships, interface reference relationships, and lineage relationships to obtain change subgraphs and change type codes (e.g., addition, deletion, modification). Field co-occurrence relationships include, for example, associations between personnel and surveillance images, personnel and case details, and personnel and activity trajectories. Field lineage relationships are the lineage relationships between object data after processing methods such as copying, standardization, fusion, statistics, tag calculation, and interface publishing.
[0018] For objects that have not changed, the snapshot data from the previous version is reused, so that the scope of AI inference directly depends on the incremental difference results.
[0019] S103. Construct a desensitization context for each object in the change subgraph, obtain a node semantic vector by semantic encoding of the desensitization context, and construct a semantic evidence vector containing the node semantic vector, semantic candidate category probability and semantic uncertainty. An object's de-identification context includes at least field names, comments, table names, adjacent fields, interface descriptions, SQL snippets, sample patterns, and variation types. A semantic encoder can process this de-identification context into a vector representation with inherent meaning and relationships—a machine-understandable expression.
[0020] The semantic candidate category probability represents the normalized confidence distribution of a node's semantics belonging to various predefined semantic categories. Essentially, it represents the posterior probability estimate of a node's semantics belonging to the i-th semantic category under the current context or features, and can be output by the semantic encoder. Semantic uncertainty, on the other hand, represents a quantitative measure of the confidence level in its own output semantic representation or category prediction result. Semantic uncertainty can be calculated using the information entropy formula.
[0021] Based on node semantic vectors, semantic candidate category probabilities, and semantic uncertainty, semantic evidence vectors can be constructed as initial values for subsequent field co-occurrence graph nodes and as evaluation items for comprehensive risk calculation.
[0022] S104. Using the changed subgraph as the update node range and the semantic evidence vector as the node initialization value, construct the field co-occurrence graph and edge features. Based on the edge features, update the node relevance through the graph attention mechanism to generate co-occurrence evidence composed of node semantic combination sensitivity score and main contribution edge. Objects in the change subgraph are used as nodes in the field co-occurrence graph, and co-occurrence relationships are used as edges in the field co-occurrence graph. Edge features between nodes are defined, which may include field co-occurrence frequency, semantic distance, call frequency, relationship type, semantic vectors of both endpoints, and change type encoding. Based on node and edge features, the corresponding field co-occurrence graph can be constructed.
[0023] Based on edge features and node semantics, a graph attention mechanism can be used to learn and quantify the relevance of adjacent nodes. Graph attention networks learn edge contributions from edge features such as co-occurrence frequency, relationship type, business distance, and semantic similarity. The core technology of the graph attention mechanism is to aggregate information by adaptively learning neighbor node weights, which can dynamically distinguish the importance of neighbor nodes without relying on a fixed graph structure.
[0024] Optionally, the co-occurrence frequency, semantic distance, call frequency, relationship type, semantic vectors of both ends, and change type are encoded and concatenated into edge features of the field co-occurrence graph. The node relevance after normalization of the edge features is calculated through a graph attention mechanism. Based on the node relevance, a Sigmoid gated network is used to calculate and output the node semantic combination sensitivity score. Relationship edges whose node semantic combination sensitivity scores exceed a preset threshold are selected to obtain the main contributing edges.
[0025] For example, the unnormalized relevance e_ij = LeakyReLU(aTT[W_hh_i||W_hh_j||W_xx_ij]) is calculated using a graph attention network, where W_hh_i, W_hh_j, and W_xx_ij represent the semantic weights of node i, node j, and edge features, respectively. Normalization is performed within the neighborhood of node i, and the node representation h_i′ = σ(Σ_(j∈N(i))α_ij·W_vh_j), where α_ij represents the normalized relevance, and W_vh_j represents the weight of node j. A combined sensitivity score is then generated using a Sigmoid-gated network, for example, represented as C_i = Sigmoid(w_c^T[h_i′||Pool({h_j′})] + b_c). The main contributing edge can be determined by the combined semantic sensitivity score of the node; when the combined sensitivity score exceeds a threshold, it can be considered a main contributing edge (i.e., a high-contribution edge).
