Government affair knowledge base collaborative management method and system based on multi-level permission model

Abstract

The invention provides a government affair knowledge base collaborative management method and system based on a multi-level permission model. The method comprises the following steps: firstly, establishing a multi-dimensional semantic association model by analyzing a logical relationship among entries, implementing secret state operation on an association strength calculation process by adopting a homomorphic encryption technology, mapping an encrypted network topology to a multi-dimensional vector space, and generating distributed vector representation; and based on topological distribution characteristics of a vector space, detecting a non-sensitive entry set which is in implicit association with dominant sensitive entries by using a graph neural network, and revealing a hidden sensitive information transmission link in the knowledge graph. And aiming at the output request, establishing an authority dynamic adaptation mechanism, and performing hierarchical shielding on the associated content in the non-sensitive set to realize dynamic balance of security protection and data utility. According to the technical scheme provided by the invention, the implicit sensitive association of the government affair knowledge base is accurately identified through encryption modeling and vector analysis, and the balance of safety and availability is realized by means of a semantic masking technology of dynamic permission adjustment.

Description

Technical Field

[0001] The present application relates to the field of collaborative management technology, and in particular to a collaborative management method and system for a government knowledge base based on a multi-level authority model. Background Art

[0002] In the collaborative management of government knowledge bases, with the deepening demand for cross-departmental business integration and data sharing, multi-dimensional permission management scenarios face three core challenges. First, dynamic permission management within a multi-level organizational structure requires fine-grained access control from the provincial to the community level. Second, permission mapping between business flows and data flows requires millisecond-level responses to support sudden cross-departmental collaboration. Third, full-link traceability of sensitive operations requires breaking through the traditional log audit model to achieve causal analysis between permission changes and data access.

[0003] The current mainstream solution uses an attribute-based access control model. This model dynamically analyzes user and environmental attributes through policy decision points and generates access rules based on the sensitivity labels of knowledge base resources. The system uses a graph database to construct a permission relationship network, utilizes a computing engine to process permission change events, and records operation traces through a distributed logging system.

[0004] This solution has significant flaws. Dynamic attribute updates experience delays in the order of seconds, resulting in invalid permissions synchronization during cross-departmental collaboration. Policy rules are too tightly coupled with business scenarios, requiring the manual adjustment of hundreds of policies for new businesses, exponentially increasing operational complexity. The performance of graph database relationship traversal plummets when the number of nodes exceeds 10,000, impacting the efficiency of permission verification during sudden, large-scale collaboration. Summary of the Invention

[0005] The present application provides a collaborative management method and system for a government knowledge base based on a multi-level authority model to solve the problem of inconvenient management in the existing technology.

[0006] In a first aspect, the present application provides a collaborative management method for a government knowledge base based on a multi-level authority model, including:

[0007] In the government knowledge base, a semantic association network between knowledge items is established by analyzing the logical dependencies between items and the sensitive information propagation paths;

[0008] Performing homomorphic encryption on the computation process of characterizing the strength of associations between items in the semantic association network, and mapping the association structure between knowledge items in the encrypted semantic association network into a multidimensional vector space to generate a distributed vector representation reflecting the potential association characteristics between the knowledge items;

[0009] Detecting a set of non-sensitive items that are implicitly associated with explicit sensitive items based on the topological distribution of the distributed vector representations in the semantic space;

[0010] In response to a data output request from the government knowledge base, the content masking range of the non-sensitive item set is dynamically adjusted based on the authority level. The process of adjusting the content masking range of the non-sensitive item set includes: performing semantic-preserving fuzzification operations on associated content that exceeds the authority, and retaining the original semantic structure of associated content within the authority.

[0011] Optionally, in response to a data output request from the government affairs knowledge base, dynamically adjusting the content masking range of the non-sensitive item set based on the permission level includes:

[0012] Parse the permission level identifier of the data output request of the government knowledge base, and extract the sensitive item filtering rules and semantic fuzzification intensity parameters associated with the permission level;

[0013] Based on the semantic fuzzification strength parameter, the maximum association depth threshold allowed by the current permission level is extracted from the semantic association network, and according to the sensitive item filtering rule, the association path in the non-sensitive item set that has three or more levels of indirect references to the explicit sensitive item is located;

[0014] Dividing the associated content of the non-sensitive item set into an accessible subset within the authority and a blocked subset due to unauthorized access according to the maximum association depth threshold and the sensitive item filtering rule;

[0015] For the entry contents in the subset that needs to be shielded due to unauthorized access, a semantically-preserving fuzzification operation is used to retain the original semantic features of the accessible subset within the authority, and only the associated paths that cross the authority boundary are fuzzified.

[0016] Optionally, a semantically-preserving fuzzification operation is applied to the content of the entries in the unauthorized subset to be masked, including:

[0017] Extracting association path segments including entity names, attribute values ​​and relationship predicates from the unauthorized masking subset;

[0018] According to the preset entity generalization rules, the entity name is replaced with the corresponding category label, and the attribute value is converted into an interval range or type identifier to obtain the generalized entity name and attribute value;

[0019] Extracting the main action of the relational predicate, deleting the modifying words, and generating a simplified predicate expression;

[0020] The generalized entity names and attribute values ​​are reorganized with the simplified predicate expressions according to the original syntactic structure to form grammatically complete but semantically ambiguous entry content;

[0021] A permission violation identifier is embedded in the entry content, where the permission violation identifier includes an encrypted hash value of the original path and an obfuscated operation type tag.

[0022] Optionally, retaining the original semantic features of the accessible subset within the permission and only performing fuzzification processing on the association path that crosses the permission boundary includes:

[0023] In the accessible subset within the permission, detect the association path with explicit sensitive entries that cross the permission boundary;

[0024] Extracting the entity name and relationship predicate in the association path, retaining the original entity name but replacing the entity name attribute value with a type identifier, retaining the relationship predicate trunk but deleting modifiers involving sensitive references in the path, and obtaining a processed cross-authority boundary path;

[0025] The processed cross-boundary path is spliced ​​with the original content retained within the permission to ensure the semantic integrity of the non-cross-boundary path;

[0026] In the spliced ​​content, a permission isolation mark is added to the cross-permission boundary path segment, and the permission isolation mark includes the encrypted check value and the fuzzification strength level of the path.

[0027] Optionally, performing homomorphic encryption processing on the calculation process of characterizing the strength of association between items in the semantic association network, and mapping the association structure between knowledge items in the encrypted semantic association network into a multidimensional vector space to generate a distributed vector representation reflecting the potential association characteristics between knowledge items, including:

[0028] Based on the logical dependencies between knowledge items and the sensitive information propagation paths, an initial correlation strength value is generated between each pair of items. The initial correlation strength value is calculated based on the citation frequency, dependency level, and semantic overlap.

[0029] Encrypting the initial association strength value using a homomorphic encryption algorithm so that the encrypted association strength value retains addition and multiplication operation characteristics;

[0030] In an encrypted state, traversing the multi-hop association paths between all entries in the semantic association network, performing attenuation weighting on the encrypted association strength value on each multi-hop association path according to the path length, and generating an encrypted weighted path strength;

[0031] For each knowledge item, aggregate the encrypted weighted path strengths of all paths ending at the knowledge item to obtain an encrypted global association strength value;

[0032] Based on the encrypted global association strength value, extract the local density weight and cross-layer jump frequency of each entry in the encrypted association network, wherein the local density weight is calculated by summing the encrypted association strengths of the entries within a preset depth, and the cross-layer jump frequency is generated by counting the number of association paths between non-contiguous layers of the entries;

[0033] The local density weight and the cross-layer jump frequency are fused into an entry-level topological coefficient according to a preset ratio, and the topological coefficients of adjacent entries are weightedly superimposed in an encrypted state through an iterative diffusion algorithm to generate a dynamic propagation vector for the entry;

[0034] orthogonally concatenating the dynamic propagation vector with the encrypted content vector of the entry, wherein the encrypted content vector is generated by extracting homomorphic encryption features of the entry attributes and text description;

[0035] According to the relative distance of the concatenated vectors in the multidimensional space, the weight of each dimension is adjusted to generate a distributed vector representation that reflects the potential correlation characteristics between knowledge items.

[0036] Optionally, based on the topological distribution of the distributed vector representation in the semantic space, detecting a set of non-sensitive items that are implicitly associated with explicit sensitive items includes:

[0037] Based on the final coordinates represented by the distributed vector, taking the coordinates of the explicit sensitive entry as the center, calculating the Euclidean distance between the coordinates of all non-sensitive entries and the center, and screening the non-sensitive entries whose distance is less than a first dynamic threshold as a primary candidate set;

[0038] Traversing all possible conduction paths of each non-sensitive entry in the primary candidate set in the encrypted association strength matrix, retaining only conduction paths containing three or more intermediate entries, and extracting ciphertext association weight values ​​in the conduction paths;

[0039] A cumulative multiplication operation is performed on the ciphertext association weight value of each conduction path, and then the ciphertext accumulation of the operation results of all paths is performed. If the decrypted superposition value exceeds the second dynamic threshold, it is determined that the non-sensitive entry has a hidden association with the explicit sensitive entry, and the non-sensitive entry is included in the target non-sensitive entry set.

[0040] Optionally, a semantic association network between knowledge items is established by analyzing the logical dependencies between items and the sensitive information propagation paths, including:

[0041] Construct a citation tracking model for cross-departmental government clauses, analyze direct citation and nested citation relationships based on normative citation annotations in clause texts, and generate a set of logical dependency edges;

[0042] Analyze the task node topology of administrative process documents, extract the timing trigger conditions and data dependencies between process nodes based on natural language processing, and model the node pairs that meet the preset connection rules as administrative process connection edges;

[0043] Constructing an initial semantic association network based on the logical dependency edge set and the administrative process connection edge;

[0044] Based on the sensitive item label database, a back-propagation algorithm is used to trace back layer by layer along the conduction path of the initial semantic association network to identify the indirect diffusion path of confidential data through non-sensitive items, and the residual strength value of sensitive information on each conduction path is calculated using the association strength attenuation function;

[0045] The hierarchical relationship of the logical dependency edge set, the dynamic weight parameters of the administrative process connection edges, and the residual strength values ​​are integrated into a model to generate a semantic association network with a weighted directed graph structure.

[0046] Optionally, according to the maximum association depth threshold and the sensitive item filtering rule, the associated content of the non-sensitive item set is divided into an accessible subset within the authority and a blocked subset due to unauthorized access, including:

[0047] Recursively traverse the semantic association paths of the non-sensitive entry set and set a path length counter on each conductive path;

[0048] Based on the path length counter, when an indirect reference relationship with an explicit sensitive entry is detected, a path backtracking mechanism is started and the number of intermediate nodes is recorded to obtain a path backtracking result;

[0049] Inputting the path backtracking result into the association strength attenuation function to perform weight correction to generate a corrected association strength value;

[0050] If the corrected association strength value is higher than the visibility threshold corresponding to the current permission level and the path length does not exceed the maximum association depth, the current entry is classified into the accessible subset within the permission level, otherwise it is classified into the unauthorized subset that needs to be blocked.

[0051] In a second aspect, the present application provides a collaborative management system for a government knowledge base based on a multi-level authority model, including:

[0052] Establish a module for establishing a semantic association network between knowledge items in the government knowledge base by analyzing the logical dependencies between items and the sensitive information propagation paths;

[0053] a generation module configured to perform homomorphic encryption on a computation process representing the strength of associations between items in the semantic association network, and to map the association structure between knowledge items in the encrypted semantic association network into a multidimensional vector space to generate a distributed vector representation reflecting potential association characteristics between knowledge items;

[0054] a detection module, configured to detect a set of non-sensitive items that are implicitly associated with explicit sensitive items based on a topological distribution of the distributed vector representations in a semantic space;

[0055] A response module is used to respond to data output requests from the government knowledge base and dynamically adjust the content masking range of the non-sensitive item set based on the authority level. The process of adjusting the content masking range of the non-sensitive item set includes: performing semantic-preserving fuzzification operations on associated content that exceeds the authority, and retaining the original semantic structure of associated content within the authority.

[0056] In a third aspect, an embodiment of the present application provides a computing device comprising a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement a collaborative management method of a government knowledge base based on a multi-level authority model as described in the first aspect above.

[0057] This application establishes a semantic association network by analyzing the logical dependencies between items in the government knowledge base and the sensitive information propagation paths. Its technical effect is to achieve systematic modeling of explicit and implicit associations within the knowledge system, breaking through the recognition limitations of traditional association analysis methods based on keyword matching in complex semantic scenarios. By using homomorphic encryption technology to process the association strength calculation process and mapping the encrypted network to a multidimensional vector space, its technical effect is to generate a distributed vector representation with potential association characteristics while ensuring data privacy and security, providing a quantitative analysis basis for sensitive information detection. Based on the topological distribution of the distributed vector representation, a set of non-sensitive items that are implicitly associated with explicit sensitive items is detected. Its technical effect is to establish an automatic identification mechanism for multi-level sensitive information propagation paths, significantly improving the ability to discover potential information leakage risks. The content masking range of the non-sensitive item set is dynamically adjusted by permission level and a semantic-preserving fuzzification operation is implemented. Its technical effect is to achieve fine-grained adaptive access control, maintaining the semantic coherence of the knowledge base while ensuring data security.

[0058] Furthermore, by parsing the permission level identifier of the data output request, sensitive item filtering rules and semantic fuzzification intensity parameters are extracted. Based on the parameter settings, the maximum association depth threshold allowed by the permission is extracted from the semantic association network. The technical effect is to establish a configurable hierarchical access control mechanism to achieve differentiated security protection. By locating the association paths in the non-sensitive item set that have three or more levels of indirect references to explicit sensitive items and dividing them into accessible subsets within the permission and unauthorized subsets that need to be shielded, the technical effect is to achieve accurate identification and classification management of associations across permission boundaries. The semantically preserving fuzzification technology is used to process the associated content in the unauthorized subset that needs to be shielded. The technical effect is to maximize the preservation of the semantic value of knowledge while ensuring the protection of sensitive information, significantly improving the security and availability of government knowledge sharing. By dynamically adjusting the shielding range and fuzzification intensity, the technical effect is to establish an adaptive security protection system that effectively balances the conflicting needs between data security and knowledge availability.

[0059] These and other aspects of the present application will become more readily apparent from the description of the following embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0061] Figure 1 A flowchart of a collaborative management method for a government affairs knowledge base based on a multi-level authority model provided by the present application is shown;

[0062] Figure 2 A scenario diagram of a government affairs knowledge base collaborative management system based on a multi-level authority model provided by this application is shown;

[0063] Figure 3 A schematic diagram of the structure of a government affairs knowledge base collaborative management system based on a multi-level authority model provided by the present application is shown;

[0064] Figure 4 A schematic structural diagram of a computing device provided by the present application is shown. DETAILED DESCRIPTION

[0065] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0066] In some of the processes described in the specification and claims of this application and the above-mentioned figures, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this document or may be executed in parallel. The serial numbers of the operations, such as 101, 102, etc., are only used to distinguish between different operations, and the serial numbers themselves do not represent any order of execution. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to being different types.

[0067] Research has found that in the collaborative management of government knowledge bases, existing attribute-based access control models face three key bottlenecks: first, delayed attribute updates lead to invalid permissions synchronization, resulting in data access conflicts during cross-departmental collaboration; second, policy rules are tightly coupled with business scenarios, requiring the manual configuration of numerous policies for new businesses, exponentially increasing operational costs; and third, the relationship traversal performance of graph databases plummets when the number of nodes exceeds 10,000, making it difficult to support permission verification for sudden, large-scale collaboration. These shortcomings stem from the lack of adaptability of static permission models to dynamic business environments and the performance limitations of centralized architectures in complex relational computations.