[0026] S105. Determine the node business label based on the candidate category probability in the semantic evidence vector, use the node business label as the seed node, perform risk transmission calculation in the label graph composed of label nodes, node relationship types and node semantic combination sensitivity scores, and obtain the label graph score. The label graph is a graph constructed by using labels as nodes, node relationship types (i.e., label dependencies) as edges, and combined sensitivity scores as relationship weights. Node relationship types can include hierarchical relationships, synonymous relationships, combined relationships, business scenario relationships, sensitivity enhancement relationships, and lineage dependency relationships, etc.
[0027] Based on the candidate category probabilities in the semantic evidence vector, the hit business label can be determined. The business label is used as a seed node, and a traversal search is performed in the label graph to calculate the label graph score.
[0028] Preferably, when the semantic combination sensitivity score of the seed node or the probability of the semantic candidate category of the seed node exceeds a preset threshold, risk transmission calculation is triggered, and the weight of the relationship edge of the label node in the label graph is determined by the Sigmoid gating network to obtain the label graph score.
[0029] The semantic candidate category probability and combined sensitivity score are used as gating conditions for triggering label relationships, and the label relationship edge weights are iteratively calculated to quantify the label relationship score. For example, a Sigmoid gating network iteratively calculates the gating value κ_rs = Sigmoid(w_t^T[z_r||z_s||g_rs||C_o]+b_t), and the relationship edge weights are determined based on the gating value. z_r represents the label node, z_s represents the relationship edge, g_rs represents the relationship type, and C_o represents the combined sensitivity score. When the semantic combined sensitivity score of a node or the semantic candidate category probability of a seed node reaches a corresponding threshold, the combined risk calculation is initiated, and the label graph score is obtained by combining the relationship edge weights.
[0030] S106. Based on the changed subgraph, determine the data lineage relationship edges, construct lineage propagation features for each lineage relationship edge, and use the node semantic combination sensitivity score and the label graph score as lineage propagation enhancement coefficients to calculate the path risk of each lineage relationship edge. Then, weight and aggregate the path risks to obtain the node lineage propagation risk. Based on the objects in the transformed subgraph, relational edges with kinship relationships are determined. Kinship relationships are the associations of data after processing methods such as copying, splicing, association, label calculation, desensitization, and anonymization. The kinship propagation characteristics include processing type, path length, field retention rate, aggregation granularity, publication scope, and prior evidence.
[0031] By combining the sensitivity scores of node semantic combination and the score of label graph for the bloodline propagation characteristics, the risk of a single bloodline propagation path (i.e. the risk of data object processing) is determined. By aggregating multiple paths, the comprehensive bloodline propagation risk can be obtained.
[0032] Optionally, the weights of the bloodline propagation features in the change subgraph are determined by a Sigmoid gating network. Based on the weights of the bloodline edges, the semantic combination sensitivity scores of the nodes and the label graph scores are weighted and summed to obtain the path risk of the bloodline edges.
[0033] For example, the gating value g_e = Sigmoid(w_l^Tv_e + b_l) is calculated using a Sigmoid gating network, where Tv_e represents the lineage propagation feature, w_l represents the base weight, and b_l represents the risk increment. The risk of a single path p is represented as R_p = R_u·∏_(e∈p)g_e·(1+ξ_cC_o+ξ_tT_o), where ξ_c and ξ_t represent the combined sensitivity score weight and the label graph score weight, respectively, and R_u represents the base lineage propagation risk ratio. Aggregating multiple lineage propagation paths yields the comprehensive lineage propagation risk of a node.
[0034] S107. Construct a data-sensitive evidence matrix based on the semantic evidence vector, the co-occurrence evidence, the tag graph score, and the node lineage transmission risk. Combine this with prior weight boundary constraints to determine the comprehensive risk and risk level classification of public security data.
[0035] The data sensitivity evidence matrix includes semantic evidence vectors, co-occurrence evidence, label graph scores, and node lineage propagation risk. Each item serves as an indicator for evaluating data sensitivity. Combined with dynamic weight values, a comprehensive risk can be calculated, and the risk level is determined based on the comprehensive risk value. The dynamic weight values are adaptive weight values under the constraints of the prior weight boundaries. Simultaneously, the risk category probability and uncertainty can be output. When there are conflicts in the category probabilities or the risk uncertainty is high, the existing data sensitivity level can be maintained, and the data can be transferred to manual review.
[0036] The prior weight boundary constraints are pre-set upper and lower weight limits. Under the constraints of the upper and lower weight limits, the dynamic weights can be determined by a deep learning model.
[0037] Optionally, the constrained dynamic weights of each element of the data-sensitive evidence matrix are generated based on the object context and the consistency requirements between evidence in the change subgraph.
[0038] It is understood that in this embodiment, the configuration of network-related parameters such as graph attention network and gating network, as well as the weight values in specific risk assessment, can be learned from historical review samples. Manual configuration only limits the edge type, the range of legal values, and the minimum sample size.
[0039] In this embodiment, based on semantic evidence vectors, the risk of public security data is comprehensively assessed through evidence from dimensions such as field co-occurrence, tag graph propagation, and kinship propagation. This can ensure the accuracy, reliability, and rationality of data security level classification and avoid inaccurate data sensitivity judgments caused by incomplete fields and static annotations.
[0040] In one embodiment, such as Figure 2 As shown, step S107, which determines the comprehensive risk and risk level classification of public security data, also includes...
[0041] S201. When the risk level classification signature verification passes, the risk level is mapped to field desensitization strategy, interface access permission, encrypted storage strategy, access audit strategy, export approval strategy and data sharing scope strategy and executed. S202. Construct and output the evidence chain containing object identifier, incremental version number, model version number, semantic evidence vector, contribution edge set, label propagation path, lineage propagation path, dynamic weight, risk level classification and policy execution receipt.
[0042] The risk level classification of public security data can include types such as internal, sensitive, highly sensitive, and core. If the risk level passes signature verification, the level will be mapped to field anonymization, interface permissions, encryption, auditing, export approval, and sharing scope policies. Simultaneously, object identifiers, incremental versions, model versions, semantic evidence, main contribution edges, tag propagation paths, lineage paths, dynamic weights, final levels, and policy execution receipts will be chained together to output an evidence chain.
[0043] In another embodiment, feedback samples are constructed based on policy execution receipts, false positive / false negative markers, and manual review labels, and can be back-connected to the change subgraph by object identifier (oid). These feedback samples are used to update node semantic vectors, graph attention parameters, and dynamic weights in the comprehensive risk calculation. This feedback loop allows the security execution results to serve as the basis for training and calibration in the next round of incremental classification.
[0044] In some embodiments, a deep learning model consisting of a semantic encoder, graph attention network, label graph gating network, kinship transmission gating network, and evidence fusion network can be used. After training the deep learning model based on historical samples, the metadata collected by the data platform is used to conduct a risk assessment and output the corresponding comprehensive risk score and risk level of public security data.
[0045] Loss functions for deep learning models can include classification loss, co-occurrence edge supervision loss, lineage propagation consistency loss, and output consistency loss within the same evidence chain. In the evidence fusion network, dynamic weights can be calculated based on the evidence vectors and determined using the Clamp function within validated upper and lower bounds. A non-zero lower bound is set for manually verified high-level lineage evidence, formal interface contracts, and historically verified evidence.
[0046] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.
[0047] Figure 3 This is a schematic diagram of a dynamic classification system for public security data provided in an embodiment of the present invention. The system includes: Data acquisition module 310 is used to collect metadata from the public security data platform and generate a version snapshot for the metadata; Optionally, a global object identifier, version number, collection timestamp, and source credibility can be set for each metadata object.
[0048] The change subgraph generation module 320 is used to determine the data change set based on the current version snapshot and historical version snapshots, and to perform a traversal search on the data change set in a heterogeneous graph composed of field lineage, interface reference relationship and field co-occurrence relationship to obtain the change subgraph; The semantic evidence construction module 330 is used to construct a desensitized context for each object in the change subgraph, obtain a node semantic vector by semantic encoding of the desensitized context, and construct a semantic evidence vector containing the node semantic vector, semantic candidate category probability and semantic uncertainty. The co-occurrence evidence construction module 340 is used to use the changed subgraph as the update node range, the semantic evidence vector as the node initialization value, construct the field co-occurrence graph and edge features, update the node relevance based on the edge features through a graph attention mechanism, and generate co-occurrence evidence composed of node semantic combination sensitivity score and main contribution edge. Optionally, the co-occurrence evidence construction module 340 includes: The co-occurrence frequency, semantic distance, call frequency, relation type, semantic vectors of both ends, and change type are encoded and concatenated into the edge features of the field co-occurrence graph. The node relevance after the edge features are normalized is calculated through the graph attention mechanism. Based on the node relevance, a Sigmoid gated network is used to calculate and output the node semantic combination sensitivity score. Relationship edges whose node semantic combination sensitivity scores exceed a preset threshold are selected to obtain the main contributing edges.