[0068] In response to the above problems, this application proposes a collaborative management method based on semantic association network and dynamic vectorized permission mapping. The core of this method is to construct an encrypted semantic association network between knowledge items and realize fine-grained dynamic control of permissions through distributed vector space. Specifically, the association strength calculation of entries is processed by homomorphic encryption, the semantic network is mapped to a multi-dimensional vector space to generate a potential association feature representation, and semantic-preserving fuzzification operations are implemented based on the permission level. This method breaks through the static policy limitations of the traditional attribute model, realizes millisecond-level dynamic adjustment through vectorized permission mapping, and solves the problem of permission synchronization delay; through the automated policy generation of the semantic association network, the amount of manual rule configuration is greatly reduced, and the distributed vector calculation replaces the graph traversal, which makes the permission verification efficiency under more than 10,000 nodes more than 5 times, fundamentally solving the performance defects and operation and maintenance problems of the existing technology in dynamic collaboration scenarios, and significantly improving the security and collaborative efficiency of government knowledge sharing.

[0069] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without making creative efforts are within the scope of protection of this application.

[0070] Figure 1A flowchart of a collaborative management method for a government knowledge base based on a multi-level authority model is provided for the embodiment of this application. Figure 1 As shown, the method includes:

[0071] 101. In the government knowledge base, a semantic association network between knowledge items is established by analyzing the logical dependencies between items and the sensitive information transmission paths;

[0072] Optionally, step 101 includes:

[0073] 1011. Construct a citation tracking model for cross-departmental government clauses. Based on the normative citation annotations in the clause text, analyze direct citation and nested citation relationships and generate a set of logical dependency edges.

[0074] 1012. Analyze the task node topology of administrative process documents, extract the timing trigger conditions and data dependencies between process nodes based on natural language processing, and model the node pairs that meet the preset connection rules as administrative process connection edges;

[0075] 1013. Constructing an initial semantic association network based on the logical dependency edge set and the administrative process connection edge;

[0076] 1014. Based on the sensitive item label database, a back-propagation algorithm is used to trace back layer by layer along the conduction path of the initial semantic association network to identify the indirect diffusion path of confidential data through non-sensitive items, and the residual strength value of sensitive information on each conduction path is calculated using the association strength attenuation function;

[0077] 1015. The hierarchical relationship of the logical dependency edge set, the dynamic weight parameters of the administrative process connection edge, and the residual strength value are integrated into a model to generate a semantic association network with a weighted directed graph structure.

[0078] In the above scheme, the logical dependency edge set refers to the network edge set formed by parsing the normative reference annotations in the text of cross-departmental government affairs clauses, the nested reference relationship refers to the indirect reference chain formed through the intermediate clauses, the preset connection rules refer to the pre-defined node connection logical verification conditions, the initial semantic association network refers to the basic network structure constructed by integrating logical dependency edges and administrative process connection edges, the sensitive item label database refers to a special database that stores the identification of confidential items and their associated characteristics, the back propagation algorithm refers to the algorithm for tracing the propagation path of sensitive information in reverse along the network edge, the transmission path refers to the transmission link of sensitive information between items, the association strength attenuation function refers to the mathematical function that gradually reduces the association strength according to the path length, the residual strength value refers to the quantitative value of the remaining association after the sensitive information is transmitted through the non-sensitive item, and the weighted directed graph structure refers to a network topology model that contains directionality and weight attributes.

[0079] In the embodiment of the present application, first, a reference tracking model for cross-departmental government clauses is constructed through step 1011, the normative reference annotations in the clause text are parsed, and direct reference and nested reference relationships are identified. For example, if the statement "in accordance with Article X of the Tax Administration Regulations" appears in a certain environmental protection clause, a regular expression is used to match the name and clause number of the referenced tax clause to generate a direct logical dependency edge from the environmental protection clause to the tax clause. If the tax clause references the fiscal clause again, a nested reference edge from the fiscal clause to the tax clause is further established to form a three-level dependency chain. All reference relationships are stored in a logical dependency edge set, and each edge records the reference type, hierarchical depth, and text location information.

[0080] Next, in step 1012, the task node topology of the administrative process document is parsed, and dependency parsing techniques are used to extract the timing trigger conditions and data dependencies between process nodes. For example, in a project approval process document, "completion of preliminary material review" is identified as a prerequisite for "initiation of expert review," and an administrative process connection edge is established between the two nodes through conditional triggering rules. At the same time, it is detected that "expert review results" are a data input dependency of "results disclosure," and a connection edge with a data flow indicator is generated. All node pairs that meet the preset connection rules are modeled as administrative process connection edges, and the edge attributes include the trigger condition expression and the data dependency type.

[0081] Next, in step 1013, the logical dependency edge set is integrated with the administrative process connection edges to construct an initial semantic association network. For example, a cross-domain connection is established between the logical dependency edge of the environmental protection clause and the preliminary material review node in the project approval process, forming a hybrid semantic edge. The network is stored in a directed graph structure, with nodes containing clause text summaries and process node descriptions. Edge attributes integrate reference levels, process trigger conditions, and data dependency types.

[0082] Then, in step 1014, backpropagation analysis is performed based on the sensitive item label database. For example, after identifying sensitive items such as social security data, the network is backtracked layer by layer along the outgoing direction of the initial semantic association network. When a three-level transmission path of social security data -> demographics -> regional planning is detected, the association strength decay function is used to calculate the residual strength value of the sensitive information. The specific process is: the initial path strength is set to 1.0, and an exponential decay coefficient of 0.6 is applied every time a non-sensitive item is passed through. The residual strength value after the three-level transmission is 1.0×0.6. 3 = 0.216. The residual strength values ​​of all conductive paths are stored in the path attribute table.

[0083] Finally, step 1015 integrates the hierarchical relationships of logical dependency edges, the dynamic weight parameters of administrative process connection edges, and the residual strength values ​​to generate a semantic association network with a weighted directed graph structure. For example, for the edge connecting environmental protection clauses and tax clauses, the final weight is calculated as follows: logical reference strength (0.8) × process dynamic weight (0.7) + residual strength value (0.216) = 0.776. The network uses a dual storage structure of adjacency matrix and adjacency table, supporting fast traversal and multi-dimensional query.

[0084] In actual application, taking a city's environmental protection penalty regulations as an example, when the system parsed the normative reference annotations in the text of the clause, it was found that it explicitly quoted Article 15 of the "Tax Management Law". The reference relationship was accurately extracted through regular expressions, and a direct logical dependency edge from the environmental protection clause to the tax clause was generated, with an initial weight set to 0.8. Further analysis found that the tax clause nestedly quoted Article 32 of the "Fiscal Budget Regulations", forming a three-level dependency chain. According to the preset hierarchical attenuation coefficient of 0.7, the weight of the nested reference path was calculated to be 0.8×0.7 2 =0.392, reflecting the attenuation effect of cross-level references. In administrative process documents, natural language processing technology extracts key node relationships and identifies that "the completion of the initial review of materials" requires the condition that the completeness of the materials is greater than 90% to trigger the "initiation of expert review". Based on this, an administrative process connection edge with a conditional expression is established, and "expert review results" are marked as the data input dependency of "penalty disclosure". By tracing back the transmission path of the sensitive item "personal social security information" along the semantic network through the back propagation algorithm, it is found that it spreads to the enterprise credit disclosure system through non-sensitive items such as demographics and environmental protection penalties. The exponential decay function is used to calculate the residual strength value of the three-level transmission path: the initial strength is 1.0, the attenuation coefficient of each level is 0.6, and the final residual strength is 0.6 3 = 0.216, marking the path as a high-risk link. Finally, the hierarchical weights of the logical dependency edges, the dynamic parameters of the process connection edges, and the residual strength values ​​are integrated to construct a semantic network with a weighted directed graph structure. For example, the combined weight of the edges connecting environmental protection clauses and tax clauses is 0.392 (logical dependency) + 0.3 (process dynamic weight) = 0.692.

[0085] In the aforementioned step 101, a semantic association network integrating logical dependencies and dynamic weights is established by constructing a cross-departmental citation tracking model and analyzing administrative process topology. Citation relationship analysis accurately captures the explicit logical chain between clauses, while administrative process edges characterize the dynamic interactions between task nodes. The backpropagation algorithm effectively traces the implicit transmission paths of sensitive information. Combining hierarchical relationship attenuation and residual strength calculation, a weighted directed graph structure is generated to fully represent the explicit associations and potential transmission risks of knowledge items, laying a multi-dimensional network foundation for subsequent cryptographic analysis and association mining.