[0049] The tag graph reasoning module 350 is used to determine the node business tag based on the candidate category probability in the semantic evidence vector, and use the node business tag as a seed node to perform risk transmission calculation in the tag graph composed of tag nodes, node relationship types and node semantic combination sensitivity scores to obtain the tag graph score. Preferably, the tag map inference module 350 includes: When the semantic combination sensitivity score of the seed node or the probability of the semantic candidate category of the seed node exceeds a preset threshold, risk transmission calculation is triggered, and the weight of the relationship edge of the label node in the label graph is determined by the Sigmoid gating network to obtain the label graph score.
[0050] The bloodline propagation reasoning module 360 is used to determine the data bloodline relationship edges based on the changed subgraph, construct bloodline propagation features for each bloodline relationship edge, and use the node semantic combination sensitivity score and the label graph score as bloodline propagation enhancement coefficients to calculate the path risk of each bloodline relationship edge, and weight and aggregate the path risks to obtain the node bloodline propagation risk. Optionally, the bloodline propagation inference module 360 includes: The weights of the bloodline propagation features are determined by a Sigmoid gating network in the change subgraph. Based on the weights of the bloodline edges, the semantic combination sensitivity scores of the nodes and the label graph scores are weighted and summed to obtain the path risk of the bloodline edges.
[0051] The risk fusion output module 370 is used to construct a data-sensitive evidence matrix based on the semantic evidence vector, the co-occurrence evidence, the tag graph score, and the node lineage transmission risk, and to determine the comprehensive risk and risk level classification of public security data by combining prior weight boundary constraints.
[0052] Optionally, the constrained dynamic weights of each element of the data-sensitive evidence matrix are generated based on the object context and the consistency requirements between evidence in the change subgraph.
[0053] In one embodiment, the risk fusion output module 370 further includes...
[0054] The evidence chain generation module is used to map the risk level to field desensitization strategy, interface access permission, encrypted storage strategy, access audit strategy, export approval strategy, and data sharing scope strategy and execute them when the risk level classification signature verification passes; construct and output the evidence chain containing object identifier, incremental version number, model version number, semantic evidence vector, contribution edge set, label propagation path, lineage propagation path, dynamic weight, risk level classification, and strategy execution receipt.
[0055] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the system and modules described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.
[0056] Figure 4 This is a schematic diagram of an electronic device according to an embodiment of the present invention. The electronic device is used for public security data security classification. Figure 4 As shown, the electronic device 40 of this embodiment includes: a memory 410, a processor 420, and a system bus 430. The memory 410 includes an executable program 4101 stored thereon. As those skilled in the art will understand, Figure 4 The electronic device structure shown does not constitute a limitation on the electronic device and may include more or fewer components than shown, or combine certain components, or have different component arrangements.
[0057] The following is combined Figure 4 A detailed description of each component of the electronic device 40 is provided below: The memory 410 can be used to store software programs and modules. The processor 420 executes various functional applications and data processing of the electronic device by running the software programs and modules stored in the memory 410. The memory 410 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, application programs required for at least one function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the electronic device (such as cached data), etc. In addition, the memory 410 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device.
[0058] The memory 410 contains an executable program 4101 for a data classification method. This executable program 4101 can be divided into one or more modules / units, which are stored in the memory 410 and executed by the processor 420 to implement sensitive classification of public security data, etc. Each module / unit can be a series of computer program instruction segments capable of performing a specific function, describing the execution process of the executable program 4101 in the electronic device 40. For example, the executable program 4101 can be divided into functional modules such as a data acquisition module, a change subgraph generation module, a semantic evidence construction module, a co-occurrence evidence construction module, a tag graph reasoning module, a kinship transmission reasoning module, and a risk fusion output module.
[0059] The processor 420 is the control center of the electronic device. It connects various parts of the electronic device via various interfaces and lines. By running or executing software programs and / or modules stored in the memory 410, and by calling data stored in the memory 410, it performs various functions and processes data, thereby monitoring the overall status of the electronic device. Optionally, the processor 420 may include one or more processing units; preferably, the processor 420 may integrate an application processor and a modem processor, wherein the application processor mainly handles the operating system, application programs, etc., and the modem processor mainly handles wireless communication. It is understood that the modem processor may also not be integrated into the processor 420.
[0060] The system bus 430 is used to connect various functional components inside the computer, transmitting data, address, and control information. Its type can be, for example, a PCI bus, an ISA bus, or a CAN bus. Instructions from the processor 420 are transmitted to the memory 410 via the bus, and the memory 410 sends data back to the processor 420. The system bus 430 is responsible for data and instruction exchange between the processor 420 and the memory 410. Of course, the system bus 430 can also connect to other devices, such as network interfaces and display devices.
[0061] In this embodiment of the invention, the executable program executed by the processor 420 included in the electronic device includes: Collect metadata from the public security data platform and generate a version snapshot for the metadata; Based on the current version snapshot and historical version snapshots, determine the set of data changes, and perform a traversal search on the set of data changes in a heterogeneous graph composed of field lineage, interface reference relationship and field co-occurrence relationship to obtain the change subgraph; For each object in the change subgraph, a desensitization context is constructed. The desensitization context is semantically encoded to obtain a node semantic vector. A semantic evidence vector containing the node semantic vector, semantic candidate category probability, and semantic uncertainty is constructed. The changed subgraph is used as the updated node range, and the semantic evidence vector is used as the node initialization value. A field co-occurrence graph and edge features are constructed. Based on the edge features, the node relevance is updated through a graph attention mechanism to generate co-occurrence evidence composed of node semantic combination sensitivity scores and main contribution edges. Based on the candidate category probabilities in the semantic evidence vector, the node business label is determined. The node business label is used as a seed node. Risk transmission calculation is performed in the label graph composed of label nodes, node relationship types and node semantic combination sensitivity scores to obtain the label graph score. Based on the changed subgraph, the data lineage relationship edges are determined. Lineage propagation features are constructed for each lineage relationship edge. The node semantic combination sensitivity score and the label graph score are used as lineage propagation enhancement coefficients. The path risk of each lineage relationship edge is calculated. The path risks are weighted and aggregated to obtain the node lineage propagation risk. Based on the semantic evidence vector, the co-occurrence evidence, the tag graph score, and the node lineage transmission risk, a data-sensitive evidence matrix is constructed. Combined with prior weight boundary constraints, the comprehensive risk and risk level classification of public security data are determined.
[0062] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and modules described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.
[0063] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.
[0064] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A dynamic classification method for public security data, characterized in that, include: Collect metadata from the public security data platform and generate a version snapshot for the metadata; Based on the current version snapshot and historical version snapshots, determine the set of data changes, and perform a traversal search on the set of data changes in a heterogeneous graph composed of field lineage, interface reference relationship and field co-occurrence relationship to obtain the change subgraph; For each object in the change subgraph, a desensitization context is constructed. The desensitization context is semantically encoded to obtain a node semantic vector. A semantic evidence vector containing the node semantic vector, semantic candidate category probability, and semantic uncertainty is constructed. The changed subgraph is used as the updated node range, and the semantic evidence vector is used as the node initialization value. A field co-occurrence graph and edge features are constructed. Based on the edge features, the node relevance is updated through a graph attention mechanism to generate co-occurrence evidence composed of node semantic combination sensitivity scores and main contribution edges. Based on the candidate category probabilities in the semantic evidence vector, the node business label is determined. The node business label is used as a seed node. Risk transmission calculation is performed in the label graph composed of label nodes, node relationship types and node semantic combination sensitivity scores to obtain the label graph score. Based on the changed subgraph, the data lineage relationship edges are determined. Lineage propagation features are constructed for each lineage relationship edge. The node semantic combination sensitivity score and the label graph score are used as lineage propagation enhancement coefficients. The path risk of each lineage relationship edge is calculated. The path risks are weighted and aggregated to obtain the node lineage propagation risk. Based on the semantic evidence vector, the co-occurrence evidence, the tag graph score, and the node lineage transmission risk, a data-sensitive evidence matrix is constructed. Combined with prior weight boundary constraints, the comprehensive risk and risk level classification of public security data are determined.
2. The method according to claim 1, characterized in that, The process of collecting metadata from the public security data platform and generating version snapshots for the metadata also includes: Set a global object identifier, version number, collection timestamp, and source credibility for each metadata object.