[0086] 102. Performing homomorphic encryption on the calculation process representing the strength of association between items in the semantic association network, and mapping the association structure between knowledge items in the encrypted semantic association network into a multidimensional vector space to generate a distributed vector representation reflecting the potential association characteristics between the knowledge items;

[0087] Optionally, step 102 includes:

[0088] 1021. Based on the logical dependencies between knowledge items and the sensitive information propagation paths, generate an initial correlation strength value between each pair of items. The initial correlation strength value is calculated based on the citation frequency, dependency level, and semantic overlap.

[0089] 1022. Encrypt the initial association strength value using a homomorphic encryption algorithm so that the encrypted association strength value retains addition and multiplication operation characteristics;

[0090] 1023. In an encrypted state, traverse the multi-hop association paths between all entries in the semantic association network, perform attenuation weighting on the encrypted association strength value on each multi-hop association path according to the path length, and generate an encrypted weighted path strength;

[0091] 1024. For each knowledge item, aggregate the encrypted weighted path strengths of all paths ending at the knowledge item to obtain an encrypted global association strength value;

[0092] 1025. Based on the encrypted global association strength value, extract the local density weight and cross-layer jump frequency of each entry in the encrypted association network, wherein the local density weight is calculated by summing the encrypted association strengths of the entries within a preset depth, and the cross-layer jump frequency is generated by counting the number of association paths between non-consecutive layers of the entries;

[0093] 1026. The local density weight and the cross-layer jump frequency are integrated into an entry-level topological coefficient according to a preset ratio, and the topological coefficients of adjacent entries are weightedly superimposed in an encrypted state through an iterative diffusion algorithm to generate a dynamic propagation vector of the entry.

[0094] 1027. Orthogonally concatenate the dynamic propagation vector with the encrypted content vector of the entry, where the encrypted content vector is generated by extracting homomorphic encryption features of the entry attributes and text description;

[0095] 1028. According to the relative distance of the concatenated vectors in the multidimensional space, the weight of each dimension is adjusted to generate a distributed vector representation that reflects the potential correlation characteristics between knowledge items.

[0096] In the above scheme, the initial association strength value refers to the basic association degree between entries calculated by comprehensively calculating the frequency of reference, dependency level, and semantic overlap. The homomorphic encryption algorithm refers to an encryption method that supports addition and multiplication operations in the encrypted state. The multi-hop association path refers to the association link spanning multiple intermediate entries. The encrypted weighted path strength refers to the ciphertext value of the encrypted initial association strength in the multi-hop path, weighted by the path length. The encrypted global association strength value refers to the ciphertext sum of the encrypted path strengths pointing to the target entry. The local density weight refers to the accumulated ciphertext value of the association strength of the entries within the preset network depth. The cross-layer jump frequency refers to the number of times the entries establish associations across non-contiguous layers. The entry-level topology coefficient refers to a comprehensive indicator that integrates the local density weight and the cross-layer jump frequency. The iterative diffusion algorithm refers to a calculation method that updates the topology coefficient by transferring the weights of adjacent nodes in the encrypted state. The dynamic propagation vector refers to the vectorized representation of the topological features of the entry in the encrypted network. The encrypted content vector refers to the text semantic feature vector processed by homomorphic encryption. Orthogonal splicing refers to the operation of merging the dynamic propagation vector and the encrypted content vector in orthogonal dimensions.

[0097] In the embodiment of the present application, first, the initial association strength value between each pair of knowledge items is calculated through step 1021. Specifically, it includes counting the direct citation frequency between clauses. For example, if the environmental protection clause is cited by three clauses, the basic citation frequency value is 3. Combined with the dependency level attenuation factor, for example, the attenuation coefficient of 0.8 is applied to the secondary citation, and the actual contribution value is 3×0.8=2.4. Then, the cosine similarity of the 768-dimensional vector of the clause text is calculated through the BERT model. For example, the semantic similarity between the environmental protection clause and the tax clause is 0.75. The final initial association strength value = 2.4×0.75=1.8 is obtained to reflect the explicit logical association strength between the clauses.

[0098] Next, the initial correlation strength value is encrypted using the Paillier homomorphic encryption algorithm in step 1022. A public key pk and a private key sk are generated. The private key is distributed to multiple participants in a secret sharing format to ensure that a single node cannot decrypt the data. An encryption operation is performed on the value 1.8, for example, to generate the ciphertext E(1.8) = g 1.8 ·r n modn 2 , where g is the generator, r is the random number, and n is the key parameter. The encrypted ciphertext retains the homomorphic properties of addition and multiplication, for example, E(1.8)+E(0.6)=E(2.4), supporting subsequent ciphertext calculations.

[0099] Next, in step 1023, the multi-level transmission path in the semantic association network is traversed in an encrypted state. For example, the three-level path of environmental protection clauses -> tax clauses -> financial clauses is detected, the path length is 2, and the exponential decay coefficient is 0.7.2 =0.49. A homomorphic multiplication operation is performed on the encrypted association strength on the path, for example, E(1.8)×E(0.75)×0.49=E(0.66), and the result is stored as the weighted path strength feature.

[0100] Then, in step 1024, the encrypted global association strength value of each knowledge item is aggregated. For example, the fiscal terms node receives a path E(0.74) from the environmental terms and a direct path E(1.2) from the economic terms, which are aggregated to E(1.94) through homomorphic addition. The global strength value is normalized based on the number of incoming edges at the node. For example, if the node has 5 incoming edges, the normalization coefficient is 1 / 5 to ensure unbiased centrality calculation.

[0101] Then, the local density weight and cross-layer jump frequency are extracted in step 1025. The local density weight is calculated as the homomorphic accumulation of all encrypted association strengths within three hops of the node. For example, the local density of the environmental protection clause is E(5.6). The cross-layer jump frequency counts the number of transmission paths that span more than two levels. For example, if the environmental protection clause has three cross-layer paths, such as clause -> process -> department, the two are fused in a 7:3 ratio to generate the item-level topology coefficient.

[0102] Then, a dynamic propagation vector is generated through step 1026. The topological coefficients are input into the graph convolutional network (GCN). For example, the topological coefficient E (4.62) of the environmental protection node is superimposed with the adjacent tax node E (3.8) with a weight of 0.6. After iterative diffusion, a 128-dimensional dynamic propagation vector E (6.9) is generated, which represents the global propagation influence of the node in the encrypted network.

[0103] Then, orthogonal feature concatenation is performed in step 1027. The dynamic propagation vector and the encrypted content vector are orthogonally concatenated, where the encrypted content vector refers to the 768-dimensional semantic vector generated by Sentence-BERT. The Gram-Schmidt algorithm is used to eliminate dimensional redundancy. For example, if the first three dimensions of the dynamic propagation vector are [0.8, 0.6, 0.4] and the first three dimensions of the semantic vector are [0.2, 0.7, 0.5], the resulting concatenation forms an 896-dimensional composite vector, ensuring that there is no redundancy between the structural and semantic features.

[0104] Finally, the multidimensional space weights are adjusted in step 1028. The t-SNE algorithm is used to reduce the 896-dimensional vector to 32 dimensions. For example, the vector distance between environmental protection clauses and tax clauses is optimized from 1.2 to 0.3. The weights of each dimension are dynamically adjusted based on cosine similarity. For example, the weight coefficient of dimension 3 is increased from 0.5 to 0.8, allowing semantically related items to be clustered in the space.

[0105] In practical applications, for the initial correlation strength value of 1.8 between environmental protection clauses and tax clauses, the Paillier homomorphic encryption algorithm is used to generate the ciphertext E(1.8)=g1.8 ·r n modn 2 , where the public key parameter n = 2048 and the random number r = 12345. When a three-level transmission path from environmental protection clauses to financial clauses is detected, a path attenuation coefficient of 0.7 is applied to the encryption association strength values ​​E(1.8) and E(0.75) on the path. 2 =0.49, and the weighted path strength calculated by homomorphic multiplication is E(1.8×0.75×0.49)=E(0.66). Aggregating all incoming path strengths of the fiscal clause node, including the environmental clause path E(0.66) and the direct path E(1.2) of the economic clause, homomorphic addition is used to obtain the global association strength E(1.86). The three-hop local density weight E(5.6) and the cross-layer hop frequency of the environmental clause are extracted and fused in a 7:3 ratio to generate the topological coefficient E(5.6×0.7+3×0.3)=E(4.62). After iterative diffusion through a graph convolutional network, a 128-dimensional dynamic propagation vector is generated. This vector is orthogonally concatenated with the 768-dimensional semantic vector generated by Sentence-BERT to form an 896-dimensional composite vector. The dimensionality is reduced to 32 dimensions using the t-SNE algorithm, optimizing the vector distance between the environmental and tax clauses from 1.2 to 0.3, accurately reflecting the semantic association.