3. The method according to claim 1, characterized in that, The construction of the field co-occurrence graph and edge features, and the updating of node relevance based on the edge features through a graph attention mechanism, generate co-occurrence evidence consisting of node semantic combination sensitivity scores and main contributing edges, including: The co-occurrence frequency, semantic distance, call frequency, relation type, semantic vectors of both ends, and change type are encoded and concatenated into the edge features of the field co-occurrence graph. The node relevance after the edge features are normalized is calculated through the graph attention mechanism. Based on the node relevance, a Sigmoid gated network is used to calculate and output the node semantic combination sensitivity score. Relationship edges whose node semantic combination sensitivity scores exceed a preset threshold are selected to obtain the main contributing edges.
4. The method according to claim 1, characterized in that, The step of using node business tags as seed nodes and performing risk transmission calculations in a tag graph composed of tag nodes, node relationship types, and node semantic combination sensitivity scores to obtain tag graph scores includes: When the semantic combination sensitivity score of the seed node or the probability of the semantic candidate category of the seed node exceeds a preset threshold, risk transmission calculation is triggered, and the weight of the relationship edge of the label node in the label graph is determined by the Sigmoid gating network to obtain the label graph score.
5. The method according to claim 1, characterized in that, The process of constructing bloodline propagation features for each bloodline edge, and using the node semantic combination sensitivity score and the label graph score as bloodline propagation enhancement coefficients to calculate the path risk of each bloodline edge includes: The weights of the bloodline propagation features are determined by a Sigmoid gating network in the change subgraph. Based on the weights of the bloodline edges, the semantic combination sensitivity scores of the nodes and the label graph scores are weighted and summed to obtain the path risk of the bloodline edges.
6. The method according to claim 1, characterized in that, The method of combining prior weight boundary constraints to determine the comprehensive risk and risk level classification of public security data also includes: Based on the object context and evidence consistency requirements in the change subgraph, the constrained dynamic weights of each element in the data-sensitive evidence matrix are generated.
7. The method according to claim 1, characterized in that, The determination of the comprehensive risk and risk level classification of public security data also includes: Once the risk level classification signature verification passes, the risk level is mapped to field desensitization strategy, interface access permission, encrypted storage strategy, access audit strategy, export approval strategy, and data sharing scope strategy and executed. Construct and output an evidence chain that includes object identifier, incremental version number, model version number, semantic evidence vector, contribution edge set, label propagation path, lineage propagation path, dynamic weight, risk level classification, and policy execution receipt.
8. A dynamic classification system for public security data, characterized in that, include: The data acquisition module is used to collect metadata from the public security data platform and generate version snapshots for the metadata. The change subgraph generation module is used to determine the set of data changes based on the current version snapshot and historical version snapshots. It then performs a traversal search on the set of data changes in a heterogeneous graph composed of field lineage, interface reference, and field co-occurrence relationships to obtain the change subgraph. The semantic evidence construction module is used to construct a desensitized context for each object in the change subgraph, obtain a node semantic vector by semantic encoding of the desensitized context, and construct a semantic evidence vector containing the node semantic vector, semantic candidate category probability and semantic uncertainty. The co-occurrence evidence construction module is used to use the changed subgraph as the update node range, the semantic evidence vector as the node initialization value, construct the field co-occurrence graph and edge features, update the node relevance based on the edge features through a graph attention mechanism, and generate co-occurrence evidence composed of node semantic combination sensitivity score and main contribution edge. The tag graph reasoning module is used to determine the node business tag based on the candidate category probability in the semantic evidence vector, and use the node business tag as a seed node to perform risk transmission calculation in the tag graph composed of tag nodes, node relationship types and node semantic combination sensitivity scores to obtain the tag graph score. The bloodline propagation reasoning module is used to determine the data bloodline relationship edges based on the changed subgraph, construct bloodline propagation features for each bloodline relationship edge, and use the node semantic combination sensitivity score and the label graph score as bloodline propagation enhancement coefficients to calculate the path risk of each bloodline relationship edge. The path risks are then weighted and aggregated to obtain the node bloodline propagation risk. The risk fusion output module is used to construct a data-sensitive evidence matrix based on the semantic evidence vector, the co-occurrence evidence, the tag graph score, and the node lineage transmission risk, and to determine the comprehensive risk and risk level classification of public security data by combining prior weight boundary constraints.
9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of a dynamic classification method for public security data as described in any one of claims 1 to 7.
10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed, it implements the steps of a dynamic classification method for public security data as described in any one of claims 1 to 7.