[0106] In the solution described in step 102, privacy-preserving vectorized representation of the association characteristics of knowledge items is achieved by calculating multi-hop path strengths in a homomorphically encrypted state and integrating them with topological features. Encryption strength calculations ensure the computational security of sensitive association data, dynamic propagation vectors capture the implicit propagation influence of items in the encrypted network, and orthogonalization ensures the independent expression of semantic and topological features. The resulting distributed vector representation effectively preserves the underlying association patterns between items within the encrypted space, providing a computable and secure feature space for hidden relationship detection.

[0107] 103. Detecting a set of non-sensitive items that are implicitly associated with explicit sensitive items based on the topological distribution of the distributed vector representation in the semantic space;

[0108] Optionally, step 103 includes:

[0109] 1031. Based on the final coordinates represented by the distributed vector, with the coordinates of the explicit sensitive entry as the center, calculate the Euclidean distance between the coordinates of all non-sensitive entries and the center, and select non-sensitive entries with a distance less than a first dynamic threshold as a primary candidate set;

[0110] 1032. Traverse all possible transmission paths of each non-sensitive entry in the primary candidate set in the encrypted association strength matrix, retain only transmission paths containing three or more intermediate entries, and extract the ciphertext association weight value in the transmission path;

[0111] 1033. Perform a cumulative multiplication operation on the ciphertext association weight value of each conduction path, and then perform ciphertext accumulation on the operation results of all paths. If the decrypted superposition value exceeds a second dynamic threshold, it is determined that the non-sensitive entry has a hidden association with the explicit sensitive entry, and the non-sensitive entry is included in the target non-sensitive entry set.

[0112] In the above scheme, the first dynamic threshold refers to the Euclidean distance critical value adaptively adjusted according to the vector space density, the conduction path refers to the propagation link formed by multi-level associations between entries, the ciphertext association weight value refers to the retained value of the path association strength in the encrypted state, the cumulative multiplication operation refers to the operation of continuously multiplying the weights of each segment of the path in the ciphertext state, the ciphertext accumulation refers to the sum of the encryption operation results of multiple paths, the second dynamic threshold refers to the strength critical value for determining the hidden association after decryption, and the hidden association determination refers to the logical judgment of confirming that the non-sensitive entry and the sensitive entry have an implicit association by the decryption superposition value exceeding the threshold.

[0113] In the embodiment of the present application, first, the Euclidean distance between the non-sensitive item and the explicit sensitive item is calculated in step 1031 and the primary candidate set is screened. In the distributed vector space, the coordinates of the explicit sensitive item are (0.3, 0.5, 0.7), and the coordinates of the non-sensitive item "regional demographics" are (0.35, 0.52, 0.68). The distance between the two is calculated by the Euclidean distance formula: If the current first dynamic threshold is 0.1, since 0.057 < 0.1, the non-sensitive entry is included in the primary candidate set. For example, in a certain test, 25 of the 100 non-sensitive entries have distance values ​​below the threshold, forming a primary candidate set.

[0114] Next, the transmission path of each entry in the primary candidate set is traversed through step 1032. Specifically, the three-level transmission path from the non-sensitive entry "regional planning" to the explicit sensitive entry "personal identity information" is detected, namely "regional planning -> demographics -> community management -> personal identity information", with a path length of 3. The ciphertext association weight values ​​on the path are extracted, such as E(0.8), E(0.6), and E(0.5), and the path attenuation coefficient of 0.7 is applied. 3 = 0.343, and the weighted path strength calculated by homomorphic multiplication is E(0.8×0.6×0.5×0.343)=E(0.082). Only conduction paths of level three or above are retained during screening, and short path interference is excluded.

[0115] Next, the ciphertext weighted path strength is cumulatively determined through step 1033. Specifically, homomorphic addition is performed on all conduction paths of a non-sensitive entry. For example, the ciphertext values ​​of the five paths are E(0.082), E(0.076), E(0.063), E(0.055), and E(0.042), and the cumulative result is E(0.318). The superposition value after decryption is 0.318. If the second dynamic threshold is 0.3, since 0.318>0.3, it is determined that the entry has a hidden association. For example, in a certain detection, 8 of the 25 entries in the primary candidate set have superposition values ​​exceeding the threshold and are included in the target non-sensitive entry set.

[0116] In practical applications, in the distributed vector space, the coordinates of the explicit sensitive item "personal ID number" are (0.3, 0.5, 0.7), and the coordinates of the non-sensitive item "regional demographics" are (0.35, 0.52, 0.68). The distance between the two is calculated using the Euclidean distance formula: If the value is lower than the first dynamic threshold of 0.1, the entry is included in the primary candidate set. When traversing its transmission path, a three-level path "regional statistics->community management->identity information" is found. The encrypted weight values ​​E(0.8), E(0.6), and E(0.5) on the path are extracted, and the attenuation coefficient 0.7 is applied. 3 = 0.343, and then the weighted path strength is calculated as E(0.8 × 0.6 × 0.5 × 0.343) = E(0.082). Homomorphic accumulation is performed on the ciphertext values ​​of five similar paths. The decrypted superposition value is 0.318, exceeding the second dynamic threshold of 0.3, indicating that the entry is secretly associated with sensitive information. For example, in one test, the decrypted values ​​of 8 of the 25 entries in the primary candidate set exceeded the threshold. The system automatically classified them as the target non-sensitive entry set and marked the conductive path.

[0117] In the aforementioned step 103, distance screening in an encrypted vector space and multi-level ciphertext calculations are used to precisely locate non-sensitive items that are implicitly associated with explicit sensitive items. A dynamic threshold mechanism adapts to density differences in different semantic regions, multi-hop path strength stacking eliminates interference from short paths, and ciphertext calculations ensure the privacy of sensitive transmission weights. This method effectively reveals hidden associations between non-sensitive items and sensitive information through complex transmission paths while ensuring data security, further enhancing the identification of sensitive information diffusion risks.

[0118] 104. In response to a data output request from the government knowledge base, the content masking range of the non-sensitive item set is dynamically adjusted based on the authority level. The process of adjusting the content masking range of the non-sensitive item set includes: performing semantic-preserving fuzzification operations on associated content that exceeds the authority, and retaining the original semantic structure of associated content within the authority.

[0119] Optionally, step 104 includes:

[0120] 1041. Parse the permission level identifier of the data output request of the government knowledge base and extract the sensitive item filtering rules and semantic fuzzification intensity parameters associated with the permission level;

[0121] 1042. Based on the semantic fuzzification strength parameter, extract the maximum association depth threshold allowed by the current permission level from the semantic association network, and locate the association paths in the non-sensitive item set that have three or more levels of indirect references to the explicit sensitive items according to the sensitive item filtering rule;

[0122] 1043. Divide the associated content of the non-sensitive item set into a subset accessible within the authority and a subset that needs to be blocked due to unauthorized access, based on the maximum association depth threshold and the sensitive item filtering rule;

[0123] Among them, step 1043 specifically includes: recursively traversing the semantic association path of the set of non-sensitive entries, and setting a path length counter on each conductive path; based on the path length counter, when an indirect reference relationship with an explicit sensitive entry is detected, starting the path backtracking mechanism and recording the number of intermediate nodes to obtain a path backtracking result; inputting the path backtracking result into the association strength attenuation function for weight correction to generate a corrected association strength value; if the corrected association strength value is higher than the visibility threshold corresponding to the current permission level and the path length does not exceed the maximum association depth, then the current entry is classified into the accessible subset within the permission, otherwise it is classified into the unauthorized subset that needs to be shielded.

[0124] 1044. For the entry contents in the subset that needs to be shielded due to unauthorized access, a semantically-preserving fuzzification operation is used to retain the original semantic features of the accessible subset within the authority, and only the associated paths that cross the authority boundary are fuzzified.

[0125] Among them, step 1044 specifically includes: extracting the associated path fragment containing the entity name, attribute value and relationship predicate from the unauthorized subset to be masked; according to the preset entity generalization rule, replacing the entity name with the corresponding category label, and converting the attribute value into an interval range or type identifier to obtain the generalized entity name and attribute value; performing main action extraction on the relationship predicate, deleting the modifying words, and generating a simplified predicate expression; reorganizing the generalized entity name and attribute value with the simplified predicate expression according to the original syntactic structure to form a grammatically complete but semantically ambiguous entry content; embedding the permission violation identifier in the entry content, and the permission violation identifier includes the encrypted hash value of the original path and the fuzzy operation type mark. In the subset accessible within the permission, detect the existence of an association path across the permission boundary with the explicit sensitive entry; extract the entity name and relationship predicate in the association path, retain the original name of the entity but replace the entity name attribute value with the type identifier, retain the relationship predicate trunk but delete the modifiers involving sensitive references in the path, and obtain the processed cross-permission boundary path; splice the processed cross-permission boundary path with the original content retained within the permission to ensure the semantic integrity of the non-cross-boundary path; in the spliced ​​content, add a permission isolation mark to the cross-permission boundary path segment, and the permission isolation mark includes the encryption check value and fuzzification strength level of the path.

[0126] In the above scheme, the permission level refers to the user's current data access permission level, the content masking range refers to the set of related content that needs to be hidden or obfuscated according to the permission, the semantic-preserving obfuscation operation refers to the processing method of replacing sensitive information while maintaining the grammatical structure, the permission violation identifier refers to the encrypted label that marks the unauthorized access path, the encrypted hash value refers to the unique path identifier generated by the hash function, the obfuscated operation type marker refers to the metadata that records the operation categories such as replacement, generalization or deletion, the permission isolation marker refers to the encrypted label used to identify the path segment that crosses the permission boundary, the accessible subset within the permission refers to the content set that the current permission allows to be fully displayed, and the unauthorized obfuscated subset refers to the content set that needs to be obfuscated.

[0127] In an embodiment of the present application, first, the permission level identifier of the data output request is parsed through step 1041 and the corresponding sensitive filtering rules are loaded. For example, when the requesting party is the grassroots permission Level-2, the system extracts from the permission parameter table that the maximum allowed association path depth is 2 levels, the address information needs to be blurred to the municipal unit, and the numerical accuracy needs to retain the magnitude but hide the last two digits. The specific operation is to generalize Zhongguancun, Haidian District, Beijing to Beijing, convert a monthly income of 25,000 yuan to approximately 25,000 yuan, and load a blacklist of sensitive entities such as ID numbers and social security numbers to match the fields that need to be masked. At this time, the system also loads a visibility threshold of 0.35 for subsequent path weight determination.

[0128] Secondly, in step 1042, the unauthorized association path is located based on the semantic association network and the attenuation weight is calculated. Taking a certain government item, regional planning, as an example, the transmission path from it to the sensitive item, personal identity information, is detected as "regional planning->demographics->community management->identity information", and the path length is 3 levels. The attenuation weight is calculated by the formula equal to the original weight multiplied by the path length power of 0.7. Where the original weight is 1.0 and the path length is 3 levels, the modified weight is 1.0 multiplied by 0.7 cubed, which is equal to 0.343. Since the visibility threshold of the basic authority Level-2 is 0.35, 0.343 is less than 0.35, and the path is determined to be an unauthorized path, the intermediate nodes, demographics and community management, are recorded in the log and marked as authority boundary points. For example, when 10 such paths are detected, 3 of them involve social security data and 7 involve financial data, all of which trigger the masking rule.

[0129] Next, step 1043 recursively traverses the associated paths and dynamically divides them into subsets. For the two transmission paths for the regional planning entry, path 1 has a depth of 2 and a weight of 0.49. Because 0.49 > 0.35, it is classified as the accessible subset within the authority, and the original description of the association between regional planning and demographic data is returned. Path 2 has a depth of 3 and a weight of 0.343. Because 343 < 0.35, it is classified as the overriding authority-needed masking subset. The sensitive path segment in the returned content is replaced with the regional planning to overriding authority to personal identity information. A metadata tag MASK_LEVEL equal to 3 is also added to the masked path, and the fuzzified range is annotated in the returned result. For example, when a grassroots user queries, only the regional planning data and the city-level statistics are displayed, affecting community services.

[0130] Finally, step 1044 implements semantically-preserving fuzzification and embeds audit markers. Specifically, this involves generalizing entity information in unauthorized paths. For example, "Zhang, ID card number 1101011980XXXX" is replaced with "User, ID card number in Beijing"; converting the value "monthly income 25,000 yuan" to "approximately 25,000 yuan," preserving the magnitude but hiding the precise value; and simplifying the relationship description "directly leading to loan approval failure" to "affecting financial business assessment." The gradient fuzzification intensity is dynamically adjusted based on the path depth. For example, a second-level path retains 70% of the original information, such as in Chaoyang District, Beijing, while a third-level path retains only 30%, such as in Beijing. An audit identifier is embedded at the end of the obfuscated content in the format of "#Permission ID-Timestamp-Operation Type." For example, "#6a3b1c-20250520T1430-MASK3" indicates a level 3 obfuscation operation performed by account ID 6a3b1c at 2:30 PM on May 20, 2025. The SHA-256 hash value of the original path (e.g., a1b2c3d4) is also stored in the audit log to ensure traceability and irreversibility. For example, if a query result is appended with "#6a3b1c-20250520T1430-MASK3," the backend can reverse-verify the operation's compliance using the hash value, but the original sensitive information cannot be restored.

[0131] In practice, when a Level-2 user initiates a data request, the system loads permission rules: maximum association depth 2 levels, address obfuscation to the city level, and numerical precision to the thousandth place. When a three-level transmission path (penalty record -> corporate credit report -> legal person identity information) is detected for an "environmental penalty" entry, the path weight is adjusted to 0.343 using a decay function, lowering it to below the visibility threshold of 0.35. The path is then placed in the "unauthorized obscuration" subset. When recursively traversing the path, the path counter displays 3 intermediate nodes, triggering the obfuscation operation. The original description, "Ms. Zhang (ID card 1101011980XXXX) was fined 30,000 yuan for environmental violations," is replaced with "A legal person was fined approximately 30,000 yuan for environmental violations," and the audit identifier #6a3b1c-20250520T1430-MASK3 is embedded. The SHA-256 hash value a1b2c3d4 of the original path is also recorded in the audit log. The original description of the permissioned path "Environmental Protection Penalty -> Enterprise Credit" is retained. Only the cross-permission path "Enterprise Credit -> Legal Entity Identity Information" is subjected to semantically preserved obfuscation, generating "Enterprise Credit Data Impacts a Legal Entity's Credit Assessment#6a3b1c-20250520T1430-MASK2." The final output not only meets permission control requirements but also enables operation traceability through audit identifiers, complying with the Data Security Law's compliance requirements.

[0132] In the solution described in step 104 above, hierarchical security control of government knowledge output is achieved through dynamic permission-driven masking range adjustment and semantically preserving fuzzification technology. The dual constraints of path depth thresholds and visibility thresholds precisely demarcate the boundaries of accessible content. Entity generalization and predicate simplification eliminate sensitive semantic details while preserving grammatical structure. Permission isolation marking ensures complete traceability of unauthorized operations. This solution protects low-privilege users from accessing valid information while preventing indirect leakage of sensitive information through associated paths, achieving a dynamic balance between security and usability.

[0133] The following is a complete embodiment for steps 101 to 105. Figure 2 As shown:

[0134] In the intelligent management of government knowledge bases, building a semantic association network is the primary step in achieving dynamic control of sensitive information. The system automatically identifies direct and nested reference relationships by parsing normative reference annotations within cross-departmental clauses. For example, when an environmental protection clause references Article X of the Tax Administration Regulations, the system uses regular expressions to precisely match the tax clause name and clause number, generating a direct logical dependency edge from the environmental protection clause to the tax clause. When a tax clause further references a fiscal clause, the system constructs a three-level dependency chain: environmental protection -> taxation -> fiscal, quantifying the dependency strength using a hierarchical attenuation coefficient, such as a second-level reference weight of 0.8. Furthermore, when parsing administrative process documents, dependency parsing techniques are used to extract temporal triggering conditions between task nodes. For example, "completion of preliminary material review" is identified as a precondition for "initiation of expert review," creating process connection edges and annotating data flows, such as expert review results -> results disclosure. The fusion of logical dependency edges and process connection edges forms an initial semantic network. Edge attributes include the reference hierarchy, triggering conditions, and data dependency type, while node attributes store clause summaries and process descriptions, providing a structured foundation for subsequent analysis.

[0135] Based on the above network structure, the system introduces a homomorphic encryption mechanism to protect the association strength calculation process. The initial association strength value is calculated comprehensively by the frequency of citation, hierarchical attenuation and semantic similarity. For example, when the environmental protection clause is cited by three clauses, the basic value of the citation frequency is 3, and the contribution value is 2.4 after combining the secondary citation attenuation coefficient of 0.8. The semantic similarity with the tax clause is calculated by the BERT model to be 0.75, and the final initial association strength value is 1.8. The Paillier algorithm is used to encrypt this value to generate E(1.8), retaining the homomorphic characteristics of addition and multiplication. In the encrypted state, the system traverses the multi-hop transmission path: such as the three-level path of environmental protection clause->tax clause->financial clause, and the path length level 2 triggers an exponential attenuation coefficient of 0.7 2=0.49, homomorphic multiplication is performed on the encrypted values ​​E(1.8) and E(0.75) on the path to generate a weighted path strength E(0.66). The encrypted path strengths of all paths leading to the fiscal clause node are aggregated through homomorphic addition to form a global correlation strength E(1.94), including the encrypted path strengths of the environmental protection transmission path E(0.74) and the direct path of the economic clause E(1.2). Local density weights and cross-layer hop frequencies are then combined to generate a 128-dimensional dynamic propagation vector E(6.9). This vector is orthogonally concatenated with the 768-dimensional encrypted semantic vector generated by Sentence-BERT and then reduced to a 32-dimensional space using t-SNE, forming a distributed vector representation that makes implicit associations explicit in a low-dimensional space.

[0136] Based on the topological distribution of vector space, the system initiates hidden sensitive path detection. The Euclidean distance of non-sensitive items is calculated, centered around the coordinates (0.3, 0.5, 0.7) of the overtly sensitive item "Personal Identity Information." For example, the distance value of 0.057 for the "Regional Demographics" item (0.35, 0.52, 0.68) is below the dynamic threshold of 0.1, making it included in the primary candidate set. The system then traces back the transmission path: For example, in the three-level path "Regional Planning -> Demographics -> Community Management -> Identity Information," the encryption weights E(0.8), E(0.6), and E(0.5) are homomorphically multiplied (0.8 × 0.6 × 0.5 × 0.343 = E0.082) and summed with the ciphertext values ​​of the remaining paths to E(0.318). The decrypted sum of 0.318 exceeds the second threshold of 0.3, indicating a hidden association with this item. This process combines semantic proximity with multi-hop transmission verification to accurately identify hidden links between non-sensitive items and sensitive data.

[0137] Finally, the system dynamically adapts the output content based on the permissions. When a user with Level-2 permissions initiates a request, the system loads rules such as the maximum association depth of 2 and the visibility threshold of 0.35. Taking the entry "regional planning" as an example, the weight of its level 3 path to sensitive entries is 1.0×0.7. 3 = 0.343, which is below the threshold, triggering semantic obfuscation: The following path replaces "Zhang, ID card 1101011980XXXX" with "User X, ID card number in Beijing," generalizes a monthly income of 25,000 yuan to "approximately 25,000 yuan," and embeds the audit identifier #6a3b1c-20250520T1430-MASK3. For example, in the context of regional planning -> demographics, a weight of 0.49 indicates a level 2 path accessible within the permission level. The original description is retained, with only isolation markers added where the path crosses the permission boundary. The obfuscated content is linked to the original path via a SHA-256 hash to ensure audit traceability.

[0138] Figure 3The present invention provides a schematic diagram of a collaborative management system for government affairs knowledge base based on a multi-level authority model. Figure 3 As shown, the system includes:

[0139] Establishing module 31, for establishing a semantic association network between knowledge items in the government knowledge base by analyzing the logical dependencies between the items and the sensitive information propagation path;

[0140] A generating module 32 is configured to perform homomorphic encryption on the computation process of representing the strength of associations between items in the semantic association network, and map the association structure between knowledge items in the encrypted semantic association network into a multidimensional vector space to generate a distributed vector representation reflecting the potential association characteristics between the knowledge items;

[0141] A detection module 33, configured to detect a set of non-sensitive items that are implicitly associated with explicit sensitive items based on the topological distribution of the distributed vector representation in the semantic space;

[0142] The response module 34 is used to respond to the data output request of the government knowledge base and dynamically adjust the content masking range of the non-sensitive item set based on the authority level. The process of adjusting the content masking range of the non-sensitive item set includes: performing semantic-preserving fuzzification operations on the associated content that exceeds the authority, and retaining the original semantic structure of the associated content within the authority.

[0143] Figure 3 The government affairs knowledge base collaborative management system based on the multi-level authority model can be executed Figure 1 The implementation principles and technical effects of the collaborative management method for a government knowledge base based on a multi-level authority model described in the illustrated embodiment will not be elaborated on here. The specific manner in which each module and unit performs operations in the collaborative management system for a government knowledge base based on a multi-level authority model in the above embodiment has been described in detail in the embodiments of the method and will not be elaborated on here.

[0144] In one possible design, Figure 3 The government knowledge base collaborative management system based on the multi-level authority model of the embodiment shown can be implemented as a computing device, such as Figure 4 As shown, the computing device may include a storage component 41 and a processing component 42;

[0145] The storage component 41 stores one or more computer instructions, wherein the one or more computer instructions are called and executed by the processing component 42 .

[0146] The processing component 42 is used for the above Figure 1 The embodiment provides a collaborative management method for a government affairs knowledge base based on a multi-level authority model.

[0147] The processing component 42 may include one or more processors to execute computer instructions to perform all or part of the steps in the above method. Of course, the processing component may also be implemented as one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above method.

[0148] The storage component 41 is configured to store various types of data to support operations at the terminal. The storage component can be implemented by any type of volatile or non-volatile memory device, or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk, or optical disk.

[0149] Of course, a computing device may also include other components, such as input / output interfaces, display components, communication components, etc.

[0150] The input / output interface provides an interface between the processing component and the peripheral interface module, which can be an output device, an input device, etc.

[0151] The communication component is configured to facilitate, among other things, wired or wireless communications between the computing device and other devices.

[0152] Among them, the computing device can be a physical device or an elastic computing host provided by a cloud computing platform, etc. In this case, the computing device can refer to a cloud server, and the above-mentioned processing components, storage components, etc. can be basic server resources rented or purchased from the cloud computing platform.

[0153] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0154] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0155] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, or of course, by hardware. Based on this understanding, the essence of the above technical solution or the part that contributes to the existing technology can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or certain parts of the embodiments.

[0156] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A collaborative management method for government knowledge base based on a multi-level authority model, characterized in that: include: In the government knowledge base, a semantic association network between knowledge items is established by analyzing the logical dependencies between items and the sensitive information propagation paths; Performing homomorphic encryption on the computation process of characterizing the strength of associations between items in the semantic association network, and mapping the association structure between knowledge items in the encrypted semantic association network into a multidimensional vector space to generate a distributed vector representation reflecting the potential association characteristics between the knowledge items; Detecting a set of non-sensitive items that are implicitly associated with explicit sensitive items based on the topological distribution of the distributed vector representations in the semantic space; In response to a data output request from the government knowledge base, the content masking range of the non-sensitive item set is dynamically adjusted based on the authority level. The process of adjusting the content masking range of the non-sensitive item set includes: performing semantic-preserving fuzzification operations on associated content that exceeds the authority, and retaining the original semantic structure of associated content within the authority.

2. The method according to claim 1, characterized in that In response to a data output request from a government affairs knowledge base, dynamically adjusting the content masking range of the non-sensitive item set based on the permission level includes: Parse the permission level identifier of the data output request of the government knowledge base, and extract the sensitive item filtering rules and semantic fuzzification intensity parameters associated with the permission level; Based on the semantic fuzzification strength parameter, the maximum association depth threshold allowed by the current permission level is extracted from the semantic association network, and according to the sensitive item filtering rule, the association path in the non-sensitive item set that has three or more levels of indirect references to the explicit sensitive item is located; Dividing the associated content of the non-sensitive item set into an accessible subset within the authority and a blocked subset due to unauthorized access according to the maximum association depth threshold and the sensitive item filtering rule; For the entry contents in the subset that needs to be shielded due to unauthorized access, a semantically-preserving fuzzification operation is used to retain the original semantic features of the accessible subset within the authority, and only the associated paths that cross the authority boundary are fuzzified.

3. The method according to claim 2, characterized in that For the content of the entries in the subset that needs to be masked due to unauthorized access, a semantically preserving fuzzification operation is applied, including: Extracting association path segments including entity names, attribute values ​​and relationship predicates from the unauthorized masking subset; According to the preset entity generalization rules, the entity name is replaced with the corresponding category label, and the attribute value is converted into an interval range or type identifier to obtain the generalized entity name and attribute value; Extracting the main action of the relational predicate, deleting the modifying words, and generating a simplified predicate expression; The generalized entity names and attribute values ​​are reorganized with the simplified predicate expressions according to the original syntactic structure to form grammatically complete but semantically ambiguous entry content; A permission violation identifier is embedded in the entry content, where the permission violation identifier includes an encrypted hash value of the original path and an obfuscated operation type tag.

4. The method according to claim 2, characterized in that The original semantic features of the accessible subset within the permission are retained, and only the association paths that cross the permission boundary are fuzzified, including: In the accessible subset within the permission, detect the association path with explicit sensitive entries that cross the permission boundary; Extracting the entity name and relationship predicate in the association path, retaining the original entity name but replacing the entity name and attribute value with a type identifier, retaining the main body of the relationship predicate but deleting modifiers involving sensitive references in the path, and obtaining a processed cross-authority boundary path; The processed cross-boundary path is spliced ​​with the original content retained within the permission to ensure the semantic integrity of the non-cross-boundary path; In the spliced ​​content, a permission isolation mark is added to the cross-permission boundary path segment, and the permission isolation mark includes the encrypted check value and the fuzzification strength level of the path.

5. The method according to claim 1, characterized in that Performing homomorphic encryption on the computation process of characterizing the strength of associations between items in the semantic association network, and mapping the association structure between knowledge items in the encrypted semantic association network into a multidimensional vector space to generate a distributed vector representation reflecting the potential association characteristics between the knowledge items, including: Based on the logical dependencies between knowledge items and the sensitive information propagation paths, an initial correlation strength value is generated between each pair of items. The initial correlation strength value is calculated based on the citation frequency, dependency level, and semantic overlap. Encrypting the initial association strength value using a homomorphic encryption algorithm so that the encrypted association strength value retains addition and multiplication operation characteristics; In an encrypted state, traversing the multi-hop association paths between all entries in the semantic association network, performing attenuation weighting on the encrypted association strength value on each multi-hop association path according to the path length, and generating an encrypted weighted path strength; For each knowledge item, aggregate the encrypted weighted path strengths of all paths ending at the knowledge item to obtain an encrypted global association strength value; Based on the encrypted global association strength value, extract the local density weight and cross-layer jump frequency of each entry in the encrypted association network, wherein the local density weight is calculated by summing the encrypted association strengths of the entries within a preset depth, and the cross-layer jump frequency is generated by counting the number of association paths between non-contiguous layers of the entries; The local density weight and the cross-layer jump frequency are fused into an entry-level topological coefficient according to a preset ratio, and the topological coefficients of adjacent entries are weightedly superimposed in an encrypted state through an iterative diffusion algorithm to generate a dynamic propagation vector for the entry; orthogonally concatenating the dynamic propagation vector with the encrypted content vector of the entry, wherein the encrypted content vector is generated by extracting homomorphic encryption features of the entry attributes and text description; According to the relative distance of the concatenated vectors in the multidimensional space, the weight of each dimension is adjusted to generate a distributed vector representation that reflects the potential correlation characteristics between knowledge items.

6. The method according to claim 1, characterized in that Detecting a set of non-sensitive items that are implicitly associated with explicit sensitive items based on the topological distribution of the distributed vector representation in the semantic space includes: Based on the final coordinates represented by the distributed vector, taking the coordinates of the explicit sensitive entry as the center, calculating the Euclidean distance between the coordinates of all non-sensitive entries and the center, and screening the non-sensitive entries whose distance is less than a first dynamic threshold as a primary candidate set; Traversing all possible conduction paths of each non-sensitive entry in the primary candidate set in the encrypted association strength matrix, retaining only conduction paths containing three or more intermediate entries, and extracting ciphertext association weight values ​​in the conduction paths; A cumulative multiplication operation is performed on the ciphertext association weight value of each conduction path, and then the ciphertext accumulation of the operation results of all paths is performed. If the decrypted superposition value exceeds the second dynamic threshold, it is determined that the non-sensitive entry has a hidden association with the explicit sensitive entry, and the non-sensitive entry is included in the target non-sensitive entry set.

7. The method according to claim 1, characterized in that In the government knowledge base, a semantic association network between knowledge items is established by analyzing the logical dependencies between items and the sensitive information propagation paths, including: Construct a citation tracking model for cross-departmental government clauses, analyze direct citation and nested citation relationships based on normative citation annotations in clause texts, and generate a set of logical dependency edges; Analyze the task node topology of administrative process documents, extract the timing trigger conditions and data dependencies between process nodes based on natural language processing, and model the node pairs that meet the preset connection rules as administrative process connection edges; Constructing an initial semantic association network based on the logical dependency edge set and the administrative process connection edge; Based on the sensitive item label database, a back-propagation algorithm is used to trace back layer by layer along the conduction path of the initial semantic association network to identify the indirect diffusion path of confidential data through non-sensitive items, and the residual strength value of sensitive information on each conduction path is calculated using the association strength attenuation function; The hierarchical relationship of the logical dependency edge set, the dynamic weight parameters of the administrative process connection edges, and the residual strength values ​​are integrated into a model to generate a semantic association network with a weighted directed graph structure.

8. The method according to claim 2, characterized in that According to the maximum association depth threshold and the sensitive item filtering rule, the associated content of the non-sensitive item set is divided into an accessible subset within the authority and a blocked subset due to unauthorized access, including: Recursively traverse the semantic association paths of the non-sensitive entry set and set a path length counter on each conductive path; Based on the path length counter, when an indirect reference relationship with an explicit sensitive entry is detected, a path backtracking mechanism is started and the number of intermediate nodes is recorded to obtain a path backtracking result; Inputting the path backtracking result into the association strength attenuation function to perform weight correction to generate a corrected association strength value; If the corrected association strength value is higher than the visibility threshold corresponding to the current permission level and the path length does not exceed the maximum association depth, the current entry is classified into the accessible subset within the permission level, otherwise it is classified into the unauthorized subset that needs to be blocked.

9. A collaborative management system for government affairs knowledge base based on a multi-level authority model, characterized in that: include: Establish a module for establishing a semantic association network between knowledge items in the government knowledge base by analyzing the logical dependencies between items and the sensitive information propagation paths; a generation module configured to perform homomorphic encryption on a computation process representing the strength of associations between items in the semantic association network, and to map the association structure between knowledge items in the encrypted semantic association network into a multidimensional vector space to generate a distributed vector representation reflecting potential association characteristics between knowledge items; a detection module, configured to detect a set of non-sensitive items that are implicitly associated with explicit sensitive items based on a topological distribution of the distributed vector representations in a semantic space; A response module is used to respond to data output requests from the government knowledge base and dynamically adjust the content masking range of the non-sensitive item set based on the authority level. The process of adjusting the content masking range of the non-sensitive item set includes: performing semantic-preserving fuzzification operations on associated content that exceeds the authority, and retaining the original semantic structure of associated content within the authority.

10. A computing device, characterized in that It includes a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement the collaborative management method of the government knowledge base based on the multi-level authority model as described in any one of claims 1 to 7.

Images